UNIVERSITY of CALIFORNIASanta Barbara
Electron-Hole Recollisions in Driven Quantum Wells
A dissertation submitted in partial satisfaction of the
requirements for the degree of
Doctor of Philosophy
in
Physics
by
Hunter Bennett Banks
Committee in charge:
Professor Mark S. Sherwin, ChairProfessor David M. Weld
Professor Cenke Xu
December 2016
The dissertation of Hunter Bennett Banks is approved:
Professor David M. Weld
Professor Cenke Xu
Professor Mark S. Sherwin, Chair
September 2016
Copyright © 2016by Hunter Bennett Banks
iii
To my family
iv
Acknowledgements
First, I would like to thank my advisor Mark Sherwin. He let me into his group and putme on probably the coolest project a young graduate student could imagine, and for thatI am truly thankful. His support, patience, and sage advice has kept me and this workfocused, grounded, and moving forward.
The measurements in this dissertation certainly would not have been possible withoutthe FEL, so I would next like to thank the people who kept the whole thing running forover thirty years. The machine and all of its parts that Dave Enyeart, Gerry Ramian,and Nick Agladze (and others) have designed, built and improved is a testament to theiringenuity and skill. Getting to work with and around Dave has certainly been somethingI’m better for, with his razor wit and uncanny ability to fix things.
I’ve been very fortunate to have worked with some incredibly talented students andpostdocs. I’d like to thank Ben Zaks in particular—Ben took me under his wing when Ijoined the lab and taught me pretty much everything I know about optics. Thank youalso to Dan Ouellette, Devin Edwards, Jessica Clayton, Andrea Hofmann, Jordan Grace,Andrew Pierce, Dominik Peller, Jonathan Essen, Nutan Gautam, Aaron Ma, MengchenHuang, Darren Valovcin, Blake Wilson, and Chang Yun, to name just a few. The lab isin good hands with the new generation of students, Darren, Blake, and Chang.
The materials and the theoretical support provided by our collaborators have beenwonderful. Thank you to Shawn Mack and Loren Pfieffer for the amazing samples,they’ve been such source of interesting and fun physics. Thank you to Fan Yang, QileWu, and Ren-Bao Liu for the stimulating discussions and keen theoretical insight.
Thank you to John Leonard, Garrett Cole for sharing their hard-earned processingtechniques and to Brian Thibeault, Aidan Hopkins and the rest of the cleanroom staff forthe incredible facility they run. Thank you to Mike Deal, Jennifer Farrar, Guy Patterson,Rob Marquez, Rita Makogon, and the rest of the Physics Department staff who go aboveand beyond every day.
I wouldn’t’ve made it this far in grad school without all of my friends and family.You’ve been roommates, labmates, teammates, trivia-mates, golf-mates, and kept medriven and happy and curious. Thank you all for the lunches, the late nights, and thegood times. Thanks to my parents, Kathy and Ben, and my brother and sister, Brianand Katie, who have all been incredibly supportive and surprisingly willing to visit inAugust. And, of course, thanks to Miranda, who has been everything.
v
Curriculum Vitæ
Hunter Bennett Banks
Education
2016 Ph.D., Physics, University of California, Santa Barbara, California
2010 A.B., Physics, Washington University in St. Louis, St. Louis,Missouri
2006 St. Mark’s School of Texas, Dallas, Texas
Professional Experience
2012–2016 Graduate research assistant, Physics Department, UCSB
2011–2012 Teaching assistant, Physics Department, UCSB
2010–2011 Graduate research assistant, Physics Department, UCSB
2008–2010 Undergrad research assistant, Physics Department, WUSTL
Publications
“Anomalous He-gas high-pressure studies on superconducting LaO1−xFxFeAs”, Bi, W.,Banks, H. B., Schilling, J. S., Takahashi, H., Okada, H., Kamihara, Y., Hirano, M., andHosono, H. New Journal of Physics Vol. 12, 023005 (2010)
“Dependence of the magnetic ordering temperature on hydrostatic pressure for the ternaryintermetallic compounds GdAgMg, GdAuMg, EuAgMg, and EuAuMg”, Banks, H., Hillier,N. J., Schilling, J. S., Rohrkamp, J., Lorenz, T., Mydosh, J. A., Fickenscher, T., andPttgen, R. Physical Review B Vol. 81, 212403 (2010)
“Dimeric endophilin A2 stimulates assembly and GTPase activity of dynamin 2”, Ross,J. A., Chen, Y., Muller, Joachim, Barylko, B., Wang, L., Banks, H. B., Albanesi, J. P.,and Jameson, D. M. Biophysical Journal Vol. 100, 729–737 (2011)
“Dependence of magnetic ordering temperature of doped and undoped EuFe2As2 onhydrostatic pressure to 0.8 GPa”, Banks, H. B., Bi, W., Sun, L., Chen, G. F., Chen, X.H., and Schilling, J. S. Physica C: Superconductivity Vol. 471, 476–479 (2011)
vi
“High-order sideband generation in bulk GaAs”, Zaks, B., Banks, H., Sherwin, M. S.Applied Physics Letters Vol. 102, 012104 (2013)
“Terahertz Electron-Hole Recollisions in GaAs Quantum Wells: Robustness to Scatteringby Optical Phonons and Thermal Fluctuations”, Banks, H., Zaks, B., Yang, F., Mack,S., Gossard, A. C., Liu, R., Sherwin, M. S. Physical Review Letters Vol. 111, 267402(2013)
“Antenna-boosted mixing of terahertz and near-infrared radiation”, Banks, H. B., Hof-mann, A., Mack, S., Gossard, A. C., Sherwin, M. S. Applied Physics Letters Vol. 105,092102 (2014)
vii
Abstract
Electron-Hole Recollisions in Driven Quantum Wells
by
Hunter Bennett Banks
Driving semiconductor quantum wells with terahertz electric fields strong enough to
overcome the Coulomb attraction between bound electron-hole pairs leads to high-order
sideband generation (HSG). In HSG, excitons are optically-injected into quantum wells by
a weak near-infrared (NIR) laser while simultaneously being illuminated with a terahertz
field from the UCSB Free Electron Laser. The phenomenon can be described by the
so-called “three step model” developed in high-field atomic physics: (1) the electron and
hole tunnel-ionize in the strong field, (2) the now-free particles accelerate in the field,
and (3) they recollide, emitting a photon. The two lasers are continuous, so the emitted
photons are sidebands on the NIR laser. Because of the large gain of kinetic energy before
recollision, an HSG spectrum has a broad bandwidth with many more sidebands above
the NIR frequency than below. The largest spectra span over one hundred nanometers,
with over 100th order sidebands above and 20th order below.
The electron and hole must remain coherent throughout their trajectories, which can
last hundreds of femtoseconds and extend for more than fifty nanometers, if they are to
recollide. Sidebands have been observed that result from recollisions with kinetic energies
viii
far above the threshold for optical phonon emission. These high orders persist up to room
temperature. Not even quenched disorder in the quantum wells strongly attenuates the
HSG signal.
Because of this coherence, the electron and hole are very sensitive to the complete
band structure of the material. Excitation by linear NIR polarization creates both the
electron and hole in a superposition of spin-up and spin-down states with complex coeffi-
cients given by the relative orientation of the NIR polarization and the THz polarization.
Interference between these the spin-up and spin-down particles, particularly in the va-
lence band and mediated by non-Abelian Berry curvature, has large effects on both the
intensity and polarization state of the sidebands. The connection between HSG and
complete band structure points to the possibility of directly measuring both the band
dispersion relations as well as the non-Abelian Berry curvature of the host material.
ix
Contents
1 Introduction 11.1 Perturbative nonlinear optics . . . . . . . . . . . . . . . . . . . . . . . . 31.2 A brief introduction to GaAs . . . . . . . . . . . . . . . . . . . . . . . . 51.3 The three step model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.4 GaAs quantum wells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161.5 Photoluminescence and absorption . . . . . . . . . . . . . . . . . . . . . 201.6 UCSB Free Electron Laser . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2 The electron-lattice interaction in the three step model 262.1 Sample details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.1.1 Sample design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.1.2 Sample processing . . . . . . . . . . . . . . . . . . . . . . . . . . 282.1.3 Sample optical properties . . . . . . . . . . . . . . . . . . . . . . 29
2.2 Sideband measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.2.1 Optical setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.3 Theoretical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382.3.1 Three step model: Classical interpretation . . . . . . . . . . . . . 382.3.2 Full theoretical results . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4 Conclusion and remaining questions . . . . . . . . . . . . . . . . . . . . . 46
3 The Sideband Spectrometer 473.1 Optical setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.1.1 Linear measurements . . . . . . . . . . . . . . . . . . . . . . . . . 483.1.2 Nonlinear measurements . . . . . . . . . . . . . . . . . . . . . . . 53
3.2 FEL characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673.2.1 Power measurement . . . . . . . . . . . . . . . . . . . . . . . . . . 683.2.2 Frequency and linewidth . . . . . . . . . . . . . . . . . . . . . . . 71
3.3 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4 Terahertz-induced birefringence and sideband polarimetry 774.1 Sample details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.1.1 Sample growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
x
4.1.2 Sample processing . . . . . . . . . . . . . . . . . . . . . . . . . . 804.1.3 Sample optical properties . . . . . . . . . . . . . . . . . . . . . . 82
4.2 Sideband measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . 864.2.1 Experimental preparation . . . . . . . . . . . . . . . . . . . . . . 864.2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.3 Theoretical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1024.3.1 Separation into circular components . . . . . . . . . . . . . . . . . 1034.3.2 Semiclassical trajectories . . . . . . . . . . . . . . . . . . . . . . . 106
4.4 Conclusion and remaining questions . . . . . . . . . . . . . . . . . . . . . 111
A Equipment details 115A.1 Cryostat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116A.2 Sample imaging camera . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119A.3 M2 SolsTiS Ti:sapphire laser . . . . . . . . . . . . . . . . . . . . . . . . . 122
A.3.1 Polarizer attenuator . . . . . . . . . . . . . . . . . . . . . . . . . 125A.4 PhotonControl spectrometer . . . . . . . . . . . . . . . . . . . . . . . . . 125A.5 Acousto-optic modulator . . . . . . . . . . . . . . . . . . . . . . . . . . . 126A.6 Mechanical shutter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128A.7 Absorption LED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128A.8 SPEX monochromator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129A.9 Hamamatsu photomultiplier tube . . . . . . . . . . . . . . . . . . . . . . 132A.10 Acton spectrometer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132A.11 Andor electron-multiplying CCD camera . . . . . . . . . . . . . . . . . . 134A.12 Thomas-Keating energy meter . . . . . . . . . . . . . . . . . . . . . . . . 139
B FEL transport details 140
C Cleanroom processing 142C.1 Quantum well membrane process . . . . . . . . . . . . . . . . . . . . . . 143C.2 Epitaxial transfer process . . . . . . . . . . . . . . . . . . . . . . . . . . . 146C.3 Indium-tin oxide deposition . . . . . . . . . . . . . . . . . . . . . . . . . 150C.4 Dicing samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
Bibliography 152
xi
Chapter 1
Introduction
Almost thirty years ago, experimentalists studying the behavior of noble gases in strong
laser pulses generated light at 32 nm when illuminating the atoms with excitation light
at 1064 nm, the thirty-third harmonic [19]. They observed only odd order harmonics
ωn = nωdrive, where n is the order and ωdrive is the excitation frequency. Most surpris-
ingly, the harmonics from seventh to twenty-seventh order hardly decayed with increasing
order, likely the first reported harmonic plateau. Within five years, intense near-infrared
(NIR) laser pulses were generating harmonics of more than one hundredth order through
a process called high-order harmonic generation (HHG) [36, 50, 43, 40]. Light with
wavelengths around 1 µm could generate light with wavelengths less than 10 nm and the
monochromator was the limiting factor! Despite the nonlinear nature of this process,
the theoretical description for this phenomenon, the three step model, is actually quite
simple and is based on the ionization, acceleration, and recollision of an electron with its
1
parent ion [14]. Many discoveries have been made using this phenomenon, including the
creation of single 35 attosecond pulses using light with a period T = 7 fs (λ = 2 µm) and
the seeding of x-ray lasers, both table-top and free electron lasers [15, 12, 75, 37].
Inspired by HHG from the atomic community, Liu, et al., investigated the possibility
of recollision physics in a semiconductor using excitons, atom-like bound electron-hole
pairs, termed high-order sideband generation (HSG), which was later observed experi-
mentally by Zaks, et al., with the detection of eighteenth order sidebands [41, 66, 73].
Like HHG, HSG is a highly nonlinear mixing process. In HSG, however, two frequencies
of light illuminate the material, one in the NIR wavelength range, ωNIR, and the other
in the terahertz (THz), ωTHz. Through the process of HSG, many THz photons combine
with a single NIR photon to create a comb of sidebands, ωSB(n) of integer order n with
frequencies,
ωSB(n) = ωNIR + nωTHz.
The physics of atoms and excitons are very similar, despite existing at such distant
scales. The dielectric constant and effective masses bring the binding energy from 20 eV
for neon down to 10 meV for an exciton in a GaAs quantum well and the Bohr radius
from less than 1 A up to 10 nm. To stay in the limit of tunnel ionization, the driving
frequency for HSG must be less than 1 THz (4 meV), but the required field strengths
drop to 1 mV/nm (10 kV/cm).
This chapter will introduce the physics of HHG and HSG. First, it will put them into
context by describing perturbative nonlinear optics. Second, it will explain the unique
2
properties of GaAs to understand how to extend the atomic models to explain HSG.
Then, it will go into detail about the three step model and the modifications required to
move from atoms to solids. Finally, it will give background on the generation of intense
terahertz light by the UCSB Free Electron Lasers.
1.1 Perturbative nonlinear optics
To understand the significance of high-order nonlinear processes, it is important to first
introduce perturbative nonlinear optics to demonstrate how unusual double-digit THz
nonlinearities are. This subsection will follow the excellent resource Nonlinear Optics
by Robert W. Boyd [5]. Electrons in free atoms or in crystals are not bound by ex-
actly quadratic potentials, although the approximation is often made, and it is this
anharmonicity that leads to nonlinear optics [20]. In perturbative nonlinear optics, the
polarization P (t) of the medium is expanded in a Taylor series in the applied electric
field. Treating the electric field as a scalar for simplicity,
P (t) = χE(t) (1.1)
= χ(1)E(t) + χ(2)E2(t) + χ(3)E3(t) + · · · . (1.2)
For conventional linear optics, the approximation χ ≈ χ(1) is taken. For sufficiently
intense electric fields, however, the second- and third-order terms can become significant.
For real, vector-natured electric fields, the susceptibility χ(n) is a (n+1)-rank tensor that
3
can couple electric fields polarized along arbitrary axes, but such a treatment is beyond
the scope of this introduction. In general, only the lowest nonlinearity matters as the
higher order χ(n) coefficients are extremely small. For many interesting nonlinear optical
phenomena, it is important to have the largest nonlinear coefficients possible. Usually,
the lowest-order term, χ(2), dominates. For a material to have a non-zero χ(2), however, it
must not be inversion-symmetric. Second-order mixing is a well-established technique for
generating THz pulses with very strong fields [70, 58]. In inversion symmetric materials,
such as (100)-cut GaAs, χ(3) is the lowest-order nonzero nonlinearity. Other planes in
GaAs, such as (110), do have non-zero χ(2), however, but they will not be used in this
dissertation.
The multiplication of sinusoidal fields leads to the generation of new frequencies based
on the sum and difference of the initial frequencies. When the weak frequency is in the
NIR, around 400 THz, and the strong frequency is in the THz, around 0.5 THz, then
sum- and difference-frequency generation become sideband generation. Perturbative THz
sideband generation, first discovered at UCSB using the Free Electron Laser, has been
used to study semiconductor heterostructures since 1995 [8, 34, 6, 61, 60, 7]. Because
sideband generation can be resonantly enhanced by nearby transitions, even electric-
dipole-forbidden ones, many studies have probed the energy spectrum of excitons under
various internal and external perturbations [34, 61].
To generate new frequencies efficiently, matching the phase of the participating laser
fields and their products is important. The wavelength-dependent index of refraction
4
in a material means that the photons created throughout the propagation length of the
material may not all add in phase. From the perspective of conservation laws, both en-
ergy (frequency ω) and momentum (wavevector k) must be conserved. The wavevector
mismatch, ∆k = |kf −∑
i ki| where kf is the generated wavevector and ki are the gen-
erating wavevectors, determines the lengthscale Lc over which there is efficient sideband
generation;
Lc =2
∆k. (1.3)
If there is perfect phase matching, Lc →∞, the entire length of the nonlinear crystal l
contributes to generating new frequencies, assuming an undepleted pump. For finite Lc,
destructive interference prevents all but the back portion of the nonlinear crystal, length
l mod Lc, from contributing. It takes tremendous materials and optical engineering to
efficiently generate THz sidebands in the perturbative regime [6]. Fortunately, sidebands
in GaAs quantum wells are generated efficiently enough that it is straightforward to work
in the limit where l Lc [73].
1.2 A brief introduction to GaAs
The electronic and optical properties of GaAs can be explained extremely well using
band theory. In band theory, the available energy and momentum states are calculated,
and then they are filled with electrons up to the Fermi level (i.e. until all the electrons
have been accounted for). Because they form a degenerate Fermi gas, the electronic and
optical properties are then determined by only the electrons near the chemical poten-
5
tial. The dynamics of these electrons are then treated as if the electrons and holes are
quasiparticles.
From a more experimental standpoint, semiconductor growth and processing tech-
nologies for GaAs are second only to silicon. Material growth techniques like molecular
beam epitaxy (MBE) are extremely mature, allowing for useful heterostructured materi-
als, using other elements from groups III and V, to be grown of exceptional quality. By
growing precision stacks of different materials, the electronic states of the semiconductor
can be tailored to the wishes of the user. This section and later sections about semicon-
ductor physics will follow derivations from the excellent The Physics of Low-Dimensional
Semiconductors by J. H. Davies [17].
The optical properties of a semiconductor are determined by the band structure at the
valence band maximum and the conduction band minimum. For GaAs, these two points
are both at k = 0, the Γ point, and they are separated by about 1.5 eV, corresponding
to a photon with a wavelength of 800 nm, see Fig. 1.1. The characteristic wavevector of
light in this range (about 1/µm) is much smaller than that of an electron (given by the
lattice spacing, about 1/nm). In order to conserve momentum, then, optical interband
transitions are very nearly vertical on the scale of the Brillouin zone. A photon with
wavelength λ ∼ 800 nm can, therefore, excite an electron just barely across the gap,
leaving a empty electron state behind in the valence band. This empty electron state, or
hole, in an otherwise-full band can be thought of as a positively-charged quasiparticle.
These electronic states are more complex than plane-waves, they retain some symme-
6
Electrons
Heavy holes
Light holes
Split-off holesESO ~ 340 meVEner
gy
Momentum
Egap ~ 1.5 eV
@ k = 0
Ebind ~ 4 meVExciton
Figure 1.1: Band structure of GaAs around the Γ-point. The conduction band andvalence band are separated by a minimum energy Egap ∼ 1.5eV at the k = 0 in theBrillouin zone. The conduction band only contains one species of quasiparticle, electrons(green, solid line). The valence band contains three species: heavy holes, light holes, andsplit-off holes (solid red, dashed blue, and dotted black, respectively). The maximum ofthe split-off hole band is 340 meV below the valence band maximum. If electron-electroninteractions are not neglected, the exciton appears, with a binding energy Ebind = 4 meV.
tries from the atomic orbitals they come from. The valence electrons in both gallium and
arsenic are in p orbitals. The next shell of available orbitals are s orbitals. Through the
mechanics of the tight-binding model, the p and s symmetry of the valence orbitals and
the empty orbitals will be preserved in the valence and conduction bands. The top of the
valence band is then six-fold degenerate, three for the px, py, and pz orbitals times two for
the electrons being spin 1/2 (from now on all bands will be implicitly doubly degenerate
from the electron spin). These angular momenta add by the Clebsch-Gordon rules to
create three bands, two with total angular momentum j = 3/2 and one with j = 1/2.
Spin-orbit coupling breaks the degeneracy between the two total angular momentum
states, lowering the j = 1/2 band 340 meV relative to the j = 3/2 bands. That leaves
7
only the two hole bands with j = 3/2 at the very top of the valence band. These two
bands, |j,mj〉 = |3/2,±3/2〉 and |3/2,±1/2〉, are degenerate at k = 0, but have different
effective masses, and so are named the heavy hole and light hole bands, respectively. The
bottom of the conduction band in GaAs is built from s orbital-like states, so there is only
one band.
To fully explain the optical properties of a semiconductor near the band edge, the
Coulomb interaction between charge carriers can no longer be ignored. An electron and
a hole sitting on either end of the band edge have quadratic dispersion relations and a
Coulombic attraction, creating a hydrogenic Hamiltonian,
H =~2k2
e
2m∗e+
~2k2h
2m∗h− e2
4πε
1
|re − rh|,
where ki, m∗i , and ri are the momentum, effective mass and position coordinate of
particle i, either an electron or a hole, and ε = κε0 where κ is the dielectric constant
of the material, κGaAs ∼ 13, and ε0 is the permittivity of free space. Because all of the
masses here will be “effective,” the asterisk denoting such will be dropped. The only
difference between this Hamiltonian and that of the hydrogen atom is that the effective
masses have been rescaled by the bands and the permittivity of the material is rescaled
by the dielectric constant due to screening. Transforming into the center of mass frame
gives the familiar result;
H =~2K2
2M+
~2k2
2µ− e2
4πε
1
r,
where K and M are the center of mass momentum and mass, respectively, k is the
relative momentum of the electron and hole, µ is the reduced mass memh/(me + mh),
8
and ε is the dielectric constant of the material. This Hamiltonian allows for the formation
of bound states called excitons, where the Bohr radius and binding energy are about 10
nm and 4 meV. If the electron and hole exist at the bottom and top of their respective
bands, then the exciton will actually exist inside the previously-forbidden band gap. The
exciton can even be excited directly with an optical photon with energy less than Egap,
see Fig. 1.1. Excitons are not long-lived, however, as the close proximity of the electron
and hole encourages recombination on the order of nanoseconds after creation [18, 29].
1.3 The three step model
One of the early empirical discoveries in HHG was the existence of a harmonic plateau
and cutoff. The harmonic spectrum would exhibit constant intensity for many dozens of
orders, even into the hundreds, but then would disappear very quickly at the cutoff,
ncutoff ∼3UP + Ebind
~ω(1.4)
UP =e2F 2
4mω2, (1.5)
where ncutoff is the cutoff harmonic order, UP is the ponderomotive energy, Ebind is the
ionization energy of the atom, and ω is the center frequency of the intense laser pulse
[36]. The ponderomotive energy is the average kinetic energy of a charged particle in a
sinusoidally oscillating electric field, where F is the electric field strength, and e and m
the charge and mass of the electron. This cutoff law can actually be understood very
9
EHHGEbind
(0)
(1)
(2)
(3)(1)
(2)
(3)
Ebind
ENIR
EHSG
HHG: HSG:
- Electron - Hole
Parent atom/ionEner
gy
Momentum
Figure 1.2: The three step model in HHG and HSG in momentum space. In HHG, theelectron tunnel-ionizes from its parent atom/ion, (1). Next, the electron accelerates as afree particle in the driving electric field, (2). Initially, the electron moves away from theion, but eventually the electric field turns around and so does the electron. The parention is assumed to be infinitely massive. Eventually, if the electron tunneled out at anallowed phase of the electric field, it recollides with the parent ion, (3). The electron andion recollide at the origin in real space, but the electron has gained a significant amountof energy and momentum. In HSG, the exciton must first be created, (0). Steps (1), (2),and (3) are all much the same as in HHG, but now the hole has finite mass. This figureis not to scale.
simply using a semiclassical three step model proposed by Corkum [14]. This model was
made fully quantum using the strong field approximation by Lewenstein, et al. [39]. This
section follows Atoms in Intense Laser Fields by C. J. Joachain, N. J. Kylstra, and R.
M. Potvliege for descriptions of HHG and other atomic phenomena [28].
The fundamental physical difference between HHG and HSG is that the exciton must
be created before it can be ionized by the driving field, the “zeroth” step of HSG, see
Fig. 1.2. This new laser field changes the frequency-order relationship in an HSG spec-
trum such that ωSB = ωNIR+nωTHz. Adding this excitation step means that the properties
of the newly-created electron-hole pair can be controlled. The NIR frequency could be
10
tuned to create excitons, or a free electron-hole pair with non-zero individual momenta.
The NIR polarization can be changed to excite electrons and holes with different spin
states. Xie, et al. predict there will be a plateau at positive and negative orders when
free electron-hole pairs are created [63].
In the first step, the exciton tunnel-ionizes in the strong driving field, see Fig. 1.2.
Ionization via tunneling is not a resonant effect. The Coulomb potential can be warped
by a strong enough electric field to actually cause the electron to tunnel through the
Coulomb barrier from the hole. Tunnel ionization occurs in a regime defined by the
Keldysh parameter,
γ =ttunnel
Tdrive
(1.6)
= ω
√2µEbind
eF(1.7)
where ttunnel is the tunneling time and Tdrive is the laser period [32]. Recall that in
the center of mass frame, the hole is stationary and the electron has the reduced mass
µ = memh/(me +mh). If γ < 0.5, the exciton will ionize only via tunnel ionization, [26].
The tunnel-ionization step is largely similar in HHG and HSG as it was generalized by
Keldysh. In HSG, however, the THz fields that can be easily achieved are much larger,
relative to the Coulomb barrier in an exciton, than the pulsed laser fields are relative
to the barrier in atoms. A sinusoidal field of 35 kV/cm in GaAs, easily generated by
the UCSB Free Electron Laser, is strong enough to classically ionize, i.e. no tunneling is
required, the most tightly bound excitons for the large majority of the field’s period.
11
In the second step, the now-free electron accelerates in the field. As a free charged
particle in an oscillating field, the ponderomotive energy becomes the dominant energy
scale, defined in Eq. 1.5. Initially, the electron accelerates away from the hole, but
eventually the driving field changes sign and drives the electron back towards the hole.
Whether the two recollide depends on the phase of the driving field when tunnel ionization
occurred, based on solving Newton’s equation for the electron:
µx = eF cos(ωt+ ϕ) (1.8)
x(t = 0) = 0 (1.9)
x(t = 0) = 0 (1.10)
where x(t) is the position of the electron as a function of time and the intial position and
velocity of the electron is zero at t = 0, and ϕ is the phase of the driving electric field
at t = 0. If ϕ is in the interval when the electric field magnitude is increasing, [π/2, π]
or [3π/2, 2π], then the electron gains too much kinetic energy before the driving field
turns around, and the electron never recollides with the hole. If tunnel ionization occurs
during the other two quarters of the cycle, then the electron can recollide with the hole,
possibly several times depending on ϕ.
The acceleration step is far more complicated in HSG, and a large portion of this
dissertation will address what happens during this step. The parent ion in HHG, being
thousands of times heavier than the electron, is assumed to be stationary. The hole,
however, has a finite mass, so its contribution to the final sideband energy must be
12
0 . 0 0 . 5 1 . 0 1 . 5 2 . 0
- 5 0 0
5 0 0
1 0 0 0
0 3 0 6 0 9 00
1
2
3
5 4 °1 °En
ergy a
t reco
llision
(UP)
P h a s e a t t u n n e l i n g ( ° )
l o n g s h o r t t r a j . t r a j .
1 8 °5 4 ° : s h o r t t r a j . 1 ° : l o n g t r a j .
1 8 ° : h i g h e s t o r d e r
Positi
on (n
m)
T i m e ( p s )
- 1 8 ° : n o r e c o l l i s i o n
T T H z = 1 . 8 p se l e c
t r o n
h o l e
0 . 0 0 . 5 1 . 0F r a c t i o n o f a T H z c y c l e
Figure 1.3: Solutions of the classical three step model in GaAs. (left) The kinetic energyat recollision as a function of the phase ϕ of the driving electric field at tunneling,Fdriving ∝ cos(ωt+ ϕ), in units of the ponderomotive energy UP . The electron and holehave several opportunities to collide, this derivation assumes that they do not survive pastthe first and that the bands are parabolic. Every solution here has two possible initialphases, differentiated by their trajectories, long and short. (right) Different possibletrajectories based on tunneling time for FTHz = 11 kV/cm, fTHz = 570 GHz, me = 0.063and mh = 0.1. For ϕ = 18, the electron and hole never recollide. For ϕ = 1 or ϕ = 54,the recollision has the same total kinetic energy, but the trajectories are very different.The long trajectory explores almost 1 µm of space and lasts over 1 ps. For ϕ = 18,the maximum energy trajectory, the whole duration is relatively short compared to theϕ = 1 trajectory.
13
considered at recollision. The hole contribution can be accounted for by transforming into
the center of mass frame and using the reduced mass, a fairly trivial though important
adjustment. As the electron and hole accelerate in the field, they can interact with
the host lattice, utterly unlike in HHG. During their trajectories, they gain a substantial
amount of kinetic energy, so there are many mechanisms for them to relax back across the
band non-radiatively. Accounting for scattering will be discussed in Chap. 2. The large
kinetic energies also mean that complicated subband mixing influences recollisions. A
substantial amount of theoretical analysis has gone into this step as well to understand
the possible observation of Berry phase physics and other quantum coherence effects
[68, 16, 69, 67]. Accounting for these effects will be discussed in Chap. 4.
In the third and final step, the electron and hole recollide. Being particle and antipar-
ticle, the electron and hole can radiatively recombine, emitting a photon. Part of that
emitted photon’s energy comes from reverse ionization, which explains why Ebind appears
in the expression for the cutoff order, although in HSG reverse ionization is simply energy
detuning of the NIR excitation laser from the band gap. The other part comes from the
kinetic energy gained in the driving field. If the kinetic energy at recollision is plotted as
a function of ϕ and the bands are parabolic, there is a peak in recollision kinetic energy
at approximately ϕ = 18, where the kinetic energy EK = 3.2UP , see Fig. 1.3(left). This
result is in very close agreement with the empirical cutoff law. Because of the complex
valence band of GaAs and again utterly unlike HHG, it is possible for the hole to change
state during its trajectory such that it can no longer radiatively recombine with the
14
electron.
This model also explains why the harmonic spectrum has a plateau in energy. During
one full cycle of the driving field, there are four instants when an electron can tunnel out
to create a harmonic of a given order. All of the highest orders then have the same number
of opportunities to be created. One factor of two comes from the inversion symmetry in
the system, that the positive and negative halves of the optical cycle contribute the same.
The other factor of two comes from what are termed the “short” and “long” trajectories,
see Fig. 1.3(left). The short trajectory occurs when the electron tunnels out between 18
and 90 after the peak of the driving field. The electron gains a small amount of kinetic
energy before turning around and recolliding, and the whole trajectory occurs in less than
half a cycle, see Fig. 1.3(right). The electron has the most kinetic energy at recollision.
The long trajectory occurs when the electron tunnels out between 0 and 18 after the
peak of the driving field. These trajectories can take more than a half-cycle to complete,
and the electron can gain substantially more kinetic energy during the trajectory than it
has at recollision.
Recent pulsed THz work
The observation of recollision physics in solids is not limited to GaAs and free electron
lasers. Strong, subresonant fields have created HHG in bulk ZnO, GaSe, and WSe2 [22,
51, 38]. In these works, coherent populations of electrons tunnel across the band gap when
driven with very strong pulsed mid-infrared or THz pulses. Harmonic photons are then
created via two pathways, non-parabolic band structure, including Bloch oscillations,
15
and electron-hole recollisions, depending on the frequency and field strengths involved.
Langer, et al., even observed HSG in WSe2 using a NIR excitation pulse and a THz
driving pulse [38]. By using a very short NIR pulse time-delayed through the THz pulse,
a slight delay between NIR excitation and sideband emission was observed, consistent
with the delay predicted by the acceleration step in the three step model.
1.4 GaAs quantum wells
To study excitons in GaAs, it is useful to build quantum wells (QWs) into the structure
of the material. Growing QWs restricts the effective dimensionality of the electrons
and holes, affecting both the band and the exciton structure in nontrivial ways. The
binding energy of excitons, for example, can increase by more than a factor of two
based on confinement alone, Ebind ∼ 10 meV for 5–10 nm QWs. Development in growth
technologies like molecular-beam epitaxy allows for controlled deposition of single atomic
layers in a large plane. For GaAs, the very highest quality quantum wells are grown
alternating several nanometers of GaAs and Al0.3Ga0.7As. Both materials have a direct
band gap and very similar lattice constants, but Al0.3Ga0.7As has a larger band gap by
about 400 meV. About 65% of the increase is in the conduction band, and 35% is in the
valence band. Electrons at the bottom of the conduction band in the GaAs well region
are not allowed to propagate into the AlGaAs barriers, the z-direction, and so electrons
are restricted to motion in the x-y plane.
Breaking the semi-continuous lattice symmetry in the growth direction forces electrons
16
and holes into a discrete energy spectrum. The simplest model for this situation is the
QW with infinite barriers,
V (z) =
+∞ z ≤ −L/2
0 −L/2 < z < L/2
+∞ z ≥ L/2
where V (z) is the potential and the QW has a width L. This potential requires any
wavefunctions to be free particles in the well and zero outside of it. The full set of
orthonormal wavefunctions ψn(z) and their energies En,
ψn(z) =
√
2L
sin(nπLz)
n ∈ 1, 3, 5, . . .√2L
cos(nπLz)
n ∈ 2, 4, 6, . . .
En =n2π2~2
2mL2.
There are still two other dimensions, x and y, to add to the complete wavefunction.
Because the potential does not depend on x or y, it is separable and the the wavefunctions
are plane waves in the plane of the quantum well. The total energy of the particles then
is the sum of the energies in all three dimensions,
En(kx, ky) =n2π2~2
2mL2+
~2
2m
(k2x + k2
y
).
17
The total energy of the particle is actually continuous, because kx and ky are continuous,
even though the z subbands are discrete. When the barriers are finite, the wavefunctions
and energies require numerical calculations. Lowering the barriers has important effects
on the properties of the wavefunctions. The energy of a subband still scales approximately
as n2, but the number of bound states is finite.
Calculating the electronic states in a GaAs QW requires careful consideration of the
band physics and the QW physics. Well widths in GaAs are generally tens of nanometers,
but they can be as small as a few nm. Even at their smallest, the wells are several atomic
layers thick, so the QWs perturb electron states at a much smaller energy scale than the
band gap. For GaAs/Al0.3Ga0.7As QWs, confinement modifies the conduction band in
a straightforward manner. The two electron states only differ by spin direction, which
does not couple to the confinement potential. The energy at k = 0 is raised above the
GaAs conduction band due to the confinement energy, but the effective mass remains
approximately the same. Electron states in different subbands simply stack upwards
in energy. The valence band, however, changes dramatically as a result of confinement.
First, the two hole states will shift down in energy (“up” from a hole’s perspective). Taken
with the increase in the conduction band energy, the band gap will increase relative
to bulk material. The two hole states couple differently to the confinement potential,
however, because their masses are different. The QW then breaks the heavy hole-light
hole degeneracy at k = 0. The holes are p-like, so the effective masses are actually
anisotropic. A hole excited in a px orbital has good overlap with nearby px orbitals in
18
"Heavy" holes
"Light" holes
Ener
gy
Momentum
Avoidedcrossing
Figure 1.4: Schematic valence band structure in the plane of the QW. The confinementenergy at k = 0, the peak of the subbands, is determined by the effective mass in thegrowth direction, so heavy holes do not shift downward as much compared to light holes.The effective mass normal to the growth direction, in the plane of the QW, changes fromheavy to light, and vice versa. Because the hole states are coupled, the bands do not crossbut avoid each other. The dynamics of holes accelerating through the avoided crossingmust be calculated carefully.
the x-direction, and so it is classified as “light.” That same px hole, however, has poor
overlap with the neighboring orbitals in the y- and z-directions, and so it is “heavy.”
The energy level for the QW is based on the z-direction mass. In the plane, the light
and heavy holes “switch” in a sense. The dispersion relations then actually will cross at
some finite momentum near k = 0, but because the two holes are coupled, there will be
an avoided crossing, see Fig. 1.4.
Optical interband transitions in QWs follow three selection rules. First, the transi-
tions must be vertical on the scale of the Brillouin zone because the photon momentum
is so small, which is a general rule of optical transitions in semiconductors. Next, also
similar to bulk GaAs transitions, the the spin states must follow the standard electric
dipole transition rule where the angular momentum of the states must change by one.
The hole states |3/2,+3/2〉 and |3/2,−1/2〉 therefore couple to the electron state |S, ↑〉.
19
Finally, unique to interband transitions in QWs, the subband quantum number n must
not change to ensure good wavefunction overlap. This rule is not strictly true, however.
Because the valence band and conduction band QW Hamiltonians have different masses
and barrier heights, different subband envelope functions are not strictly orthogonal,
although requiring ∆n = 0 is typically a good approximation.
1.5 Photoluminescence and absorption
With the knowledge of the types of states in bulk or heterostructured GaAs on paper,
it is important to be able to count the number of those states at a given energy E per
unit volume. This number is called the density of states n(E). Measuring the density of
states provides information about the quality of a sample. Dirty or imperfectly-grown
material will manifest as unexpected features. In general, the density of states for an
arbitrary system with available energies εi is,
n(E) =1
Ω
∑i
δ(E − εi),
where 1/Ω is the volume normalization factor. For a free fermion gas, which is a good
approximation for the electrons and holes near the band edge, the density of states
20
depends strongly on the dimensionality fo the system,
n3D =m√
2mE
π2~3
n2D =m
π~2.
The density of states is particularly useful because when several bands or transitions
contribute, their respective densities of states simply superpose. In undoped bulk GaAs,
physical density of states will have a small peak just below the Egap on top of the n3D.
Optical measurements are only sensitive to transitions between these states, which
requires a slight modification of n(E). The optical joint density of states nopt(~ω) is
then,
nopt(~ω) =2
Ω
∑i,j,k
δ(Egap + εv,i(k) + εc,j(k)− ~ω),
where εv,i(k) and εc,j(k) are the energies of the hole in state i and electron in state j,
respectively. They are both at the same momentum because the photon has negligibly
small momentum.
The optical properties of GaAs QWs are traditionally studied using two measure-
ments, photoluminescence (PL) and optical absorbance. In a PL measurement, electron-
hole pairs are injected with an energy much larger than the band gap of the material,
generally using a laser. If the lifetime of the two particles is longer than the scattering
time, then they will fall down to a state near their respective band edges. The ensemble
21
of electron and hole pairs will come to thermal equilibrium with the lattice before some-
how recombining. The particles must recombine radiatively if they are to be observed.
PL does not, therefore, directly measure the density of states of the lowest radiative
transition(s). Instead it measures the optical density of states modified by a Boltzmann
factor from the electron temperature,
SPL(ω) ∼ nopt(ω) · e−ω/kT , (1.11)
where SPL(ω) is the PL signal as a function of emission photon energy ω. The equation
also includes factors related to the strength of the transition at photon energy ω, but
those have been ignored as the transitions in focus in this dissertation have similar dipole
moments. The Boltzmann weighting is asymmetric, it favors the lower energy side of
otherwise symmetric peaks. At very low temperatures in undoped GaAs-based quantum
wells (T < 20 K), the heavy hole exciton will be the primary feature in PL [29, 1]. As
the temperature increases, the high-energy tail of the Boltzmann factor will get longer,
making higher-energy states such as the light hole exciton or free electron states apparent
[9, 54]. Eventually, as the temperature increases, the PL signal will decrease substantially
because injected electrons and holes will be much more likely to recombine nonradiatively.
Another useful spectrum is the optical absorbance of a sample. Here, a broadband
light source, such as an incandescent lamp or pulsed laser perhaps, illuminates a sample,
usually around the band gap. Light that gets absorbed, to create an exciton or a free
electron-hole pair, will either get re-radiated out of the beam path or the quasiparticles
22
1 . 5 5 0 1 . 5 5 50 . 0
0 . 5
1 . 0
T = 1 5 K
Signa
l (norm
.)
F r e q u e n c y ( e V )
P L A b s .
1 0 n m G a A s Q W sS l i g h t P Lr e d - s h i f t
8 0 0 7 9 8W a v e l e n g t h ( n m )
Figure 1.5: Comparison of PL and absorbance spectra in QWs. The PL and absorbancedata for 10 nm GaAs QWs are plotted, normalized to the peak of respective signal.The PL signal is slightly red-shifted from the absorbance peak due to weighting by theBoltzmann factor.
will relax non-radiatively. Because the light that is detected was not absorbed and re-
radiated, the absorbance signal Sabs is directly proportional to the optical joint density of
states without thermal weighting factors, see Fig. 1.5. Relatedly, the absorbance signal
is only sensitive to the absorbed light; the re-radiated light is not measured. These
measurements are then very applicable to resonant excitation experiments, the same
process that absorbance measures, such as HSG.
1.6 UCSB Free Electron Laser
Entering the regime of such nonlinear THz physics requires the use of the UCSB Free
Electron Laser (FEL). The FEL generates THz light by steering a 1 A beam of electrons
with up to 5 MeV through an permanent magnet undulator inside a high-Q laser cavity
23
[48]. The FEL cavity is approximately 6 m long, leading to a longitudinal mode spacing
of 25 MHz. Before the electron beam turns on, those modes are populated with thermal
radiation. The early electrons in the 5 µs-long electron beam emit synchrotron radiation
as they pass through the undulator and into the laser cavity at a frequency based on
their kinetic energy and the magnetic period of the undulator in a 1 GHz band. This
radiation interacts with the thermally populated modes, amplifying them. As the fields
in the modes get stronger, later electrons become more likely to radiate in phase with the
radiation in the cavity. This coherent amplification increases the power in the cavity as
in a more traditional laser, and, if the electron beam is long enough, the FEL lases. The
initial thermal modes and the quality of the electron beam determine the final frequency
and phase properties of the laser emission. There are currently two undulator/laser
cavities in the FEL electron beamline, the Far Infrared laser, tunable from 1–4 THz, and
the Millimeter (MM) laser, tunable from 200–800 GHz. The MM laser was used for all
experiments in this thesis. The laser light is directed from the FEL experiments using
a periodic focusing transport system that is pumped to about 200 mtorr to minimize
atmospheric absorption. The lenses and windows of the transport system are made of a
transparent plastic.
There are two mechanisms for coupling the THz power out of the FEL, [30, 48]. The
most straightforward is the hole coupler. Because of the relatively long wavelengths of the
THz radiation emitted by the FEL, the traditional “partially-silvered end mirror” of the
cavity is actually a curved metal mirror with a small hole bored through the center, the
24
“hole coupler.” The THz pulse duration from the hole coupler is up to about 1.5 µs long
with a peak power of about 5 kW. Because the hole coupler is a circle, the beam profile
in the transport system is nearly Gaussian. The beam profile also does not change over
the course of the pulse. The other output coupler is the silicon coupler “cavity dump”
(CD) [53]. A thin silicon wafer is placed at a beam waist in the FEL cavity. The wafer
is oriented at nearly Brewter’s angle to minimize the reflection out of the cavity. Once
the FEL is lasing, a Q-switched 532 nm pulsed laser fires at the silicon wafer, exciting an
electron-hole plasma in the silicon with a plasma frequency above the lasing frequency,
creating a temporary mirror. This mirror couples all of the THz light out of the cavity.
The THz pulse duration from the cavity dump is 40 ns long, twice the length of the
cavity, with a peak power of about 60 kW. A pulse duration of 40 ns is relatively short
given the 1 Hz repetition rate of the FEL, but it is about 10,000 periods of the THz
light, making it a quasi-continuous wave source. Because the silicon wafer changes from
insulating to pseudo-metallic, its reflection properties change dramatically. Recall, the
silicon wafer is mounted at approximately Brewster’s angle, but at a beam waist. The
incident light hits it at a range of angles, sweeping across Brewster’s angle. The electron-
hole plasma, however, reflects all angles approximately equally, so the beam profile is
much more Gaussian during the cavity dump than before it. This spatial profile change
has important consequences when measuring the instantaneous power in the cavity dump
FEL pulse. Failure to account for this change can lead to power measurements that are
off by as much as 50%.
25
Chapter 2
The electron-lattice interaction in
the three step model
Unlike in high-order harmonic generation (HHG), where the electron and ion acceler-
ate through free space, the electron and hole in high-order sideband generation (HSG)
must accelerate through a complex and dynamic lattice. Phonon interactions provide a
particularly strong mechanism for the quasiparticles to lose coherence and energy to the
lattice. There are two distinct phonon classes in III-V semiconductors like GaAs, acous-
tic phonons and optical phonons. For the energy scales of HSG, fortunately, acoustic
phonons actually should not have a large effect, with scattering rates on the order of 1
THz due to their linear dispersion relation making momentum conservation in electron-
acoustic phonons scattering difficult to achieve [33, 49, 17]. Optical phonons, however,
interact very strongly with electrons and holes, with scattering rates on the order of 10
26
THz, much faster than the 0.5 THz driving frequency, because of their relatively large
energy (36 meV for longitudinal optical phonons) and flat dispersion relation [4]. The
thermal population of optical phonons is only significant above liquid nitrogen tempera-
tures, but very hot charge carriers can emit such phonons very efficiently. It is, therefore,
reasonable to think that the full plateau of the three step model will not be observed
because of optical phonon scattering. Previous experiments only observed sidebands up
to almost exactly this 36 meV cutoff, but the host InGaAs quantum wells were highly
strained [73].
Here, we investigate the consequences of a dynamic lattice in HSG using clean, strain-
free GaAs/AlGaAs quantum wells. By minimizing other scattering pathways, we are able
to focus on the the electron-optical phonon interaction. Furthermore, the exciton absorp-
tion wavelength of these new quantum wells fit in the tuning range of the Ti:sapphire
laser from low temperature (15 K) up to room temperature. This temperature range
allows for control of the thermal population of optical phonons, meaning control of the
scattering rate that the electrons and holes experience during the acceleration step of
HSG.
27
2.1 Sample details
2.1.1 Sample design
Two quantum well samples with twenty QW repetitions were grown for this experiment.
Both samples were grown on semi-insulating GaAs substrates by molecular beam epitaxy
with no intentional doping. The 5 nm GaAs QW sample consists of 5 nm GaAs QWs with
20 nm Al0.3Ga0.7As barriers and the 10 nm QW sample consists of 10 nm Al0.05Ga0.95As
QWs with 20 nm Al0.3Ga0.7As barriers. For the experimental convenience of matching
excitonic absorption peaks in the different samples, the band gap of the wider well was
raised by adding 5% aluminum to match the 2D band gap of the narrower well. The
growth machine was not calibrated to grow 5% AlGaAs directly. Instead, it discretely
deposited five parts GaAs then one part 30% AlGaAs. Growing twenty QWs increases the
absorption of NIR photons, increasing the number of excitons created and, presumably,
the number of sideband photons created. Because quantum confinement raises the exciton
absorption lines above the band gap of bulk GaAs, optical transmission experiments like
sideband generation and absorption experiments will be entirely blocked by the substrate
unless it is removed.
2.1.2 Sample processing
The substrate was removed by a series of etches. Etching through 500 microns of substrate
requires fast etch but highly selective etch rates. A series of three wet etches were
28
50 nm GaAs 50 nm GaAs40 nm Al0.3Ga0.7As 40 nm Al0.3Ga0.7As20 nm Al0.3Ga0.7As 20 nm Al0.3Ga0.7As
20×QW region 5 nm GaAs 10 nm Al0.05Ga0.95As
10 nm Al0.3Ga0.7As 10 nm Al0.3Ga0.7As50 nm Al0.3Ga0.7As 50 nm Al0.3Ga0.7As
50 nm GaAs 50 nm GaAsEtch stop region 300 nm Al0.7Ga0.3As 300 nm Al0.7Ga0.3As
50 nm GaAs 50 nm GaAsSubstrate 500 µm GaAs 500 µm GaAs
Table 2.1: The growth structure of the two samples. The only difference is the materialin the quantum well region. Both samples are grown analog except for the Al0.05Ga0.95Aslayers in the 10 nm QW sample, which were grown digitally. The 5% aluminum alloyingraises the band gap of the quantum well to cancel out the smaller confinement energy ofthe wide quantum well.
used. First, diluted H2SO4 and H2O2 etches AlxGa1–xAs quickly regardless of aluminum
content. Then, a mixture of citric acid and H2O2 etches until an etch stop layer of
Al0.7Ga0.3As [45, 10]. Finally, the etch stop layer is removed with buffered HF. The
final result of the process was a sample approximately 5 × 7 mm in size with a thin
membrane “window” approximately 1× 2 mm in size less than 1 µm thin, supported by
the surrounding substrate frame. Full processing details can be found in Appendix C
This membrane is very fragile, but was able to withstand many cycles of vacuum pumping
and venting, thermal cycling from room temperature to 12 K, and sideband experiments
in air without damage.
2.1.3 Sample optical properties
Both photoluminescence and absorption spectra were measured to characterize the sam-
ples, see Fig. 2.1 for the optical schematics. Excitation for PL was a 3 mW HeNe laser
29
HeNe laser
Cryostat
Spectrometer + ICCD
Collection lens and fiber
Fiber from LED
Sample
Lens
PL optics:
Absorption optics:
Figure 2.1: Experimental schematic of PL and absorption measurements. PL optics : tomeasure PL, a lens focuses a HeNe laser onto the sample at an angle, the thin red line. Alens oriented normal the the sample collects the incoherent PL signal, the thick, faint redline. The collected light is sent into a fiber that is coupled to an imaging spectrometer tobe measured by an intensified CCD. Absorption optics : For the absorption measurement,an LED spot centered at 810 nm is focused on the sample through it and into a collectinglens.
focused on the sample at an angle. An objective lens normal to the sample collimated
the light into a fiber coupler that brought the light to the intensified CCD. The light for
absorption spectra was generated by an 810 nm LED focused on the sample and then
collected by an objective lens.
Photoluminescence and absorption spectra of the samples reveal the states near the
band edge, see Fig. 2.2. In the PL spectrum at low temperature, the lowest-energy
radiative state is the heavy hole exciton, which is the same in both samples, as designed.
Excitation for PL was a 3 mW HeNe laser focused on the sample at an angle. An
objective lens normal to the sample collimated the light into a fiber coupler that brought
the light to the intensified CCD. The two samples can, however, be differentiated by their
30
1 . 6 0 0 1 . 6 2 0 1 . 6 4 0 1 . 6 6 00 . 00 . 51 . 01 . 52 . 02 . 5
*
5 n m Q W s 1 0 n m Q W s
Signa
l (arb.
units)
F r e q u e n c y ( e V )
P L
A b s o r b a n c e
T = 1 2 K
*
7 7 0 7 6 0 7 5 0W a v e l e n g t h ( n m )
Figure 2.2: The PL and absorption spectrum at low temperature. Both samples havenearly identical PL spectra, as only the heavy hole exciton peak is observed. The ab-sorption spectra of both samples, shifted vertically for clarity, are much different, owingto the observation of the light hole exciton (the higher-energy, smaller peaks, denoted by*). The half linewidth of the heavy hole exciton in both samples is approximately 2.8meV. The data was published in [4]. Copyright 2013 by the American Physical Society.
absorption spectra. The light for absorption spectra was generated by an LED, centered
at 810 nm, focused on the sample and measured similar to PL. The heavy hole-light hole
(HH-LH) splitting in the valence band scales as the ratio of the effective mass to the
square of the well width, so the 5 nm GaAs QWs have a 25 meV splitting while the 10
nm AlGaAs QWs have only 10 meV splitting.
The PL and absorption spectra can be followed up to higher temperatures, see Fig. 2.3.
As temperature increases, band gap of GaAs redshifts approximately quadratically, and
the Al0.3Ga0.7As shifts similarly. The exciton lines from both experiments follow that
redshift very closely. Photoluminescence peaks become too weak to be detected by about
200 K, because the photoinjected electrons and holes relax nonradiatively much more
efficiently at high temperatures. Excitons in absorption measurements are still detectable
31
0 5 0 1 0 0 1 5 0 2 0 0 2 5 0 3 0 01 . 5 2 01 . 5 4 01 . 5 6 01 . 5 8 01 . 6 0 01 . 6 2 01 . 6 4 0
P LA b s o r p t i o n
Energ
y (eV
)
T e m p e r a t u r e ( K )
h h X
l h X
5 n m G a A s Q W s1 0 n m A l G a A s Q W s
E g ( T ) + 0 . 0 8 e V
G a A s :
Figure 2.3: Temperature dependence of the exciton peaks measured by PL and ab-sorption. The heavy hole exciton (HHX, closed points) peaks and light hole exciton(LHX, open points) redshift with increasing temperature as measured both by PL andabsorption. Exciton peaks in PL are no longer observed above approximately 200 K,but absorption remains observable up to room temperature. The trend almost exactlyfollowing the bulk GaAs band gap dependence (plotted shifted up 0.08 eV for clarity).Because the hole masses do not depend on temperature, the HH-LH splitting remainsconstant through the temperature range, although the light hole excitons become diffi-cult to resolve at the higher temperatures. The data was published in the SupplementaryMaterials of [4]. Copyright 2013 by the American Physical Society.
up to room temperature, see Fig. 2.4. The heavy hole-light hole splitting remains constant
with temperature, at least as far as light hole excitons are observable.
2.2 Sideband measurements
2.2.1 Optical setup
Sidebands were generated by collinear illumination of the sample with a NIR and THz
beam. The NIR beam was generated a continuous-wave Ti:sapphire laser (Coherent 890).
The THz beam was generated by cavity dumping the millimeter beam line of the FEL
32
1 . 5 0 0 1 . 5 2 0 1 . 5 4 0 1 . 5 6 00 . 0
0 . 5
1 . 0
*
5 n m Q W s 1 0 n m Q W s
Signa
l (arb.
units)
F r e q u e n c y ( e V )
A b s o r b a n c e
T = 2 9 4 K*
8 2 0 8 1 0 8 0 0 7 9 0W a v e l e n g t h ( n m )
Figure 2.4: Room temperature absorption spectra. The heavy hole exciton lines stilldominate the absorption spectra of both samples, but the light hole excitons (denotedby the *), are still observable. The half linewidth of the heavy hole exciton line is 4 meVin the 5 nm GaAs QWs and 5 meV in the 10 nm AlGaAs QWs. The data was publishedin [4]. Copyright 2013 by the American Physical Society.
tuned to 570 GHz (2.35 meV). This frequency was chosen because of its high power and
the accelerator’s stability. The laser field from both sources was linearly copolarized and
focused onto the same spot. Radii were approximately 100 µm and 0.5 mm for the NIR
and THz spots, respectively. The strength of the terahertz electric field was 11.5 kV/cm
inside the quantum wells. The power of the NIR laser was 39 mW (60 mW) at T = 12
K (294 K). The transmitted NIR laser power and generated sidebands were sent via free
space to a monochromator and detected by a photomultiplier tube (PMT). Sidebands
with order n ≤ 16 were measured by the average current output from the PMT during
the 40 ns FEL cavity dump pulse. Sidebands of order n > 16 were too weak to detect
by averaging the current, so the PMT signal from sideband orders n ≥ 14 was read
out with a high-speed photon counter. The overlap between analog readout and photon
33
counting scaled the two together. Finally, the sample was removed from the NIR beam
path and the laser line was heavily attenuated and measured with the PMT to determine
the sideband creation efficiency for the full sideband spectrum measurements.
2.2.2 Results
Sideband spectra spanning two dozen orders and tens of meV were observed from both
samples. In the 5 nm GaAs QW sample at 12 K, we observed down to -8th order
below and up to 28th order above the NIR laser, see Fig. 2.6(top left). The intensity
of the sidebands with negative order decreases exponentially with n. The intensity of
the sidebands with positive order decreases much more gently with no clear systematic
dependence on n. We observed a similar spectrum of sidebands in the 10 nm AlGaAs
QW sample, 2.6(top right). However, the intensity of each sideband is approximately 3
times smaller than the corresponding sideband in the 5 nm GaAs QW sample, and the
highest sideband is 22nd order rather than 28th.
Remarkably, HSG persists to room temperature in both samples. Sidebands with
orders as high as 22nd and 18th for the 5 nm GaAs and 10 nm AlGaAs QW samples,
respectively, were observed 2.6(bottom row). The sideband conversion efficiencies are
approximately one order of magnitude smaller at 294 K compared to 12 K, but otherwise
the HSG spectra have similar shapes.
The NIR frequency, fNIR dependence was also measured to both peak up the sideband
signal, and as an initial experiment to understand the “zeroth” step of HSG, exciton
34
AOM
Wavelengthmonitor
CW Ti:sapphire laser
Fast pyro
Mylar beamsplitter
Cryostat
Focusinglens
Collimatinglens
OAP with NIR laser hole
Long pass filter
Monocromator + PMT
Sample
Beamdump
Lens focused on entrance slits
Figure 2.5: Experimental schematic of the sideband measurement. First, we follow thepath of the excitation NIR light, the thin red line. The Ti:sapphire laser emits a contin-uous NIR beam. A pellicle splits off a few percent of the light and into a fiber-coupledspectrometer to measure the NIR wavelength. To keep the laser from overheating thesample, the laser is electronically shuttered by an AOM, triggered only for the duration ofthe FEL pulse. The zeroth order beam is sent into a beam dump. The NIR laser is thenfocused by a lens before traveling through a small hole bored in the off-axis parabolicmirror (OAP) to become collinear with the THz beam. The cryostat window is madeof a NIR- and THz-transparent plastic, TPX. The laser spot on the sample is about200 µm in diameter. The THz path, the thick green line with white arrows, leaves theFEL transport system exit port. Immediately about 5% of the beam is picked off by athin mylar beamsplitter and focused with an OAP onto a fast pyroelectric detector tomonitor the THz pulse shape. The majority of the beam gets focused onto the sampleby a 5” OAP onto the same location as the NIR beam. Sidebands are then sent througha collimating lens before being focused onto a monochromator and detected a PMT.
35
1 0 - 1
1 0 1
1 0 3
F T H z = 1 1 . 5 k V / c mf T H z = 2 . 3 5 m e V
T = 1 2 K
Signa
l (arb.
u.)
f N I R = 1 . 6 2 3 e V
1 . 6 0 0 1 . 6 2 0 1 . 6 4 0 1 . 6 6 0 1 . 6 8 0 1 . 7 0 0
1 0 - 1
1 0 1
1 0 3 1 0 n m A l G a A s Q W s
S i d e b a n d f r e q u e n c y ( e V )
1 . 6 4 1 0 1 . 6 4 1 5 1 . 6 4 2 00
5
1 0
Signa
l (arb.
u.)
n = 8
7 7 0 7 6 0 7 5 0 7 4 0 7 3 0W a v e l e n g t h ( n m )
5 n m G a A s Q W s
7 5 5 . 4 7 5 5 . 2
Figure 2.6: Measured sideband spectra at T = 12 K. The sideband sideband strengthread out by the PMT versus frequency for the 5 nm QW sample at T = 12 K (top left)and for the 10 nm QW sample at T = 12 K (bottom left) are plotted as blue circles.The hatched region represents the noise floor. Sideband order n = 8 is plotted on theleft to show the narrow width of the individual sidebands. This data was published in[4]. Copyright 2013 by the American Physical Society.
36
creation, see Fig. 2.7. The NIR laser frequency was scanned across as many as 60 meV
(30 nm) in approximately 1.5 meV steps. At every fNIR step, sideband orders n = 2, 4,
and 6 were measured. At room temperature, the NIR laser could scan over the entire
exciton absorption peaks region. At low temperature, however, the laser was unable to
tune over the 5 nm GaAs QW light hole exciton. In all samples at all temperatures,
there is no evidence of sideband generation when fNIR is deep in the band gap. Without
a finite density of states, no excitons can be created for HSG. Even χ(3) processes will
be weak because there is no resonant enhancement even from exciting the excitons. The
high-frequency “in-the-band tail” is much longer than the low-frequency “in-the-gap”
tail because electron and holes are still generated, though not bound as excitons but as
free pairs. They can even still be driven to recollide, but with a lower overall probability
[63].
The fNIR dependence near the exciton lines has many peaks depending on the sideband
order for both samples and temperatures. The n = 2 sideband at low temperature in both
samples is peaked approximately 5 meV below the heavy hole exciton (HHX) absorption
peak, see 2.7(a, c). The HHX peak will redshift slightly in the strong THz field due to
AC Stark effect [46], but not enough to explain the full 5 meV. The THz frequency is 2.35
meV, so the 5 meV offset is likely due to resonant enhancement of sideband generation
by the HHX absorption peak and has been seen before in HSG from bulk GaAs and
perturbative sideband generation in QWs [34, 61, 71] As the sideband order increases,
though, a secondary peak grows midway between the two exciton absorption lines in both
37
samples, eventually becoming the dominant peak, likely because resonant enhancement is
not as large an effect at higher orders. At room temperature, the two peaks are smeared
out. In the 5 nm GaAs QWs, all three orders are peaked just below the heavy hole
exciton, and the signal falls off slowly as fNIR is tuned across the light hole exciton. In
the 10 nm AlGaAs QWs, the high-frequency tail falls off much more quickly because the
hole splitting is much smaller.
2.3 Theoretical model
2.3.1 Three step model: Classical interpretation
Many features of the data can be explained qualitatively by adapting the three-step model
for HHG from atoms in intense laser fields to the phenomenon of HSG. Step zero, which
is not present in the case of HHG for atoms, is the creation of the electron-hole pairs by
the NIR laser at a particular phase of the THz field. For these experiments, the NIR laser
is tuned to the exciton resonance stemming from the highest heavy hole subband and the
lowest electron subband, thus creating carriers with in-plane masses of m∗e = 0.067me
and mh = 0.103me for the electron and hole, respectively (The hole mass is appropriate
for a 5 nm GaAs QW within the Kohn-Luttinger model [42]). Figure 2.8 illustrates the
next three steps that lead to the emission of the 28th order sideband. In the first step,
the exciton tunnel ionizes in the strong THz field at a phase 54 after the maximum of
the terahertz field. At this point, the carriers are assumed to be point particles with zero
38
0 . 0
0 . 5
1 . 0
Sideb
and s
ignal
(arb.
u)
( a )
1 . 6 0 0 1 . 6 2 0 1 . 6 4 00 . 0
0 . 5
1 . 0
N I R e n e r g y ( e V )
( c )
5 n m G a A s Q W s
( b )
1 . 5 0 0 1 . 5 2 0 1 . 5 4 0 1 . 5 6 0
T = 1 2 K1 0 n m A l G a A s Q W s
n = 2 n = 4 n = 6
A b s .
T = 2 9 4 K
( d )
7 7 0 7 6 5 7 6 0 7 5 5 7 5 0 8 2 0 8 1 0 8 0 0 7 9 0W a v e l e n g t h ( n m )
Figure 2.7: Low-order sidebands vs. NIR frequency at low temperature and room temper-ature. For each plot, the low order sideband spectra were taken at several NIR frequenciesaround the two exciton lines of the given sample. The zero-field absorption spectrum isthe filled region. The sideband data for each order are normalized to strongest of thatorder. (a) The sideband signal is peaked near the heavy hole exciton line of the 5 nmGaAs QWs at low temperature. The n = 2 sideband is most strongly peaked when fNIR
is about 5 meV below the heavy hole exciton, but n = 4 and 6 are peaked very broadly10 meV above the heavy hole exciton. Sideband creation drops off quickly as fNIR istuned into the band gap. (b) The same sample at room temperature has a much broaderpeak around the heavy hole exciton that is similar in each sideband. (c, d) The sidebandsignal in the 10 nm AlGaAs QWs shows a very similar fNIR dependence to the narrowquantum wells, but the peaked region is much smaller because the heavy hole-light holesplitting is much smaller. The small black arrow denotes the fNIR used for full sidebandspectrum measurements, Figs. 2.6 and 2.10.
39
initial momentum. The Keldysh parameter, γ = ωTHz
√2µ∗Ebind/ (eFTHz), the ratio of
the average tunneling time to the period of the driving field, is 0.2, well within the regime
of tunnel ionization [32, 26]. Here, µ∗ = 0.041me is the reduced mass of the electron-hole
pair, Ebind is the exciton binding energy, e is the electric charge, and ωTHz and FTHz are
the angular frequency and electric field strength of the terahertz radiation, respectively.
In the second step, the electron and hole are assumed to be independent and to accelerate
in the THz field classically. In the third step, the electron and hole recollide to emit an
NIR photon with larger energy than the photon that created them.
The high-order sidebands with positive n can be analyzed to extract the kinetic energy
of the electron-hole system at recollision. The sideband energy offset is given by ESB =
EK,e+EK,h+∆,where EK,i is the kinetic energy for the ith particle and ∆ is the detuning
of the NIR laser from the 2D band gap. Because this is an excitonic effect, ∆ should be
close to the binding energy of the exciton, approximately 9 meV for the 5 nm GaAs QWs
and 8 meV for the 10 nm AlGaAs QWs [24]. The 28th order sideband is 66 meV above
the NIR laser energy. Following this detuning assumption, the total kinetic energy of
the electron and hole is ∼57 meV. This energy is significantly higher than the threshold
for a charge carrier to emit a longitudinal optical (LO) phonon. Figure 2.8 illustrates
the classical trajectories (top) and kinetic energy (bottom) for the electron and hole
associated with the 28th order sideband between ionization at t = 0 and recollision. The
kinetic energy of the electron-hole system is above the LO phonon energy only for the final
50 fs before reollision. Since this is below the typical electron-LO phonon scattering time
40
- 2 00
2 04 0
0 . 0 0 . 1 0 . 2 0 . 3 0 . 4 0 . 5 0 . 60
3 06 0
H o l ePositi
on (n
m)
E l e c t r o n
Total
kineti
cen
ergy (
meV)
T i m e a f t e r t u n n e l - i o n i z a t i o n ( p s )
E S B∆
E L O
Figure 2.8: Classical calculation of electron-hole dynamics associated with 28th ordersideband. Top: trajectories of the electron (black) and hole (red) show electron-holeseparation reaches a maximum of 60 nm between tunnel ionization (0 ps) and recollision(flash). Bottom: total kinetic energy of the electron and hole. The total kinetic energyexceeds the threshold for LO phonon emission ELO (short dashed line) for only 50 fsbefore emitting a sideband with sideband offset energy ESB (long vertical arrow) = Totalkinetic energy + detuning ∆ (double arrow). The long dashed horizontal lines denote 0nm (top) and 0 meV (bottom). This calculation was published in [4]. Copyright 2013 bythe American Physical Society.
of 100 fs, a large fraction of electron-hole pairs are likely to recollide before they emit a LO
phonon—even though the electron-LO phonon scattering time is much shorter than the
period of the terahertz radiation. This analysis suggests that the fundamentally ballistic
process of HSG persists to room temperature because thermal fluctuations simply do not
have enough time to disrupt electron-hole recollisions.
Although there is no abrupt cutoff due to LO phonons, we observe far fewer side-
bands than are predicted by the ballistic three-step model of HSG. This model predicts
a sharp cutoff for the sideband offset energy: ~ωcutoff = 3.17UP + Ebind, where UP is the
ponderomotive energy [= e2F 2THz/(4µ
∗ω2THz)]. Such a cutoff is observed for the harmon-
41
ics in high-order harmonic generation experiments on atoms. Inserting the parameters
appropriate for our experiments, the ballistic three-step model predicts a cutoff 130th
order, approximately 4 times larger than the highest-order sideband reported here.
2.3.2 Full theoretical results
To investigate this discrepancy, our collaborators Fan Yang and Renbao Liu have also
performed the simplest calculation that captures the relevant quantum mechanical na-
ture of the recollision process together with the most important scattering mechanisms,
exciton dephasing and electron-LO phonon scattering. In this calculation, the intensity of
the nth order sideband is proportional to the integral in Eq. 2.2, which can be integrated
numerically and is plotted in Figs. 2.9 and 2.10.
Pn ∝∫eiS(k,t,τ)e−γτe−W(d,T,k) τ dk dt dτ (2.1)
S (k, t, τ) = −∫ t
t−τ
1
2µ∗[k + eA(t′′)]2 dt′′ −∆τ + nωTHzt. (2.2)
Here, S(k, t, τ) is the action, where µ∗, e and ωTHz were defined previously, k is the
momentum (working in units where ~ = 1), and A is the vector potential of the oscillating
terahertz field. The exciton is created at t−τ , the electron and hole evolve through τ , and
eventually annihilate at t, emitting sidebands with order n. The detuning ∆ is assumed
to be Ebind. Eq. 1 represents an extension of the model used in Ref. [41] to include the
exciton dephasing rate γ, which was estimated from the half-lidewidth of the heavy-hole
42
absorption peak (2.8 meV for both samples at 12 K; 4 meV and 5 meV for the 5 nm GaAs
and 10 nm AlGaAs QWs at 294 K, respectively), and the scattering rate due to emission
or absorption of LO phonons W (d, t,k). W is calculated from Ref. [13] and depends on
the QW width d, the temperature T and the kinetic energy |k|2 /(2µ∗) of the carrier,
which is assumed to be an electron with mass µ∗. In the calculation, ωTHz = 2π × 0.58
THz (photon energy ETHz = 2.4 meV), FTHz = 11 kV/cm, and the photon energy of
the NIR laser ENIR = 1.623 eV. The LO phonon energy is taken as 36.6 meV. In order
to isolate the effects of scattering on high-order sideband generation, many interactions,
such as the Coulomb interaction and heavy hole-light hole mixing, have been neglected.
These calculations refine the simple idea that HSG works “too fast” to be affected
by LO phonons. At low temperature, sideband intensities calculated with LO phonon
scattering begin to fall below those calculated without such scattering very close to
the threshold for LO phonon emission (see Fig. 2.9a). Above this threshold, sideband
intensities are suppressed but not eliminated. At room temperature (Fig. 2.9b), all
positive sidebands are suppressed because there is now a finite chance of absorbing an
LO phonon from the thermal bath. As at low temperatures, sidebands persist above the
threshold for LO phonon emission.
Calculations are compared with experimental data in Fig. 2.10. At low tempera-
ture, the calculations overestimate the intensities of high-order sidebands for positive
n relative to the low-order positive sidebands. At both temperatures, the calculations
approximate the observed exponential decay with n of the sidebands below the NIR laser
43
1 0 - 71 0 - 61 0 - 51 0 - 41 0 - 31 0 - 21 0 - 11 0 0
0 1 0 0 2 0 0 3 0 01 0 - 71 0 - 61 0 - 51 0 - 41 0 - 31 0 - 21 0 - 11 0 0
a )
w i t h L O
Relat
ive st
rength
(a.u.
)
T = 1 2 K w i t h o u t L O
b )
0 1 0 0 2 0 0 3 0 005
1 0
γ LO (
ps-1 )
K i n e t i c e n e r g y ( m e V )
S i d e b a n d e n e r g y o f f s e t ( m e V )
T = 2 9 4 K
0 1 0 0 2 0 0 3 0 005
1 01 5
γ LO (
ps-1 )
K i n e t i c e n e r g y ( m e V )
0 5 0 1 0 0 S i d e b a n d o r d e r 2 n
Figure 2.9: Calculations of the effects of longitudinal optical phonon scattering. HSGcalculations at (a) T = 12 K and (b) T = 294 K with (black circles) and without (redtriangles) LO phonon scattering. An arrow denotes the threshold for emitting a LOphonon in the 5 nm QW sample. Excitonic dephasing is included in all calculations.The inset is the calculated total energy-dependent scattering rate given by[13]. Thiscalculation was published in [4]. Copyright 2013 by the American Physical Society.
44
1 0 - 9
1 0 - 7
1 0 - 5
1 0 - 3 h f N I R = 1 . 6 2 3 e V
h f N I R = 1 . 5 3 1 e V
T = 1 2 K
Sideb
and c
onve
rsion
effici
ency
h f T H z = 2 . 3 5 m e V
- 2 0 0 2 0 4 0 6 01 0 - 1 0
1 0 - 8
1 0 - 6
1 0 - 4
T = 2 9 4 K
S i d e b a n d e n e r g y o f f s e t ( m e V )
T = 2 9 4 K E x p e r i m e n t C a l c u l a t i o n
- 2 0 0 2 0 4 0 6 0
T = 1 2 K- 1 0 0 1 0 2 0 3 0
1 0 n m Q W s
S i d e b a n d o r d e r 2 n
5 n m Q W s
- 1 0 0 1 0 2 0 3 0
Figure 2.10: Theory vs. experiment. The sideband sideband strength read out by thePMT versus sideband energy offset (hfSBhfNIR) for the 5 nm QW sample at T = 12K (top left) and T = 294 K (bottom left) and for the 10 nm QW sample at T = 12K (top right) and T = 294 K (bottom right) are plotted as blue circles. The hatchedregion represents the noise floor. The vertical arrows represent the energy offset when therecolliding electron and hole have enough kinetic energy to emit a longitudinal opticalphonon. This data was published in [4]. Copyright 2013 by the American PhysicalSociety.
line. However, at room temperature, the calculated intensities of the positive sidebands
agree remarkably well with experiment. The agreement of experiment and theory at
room temperature suggests that the decoherence at room temperature is so strong that
it dominates the physics missing from the model.
45
2.4 Conclusion and remaining questions
We have studied HSG in two QW samples demonstrating that electron-hole recollisions
in semiconductors are surprisingly robust, though still affected, to electron-lattice inter-
actions. In both samples, sidebands were generated from electron-hole pairs that had
more than enough kinetic energy to emit an LO phonon. Large sideband spectra were
observed in both samples at even room temperature, where the thermal population of
LO phonons is high enough that phonon absorption by electrons becomes probable. In
the 5 nm GaAs QWs at room temperature, we observed sidebands from electron-hole
pairs capable of emitting LO phonons. The classical three step model demonstrates that
recollisions happen too quickly to be affected by phonon scattering. A more complete
quantum saddle point calculation that includes both phonon scattering and excitonic
dephasing agrees very nicely with the data. These results imply that further reductions
in scattering and dephasing could increase the number of observed sidebands.
46
Chapter 3
The Sideband Spectrometer
High-order sideband generation (HSG) experiments demand a great deal of a spectrome-
try set up. Terahertz (THz) and near-infrared (NIR) wavelengths differ by three orders of
magnitude, so few optics work in both wavelength ranges. Because the THz wavelengths
are so long, and the focal spot must be as small as possible to maximize the electric field,
very large optics and lenses must be used which get in the way of NIR optics. On the
detection side, NIR sensitivities must be very high, as the highest order sidebands can
have creation efficiencies less than 10−10 and the cavity-dumped free electron laser has
a duty cycle of less then 10−7. Furthermore, HSG depends a lot on the material to be
studied, so basic optical characterization measurements must be able to be performed in
the same space to avoid unnecessary duplication of expensive equipment and software.
47
3.1 Optical setup
A complete optical measurement apparatus has been developed for use between 700 and
1000 nm. Linear experiments such as photoluminescence (PL) and differential transmis-
sion (DT) to measure absorption can characterize potential samples. Then nonlinear
THz experiments such as sideband generation and driven absorption will probe the light-
matter interaction at THz scales. Both excitation and detection optical paths have been
streamlined to make switching between measurement techniques fast. The THz beam
path has also been simplified from previous experiments to make initial alignment easy
and very stable. Samples are mounted on a three-axis translation stage in a closed-cycle
cryostat with a temperature range of 15–295 K. The sample translation stage can be
manipulated even at low temperature using specialized feedthroughs. The entire cryo-
stat itself can be translated linearly by as much as a foot to switch between samples,
measurements, or alignment. The front window of the cryostat is made of NIR- and
THz-transparent suprasil quartz, although TPX plastic or c-cut sapphire are also useful
windows. The back window is glass that has been anti-reflection coated for the NIR.
3.1.1 Linear measurements
The two linear measurements, PL and DT, are very important for understanding the
properties of a candidate material or processing technique for sideband experiments. It
is extremely important that these can be performed rapidly without misaligning the
more challenging nonlinear THz measurements. The optical table has been set up to
48
Collecting lens
HeNe laser
Sample
Cryostat
Collecting lensDetection
fiber
Photoluminescence
Differential Transmission
Focusing lens
Fiber from tungsten-halogen lamp
Pinhole spatial filter
Focusing lens
Detection fiber
Spectrometer + CCD
Spec. couplingoptics
Figure 3.1: The optical schematic for photoluminescence (PL) and differential transmis-sion (DT) measurements. Photoluminescence: A HeNe laser is focused on the samplein a cryostat at an angle of about 30. A collecting lens is placed as close to the sampleas possible to maximize signal. The PL emission is then coupled into a detection fiberwhich brings the light to an imaging spectrometer to be measured by a CCD. Differ-ential Transmission: Light from a tungsten-halogen lamp is brought by a fiber to apinhole spatial filter. The collimated light from the spatial filter is focused on the sam-ple. Transmitted light is collected by lens and coupled to a fiber to be brought to thespectrometer.
incorporate these spectroscopic tools alongside the nonlinear measurements.
Photoluminescence
As described in chapters 1 and 2, PL is a particularly attractive optical measurement
because of its simplicity. The improved apparatus is shown in Fig. 3.1(top). The whole
cryostat can be translated linearly between optical setups so that the optics to measure
PL can remain set up while other experiments are being performed in the THz beam
49
1 . 5 4 5 1 . 5 5 0 1 . 5 5 5 1 . 5 6 0 1 . 5 6 50 . 0
0 . 5
1 . 0
PL sig
nal (a
rb. u.
)
F r e q u e n c y ( e V )
T = 1 5 K
1 0 n m G a A s Q W s , u n p r o c e s s e dλe x = 6 3 2 n m
8 0 2 8 0 0 7 9 8 7 9 6 7 9 4 7 9 2W a v e l e n g t h ( n m )
Figure 3.2: A PL spectrum from undoped quantum wells (QWs) as-grown. A HeNe laserexcites the PL, λex = 632 nm. The single peak results from radiative recombination ofheavy hole excitons. The small shoulders in the peak result from well-width fluctuationsin the QWs distributed across the ten-well stack and the 200 µm laser spot. At 15 K,there is no significant population of light hole excitons about 10 meV above the heavyhole exciton peak.
path. New samples can be quickly tested to determine if they are good enough for
further experiments. A 5 mW HeNe laser (λHeNe = 632 nm) is focused on the sample at
an angle of about 30 to the sample normal with a spot size of about 200 µm. The excited
electron-hole pairs luminesce in all 4π steradians, but because of Snell’s law most of the
light that escapes does so normal to the sample surface. The luminescence is collected by
a lens with the shortest focus allowable, usually 60 mm, and fiber-coupled to an imaging
spectrometer to be measured by a Peltier-cooled, electron-multiplying CCD array. An
example PL spectrum can be seen in Fig. 3.2.
50
Differential Transmission
Undriven DT is slightly more complicated than PL, but the resulting absorption spec-
tra are more directly relevant to sideband experiments. To truly measure absorption
requires measuring both the transmitted and reflected components of the incident light.
In practice measuring absorption only with differential transmission is sufficient as the
reflected component is generally small and independent of wavelength. The absorbance
of a sample αsamp is the negative log of the light transmitted through the optics with and
without the sample:
αsamp(λ) = −10 logTsamp
Tref
(3.1)
where Tsamp(λ) and Tref are the amount of light transmitted by the sample and by the
reference (no sample), respectively, and αsample is the absorbance of the sample, reported
in dB. Differential transmission can be difficult if the sample has other NIR-absorbing
layers or if the light reflected off the front of the sample is larger than the light absorbed.
Sideband measurements already require sample substrates to be transparent to NIR light,
so that requirement must be met anyway.
The optical schematic for measuring DT can be seen in Fig. 3.1(bottom). The light
source is a tungsten-halogen lamp coupled to a multimode optical fiber. The lamp pro-
vides ample light so that the spot can be spatially filtered using a pinhole to minimize
the spot on the sample. The image of the pinhole on the sample is four times larger than
the pinhole itself. Pinholes as small as 10 µm in diameter are trivial to align, and enough
light is focused through the sample to get plenty of signal. The beam is then collected
51
1 . 5 4 0 1 . 5 5 0 1 . 5 6 0 1 . 5 7 0 1 . 5 8 02
4
6
8
1 0
Abso
rption
(dB)
F r e q u e n c y ( e V )
R a w a b s o r p t i o n L o w - p a s s f i l t e r e d
1 0 n m G a A s Q W sT = 1 5 K
8 0 5 8 0 0 7 9 5 7 9 0 7 8 5W a v e l e n g t h ( n m )
Figure 3.3: A raw and processed absorption spectrum of QWs epitaxially transferred toa sapphire substrate. The raw spectrum has a fast Fabry-Perot oscillation on top of thedesired absorption spectrum. Applying a low-pass filter in software removes the Fabry-Perot oscillation with no significant artifacts. The QWs are the same as in Fig. 3.2, andthe same shoulder in the heavy hole exciton peak can be resolved post-filtering. The lighthole exciton can also be clearly seen at about 1.565 eV. (This an example Abs spectrumfrom 1/12/16)
after the sample and coupled to a fiber sent to the imaging spectrometer to be measured
by the CCD.
An absorbance spectrum measured by DT can be found in Fig. 3.3. The primary
artifact in the spectrum, the high-frequency sinusoid on top of the whole spectrum, is di-
rectly related to the fact that Rsample was not measured. Samples with extremely smooth,
transparent substrates exhibit Fabry-Perot interference that modulates the transmission
and reflection coefficients as a function of wavelength. The highly polished, 500 µm-thick,
c-cut sapphire substrates have a Fabry-Perot fringe period of about 0.7 meV with an am-
plitude of approximately 2 dB overlaid on the actual transmitted spectrum. Because
this period is much smaller than even the narrowest excitonic absorption lines measured,
52
the fringes can be removed in software by short-pass Fourier filtering. With the fringes
removed, the heavy hole and light hole excitons become readily apparent. To measure
narrow resonances or where the signal-to-noise is not high enough to accurately measure
the fringes, depositing an anti-reflection coating on the backside of samples is necessary.
3.1.2 Nonlinear measurements
Nonlinear THz-optical measurements allow for the THz-material interaction to be mea-
sured using far more mature optical technology. Sideband experiments, such as HSG,
can probe quasiparticle dynamics at picosecond-times at the single-quasipartcle level.
Measuring differential NIR transmission in a strong driving THz field (driven differential
transmission, or DDT) can probe THz-exciton interactions.
Sideband measurement
First, the NIR and THz light must be aligned collinearly on the sample, see Fig. 3.4.
Starting with the NIR beam path, a continuous-wave Ti:sapphire laser outputs approxi-
mately 1 W of laser power across its entire tuning range, 700–1000 nm. To safely atten-
uate the beam without introducing thermal lensing, a rotating half-wave plate (HWP) is
placed before a Glan-laser polarizer. The HWP controls how much of the NIR light trans-
mits through the polarizer. The reflected portion of the beam is dumped in a beam dump
(BD). A pellicle beam splitter (BS) then picks off approximately 8% of the laser towards
a multimode fiber coupler connected to a small spectrometer to monitor the wavelength
of the NIR laser. The NIR laser then passes through an acousto-optic modulator (AOM)
53
FEL output port
Pyro detector
Mylar B
S
Polarizerattenuator
OAP mirror
ITO BCShutter
PolarizerHWP
to w
avel
ength
mon
itor
AOMFocusing lens
LP filter
BD
CW Ti:sapphire NIR laser
Sample
HWP
BD
Cryostat
Horncoupler
Pelli
cle
BS
Figure 3.4: The optical schematic for the NIR and THz excitation paths. The continuouswave (CW) Ti:sapphire laser is tuned to a particular NIR wavelength. A rotatablehalf-wave plate (HWP) followed by a linear polarizer provides consistent attenuation.Power reflected by the polarizer is sent into a beam dump (BD). A pellicle beamsplitter(BS) picks off a few percent of the laser power that is sent to a wavelength-monitoringspectrometer. The laser then goes through an acousto-optics modulator (AOM). Thezeroth order beam is sent to a BD, the first order beam then goes through a longpass(LP) filter and an achromatic focusing lens. A second HWP controls the direction oflinear polarization at the sample. A mechanical shutter blocks the beam when the FELis not lasing. Finally, the laser transmits through the indium-tin oxide (ITO) beamcombiner (BC) to be made collinear with the FEL. The THz beam path starts at theFEL output port. A mylar BS picks off some of the power and an off-axis parabolic(OAP) mirror focuses the beam on a horn-coupled pyroelectric (pyro) detector. Themajority of the beam passes through a wire-grid polarizer attenuator. Finally, the THzgets focused down with a second OAP mirror, and the ITO beam combiner reflects theTHz beam into the cryostat, collinear with the NIR laser. Both lasers are focused on thesample.
54
that picks off as much as 60% of the beam into the first-order diffracted beam. The first-
order beam has an on-off contrast ratio of 3000:1 and a rise and fall time of approximately
200 ns to quickly shutter the NIR laser so that it is only on during the FEL pulse. The
zeroth-order beam is sent to a BD. The first-order beam is then sent through a long-pass
filter. The sidebands are weak enough that amplified spontaneous emission from the NIR
laser can overwhelm the sideband signal, so it must be filtered out. The NIR laser is then
gently focused with a 500 mm achromatic lens. A rotatable achromatic HWP controls
the linear polarization of the beam, then the beam passes through a mechanical shutter.
The NIR laser then passes through a glass slide coated with indium-tin oxide (ITO) that
serves as a beam combiner (BC) and the suprasil cryostat window before being focused
on the sample with a 200 µm-diameter spot.
The THz beam path is far simpler. The THz beam exits the FEL transport system
output port before being split by a mylar BS, see App. B. About 5% of the THz power
is picked off and sent to an integrating pyroelectric detector to monitor and count FEL
pulses. The majority of the horizontally-polarized beam then passes through two wire-
grid polarizers. The first rotates to tunably attenuate the THz beam, and the second
is fixed to pass the remaining horizontally-polarized component. The THz laser is then
focused by an off-axis parabolic (OAP) mirror to a spot of about 1 mm in diameter.
Before the focus, the THz beam reflects off the ITO BC to become collinear with the
NIR beam. The ITO BC is mounted on a tip-tilt stage so that THz alignment can be
peaked up without disturbing the sensitive focusing optics.
55
HWPCollimating
lens
OAP mirror
ITO BC
Monochromator + PMT
Sample in cryostat
ND filters (optional)
NIRlaser
Laser +SBs
THzlaser
Figure 3.5: The optical schematic for sideband PMT detection path. Sidebands arecreated with a similar beam profile as the NIR laser. They are collimated by a lensafter the sample, and the polarization is rotated using the HWP for maximum gratingefficiency in the monochromator. Because PMT is easily saturated, calibrated neutraldensity (ND) filters allow for measurements of the NIR laser line and the strong low-ordersidebands.
If NIR-THz alignment is good and the experiment is well-designed, sidebands around
the NIR wavelength will be generated. There are two detectors used to measure the
sidebands, depending on their proximity to the NIR laser wavelength, a photomultiplier
tube (PMT) and a CCD. For low-order sidebands, n ≤ 10, the PMT, mounted on a
double-grating monochromator, is used. The monochromator has a very high rejection
of light more than 0.1 nm from the selected wavelength (about 100 GHz at 750 nm),
allowing for sensitive detection of low order sidebands generated from intense THz light
with frequency as low as 240 GHz. It is also a very sensitive, very fast detector, with a
single photon rise and fall time of about 10 ns. In addition, it allows for rapid THz-NIR
alignment by maximizing the signal from low-order sidebands because of its fast readout
(n = 8 is approximately the highest order that can be used). The primary drawback
of the PMT is that it is a single channel detector, so integrating enough to get good
56
7 5 0 7 6 0 7 7 0 7 8 01 0 - 4
1 0 - 2
1 0 0
1 0 2
1 0 4
1 0 6 5 n m G a A s Q W sSc
aled P
MT sig
nal (V
)
W a v e l e n g t h ( n m )
λN I R = 7 6 4 n mP N I R = 5 0 m Wf T H z = 5 4 0 G H zF T H z = 3 5 k V / c mT = 1 5 K
N I R l a s e r
1 . 6 5 0 1 . 6 4 0 1 . 6 3 0 1 . 6 2 0 1 . 6 1 0 1 . 6 0 0 1 . 5 9 0
- 1 4 t h
F r e q u e n c y ( e V )
1 2 t h
Figure 3.6: Example sideband spectrum measured by the PMT. Sidebands spanningfrom orders −14 ≤ n ≤ 12 are measured. The sideband spectrum does not end at +12,but the CCD is a better detector for higher-order sidebands. The noise floor can seenin the -12th and -14th sidebands, about 5 mV. Sidebands closer to the NIR laser, andthe NIR laser itself, are much stronger and must be attenuated, raising the local noisefloor substantially. The strength of an individual sideband is measured by integratingthe signal around a peak.
statistics is difficult.
To get to the PMT, the NIR laser and sidebands are collimated by another 500
mm achromatic lens and the polarization is rotated to horizontal by another rotating
achromatic HWP, see Fig. 3.5. Before being focused on the monochromator slits with
a 25 mm lens, the laser and sidebands may have to be attenuated. The PMT signal is
amplified with a 300 MHz bandwidth amplifier and read out by an oscilloscope with two
readout “modes,” integrating mode and photon-counting mode. In integrating mode, the
time trace is sliced to sum the PMT signal just when the FEL is active. The noise floor
of integrating mode is a few nanowatts and has a dynamic range of only about 15. To
57
measure strong signals such as the NIR laser (> 10 mW) or low order sidebands (10–100
µW), the PMT gain is turned down and the light is attenuated with calibrated neutral
density (ND) filters placed just before the final lens. By measuring the NIR laser with
a strong enough attenuator, the PMT response can be calibrated so that the sideband
efficiency, the number of nth order sideband photons created per NIR laser photon, can
be measured, see Fig. 3.6 for an example sideband spectrum. The monochromator then
scans over a sideband, one wavenumber (1/cm) at a time. It takes approximately 70
FEL pulses to measure a single sideband with decent statistics, but single FEL pulses
can be used to peak up the alignment by tweaking the ITO BC. When PMT signal is very
small, it can be used in photon counting mode, where a fast peak detector (in software
or hardware) is used to count the 50 mV spikes characteristic of a single photon from the
PMT output. Weaker sidebands can be photon counted, but a better detector for these
sidebands is the CCD.
The optical schematic for measuring a sideband spectrum with the CCD shares many
features with the PMT setup, see Fig. 3.7. The initial optics for both the NIR laser
and the THz laser are identical, but now the shutter is engaged. Even with the AOM
contrast ratio maximized, there is about a factor of 103 more NIR laser light between
FEL pulses than during. By using a mechanical shutter with an infinite contrast ratio
and a open-close time of 10 ms, the excess light can be brought down to an acceptable
factor of ten larger.
The CCD detection path changes as well. The achromatic HWP after the collimating
58
HWPCollimating
lens
OAP mirror
ITO BC
Sample in cryostat
Polarizer
SP filter
BD
Spectrometer + CCD
NIRlaser
Laser +SBs
THzlaser
Figure 3.7: The optical schematic for the sideband CCD detection path. Similar to thePMT detection path, the sidebands are collimated by the same lens and the polarizationis rotated using the HWP for maximum efficiency through the spectrometer. To filter outthe NIR laser line so that it doesn’t overwhelm the sideband signal, a polarizer cleansup the sideband polarization and a short pass (SP) filter attenuates the laser line byapproximately 70 dB without affecting sidebands above approximately sixth order.
lens rotates the NIR laser and sideband polarizations to vertical. Then, to ensure max-
imum NIR laser rejection by the angle-tuned short pass (SP) filter, the NIR laser and
sidebands are sent through a polarizer. Angle-tuning optical edge filters does not move
the edge for both s- and p-polarizations at the same rate, so a final polarizer is needed to
achieve the highest extinction coefficients. For high-quality measurements and the safety
of the CCD chip, more than 50 dB attenuation is required (the SP filter is theoretically
capable of 70 dB). A 60 mm achromatic lens focuses the transmitted sidebands onto the
entrance slits of the imaging spectrometer to be measured by the CCD.
The CCD allows for very rapid HSG spectrum acquisition rates. A sideband spectrum
spanning 6th to 48th order and and 40 dB of dynamic range can be taken in just five
FEL pulses, about six seconds, see Fig. 3.8. Because the detector is cooled to -90 C,
it can also integrate for as long as five minutes without significant dark count signal,
59
7 0 0 7 1 0 7 2 0 7 3 0 7 4 0 7 5 0 7 6 01 0 - 1
1 0 1
1 0 3
1 0 5
Signa
l (cou
nts/FE
L puls
e)
W a v e l e n g t h ( n m )
1 . 7 0 0 1 . 6 8 0 1 . 6 6 0 1 . 6 4 0
4 8 t h
F r e q u e n c y ( e V )
5 n m G a A s Q W sλN I R = 7 6 4 n mP N I R = 5 0 m Wf T H z = 5 4 0 G H zF T H z = 3 5 k V / c mT = 1 5 K
6 t h
Figure 3.8: Example sideband spectrum measured by the CCD. Sidebands of order 6 ≤n ≤ 48, spanning 40 nm, can be observed in four images covering 20 total FEL pulses,or 800 ns of THz power. The strongest four sidebands spill a little bit out of their pixelsduring readout, but 30 dB down from the peak. Overlap between this spectrum and thePMT spectrum in Fig. 3.6 to calculate the efficiency of these sidebands as well.
60
dramatically decreasing the readout-noise-dominated noise floor. The dynamic range of
the entire experiment can thus be extended to over 100 dB. The effect of the short pass
filter cutoff is apparent in this spectrum in the fast drop off above the 6th order sideband
at about 759 nm. Sideband strengths and linewidths are calculated from the area and
standard deviation of a Gaussian peak fit, respectively. Sidebands measured by both the
PMT and the CCD can be used to calibrate the CCD to measure sideband efficiency over
the full spectrum.
Taking the image of a sideband spectrum requires several steps. The CCD chip is
oriented such that the horizontal axis (1600 pixels) maps to wavelength, and the vertical
axis can be integrated over (400 pixels). Before any image is taken, the camera is set to a
temperature of −90 C, to minimize dark current, and a electron-multiplying gain of 110,
which corresponds to a signal gain of about 7 (these are simply the standard settings).
An un-binned image of a sideband spectrum contains a number of two dimensional peaks.
The peaks are very narrow in the horizontal (wavelength) direction, spanning no more
than ten pixels, but very broad in the vertical direction, spanning as many as one hundred
pixels. The size of the vertical hardware bins are then set so that the entire vertical extent
of a binned pixel can contain the peak of the strongest sideband in the image. Vertical
binning in hardware minimizes the effective readout noise by allowing a slower, more
accurate readout amplifier to be used. If the bins are made too large, however, charge
noise integrated over the many physical pixels can add significantly to the total noise. It
is also possible to bin in software, but the noise characteristics were found to be worse.
61
Two vertical bins are recorded with every image (so that the array output by the CCD is
1600×2). The “top” row contains the sideband or background measurement. The bottom
row is never illuminated, even by background sources, and so it provides a measurement
of the total noise per binned-pixel for that individual image.
With the camera and image settings determined, the next step is to take a background.
The FEL is turned off and the NIR laser is blocked so that background images can be
taken. Because images can be very long (often 40 s, but as long as five minutes), cosmic
rays can interfere with the signal (they look a lot like a sideband when vertically binned).
To remove cosmic ray artifacts, four identical images are taken, although three is the
minimum. These images are compared and any large signal spikes that are only found
in one image are removed and the remaining pixels are processed to return a mean and
standard error for every binned pixel. Taking more images allows for effective cosmic ray
removal (and better error statistics) at the expense of FEL time.
With the background saved, data is ready to be taken. An identical set of images
are taken with the NIR and FEL back on, and the same cosmic ray removal process is
performed. The background array is then subtracted from the background array. The
background-corrected signal data is then divided by the number of times the FEL fired to
to correct for data with different integration times or FEL repetition rates. Finally, the
sideband spectrum data is saved with three columns of data: wavelength, sideband signal,
and error. The error is the quadrature sum of the dark pixel noise and the standard error
of that binned pixel.
62
Sideband polarization measurement
The large optical bandwidth and dynamic range of a single CCD spectrum allow for
measurements of the polarization state of many sidebands in parallel. The complete
polarization state of light is measured by a Stokes polarimeter, a rotating achromatic
quarter-wave plate (QWP) in front of a linear polarizer [23]. Both optics are placed
immediately after the cryostat, as the high-reflectivity dielectric mirrors in the beam path
scramble polarization as a function of wavelength because the dielectric stacks induce
propagation delays that are a very complicated function of wavelength, see Fig. 3.9.
The strength of the sidebands measured in parallel on the CCD as a function of QWP
rotation angle can be fit using the Mueller calculus to obtain all four Stokes parameters
that describe the complete polarization state of the sidebands [23]. To measure the NIR
laser polarization state, the intensity transmitted through the polarizer is measured using
a standard laser power meter instead of the CCD.
Calculating the polarization state from the data requires fitting to the power measured
after the polarimeter, see Fig. 3.9. The four Stokes parameters are defined as follows:
S0 = I
S1 = pI cos 2α cos 2γ
S2 = pI sin 2α cos 2γ
S3 = pI sin 2γ,
63
NIRlaser
Laser + SBs
Sample in cryostat QWP
Polarizer
(a)
(b)
(c)0 90 180 270 360
0.0
0.5
1.0
DataFit
Sig
nal (
norm
.)
QWP angle (°)
NIR laserpol. state
γ
α
Figure 3.9: The Stokes polarimeter schematic and an example measurement. (a) TheQWP and polarizer are placed immediately after the cryostat so that all polarizationmanipulations can be done with as few intervening optics as possible. (b) Two anglesα and γ define the complete polarization ellipse. (c) An example Stokes measurement,the power of the NIR laser transmitted through the polarizer as a function of the QWProtation angle. For this measurement α = 90.8± 1 and γ = 0.4± 1.
where I is the intensity of the light, p (∈ [0, 1]) is the degree of polarization, α (∈ [0, π])
is the angle of the major axis of the polarization ellipse, and γ (∈ [−π/4, π/4]) is the
“roundness” of the polarization ellipse, see Fig. S2(a). For example, (α, γ) of horizontal
light would be (0, 0), vertical (π/2, 0), and right circular light (0, π/4). Let us define
the QWP angle θ as counter-clockwise rotation starting at the positive x-axis and orient
the polarizer so that it transmits horizontally polarized light. By measuring the power
transmitted through the polarizer as a function of QWP angle θ, the four Stokes param-
eters can be determined by fitting the transmitted power data I(θ) with the following
64
equation:
I =1
4(2S0 + S1 + S1 cosα) +
1
4(S1 + S1 cosα) cos(4θ)
− 1
2S3 sinα sin(2θ) +
1
4(S2 − S2 cosα) sin(4θ).
The different phases and frequency in θ allow for one measurement to fit all four Stokes
parameters simultaneously. For a complete derivation and more details, consult Goldstein
[23].
The NIR laser polarization state was measured by placing the QWP at the beam
waist and using a standard optical power meter to measure the intensity as a function
of QWP rotation angle, see Fig. S2(c). The sideband polarization states were measured
in parallel using the CCD spectrometer to monitor the intensity of individual sidebands.
The accuracy of this broadband polarimeter was tested by placing a rotatable polarizer
at the sample and shining collimated light from a tungsten-halogen lamp through the
beam path. The polarization state as a function wavelength was able to predict the α of
the test polarizer to within 1.
Driven absorption measurement
Sideband photons in HSG can only be created if excitons or even unbound electron-hole
pairs are created first. The exciton absorption line, in general, shifts non-monotonically
when illuminated by a strong terahertz field, so ensuring that the NIR laser wavelength
is appropriately tuned is not trivial [46, 6, 60, 74]. The difficulty of measuring driven
65
Fiber from pulsed LED
Pinhole spatial filter
Focusing lens
Spectrometer + CCD
Detectionfiber
Collectinglens
Mylar B
S
Polarizerattenuator
OAP
ITOBC
Pyro monitor
Sample in cryostat
FEL output port
Figure 3.10: The optical schematic of a driven absorption experiment. This measurementcombines the DT measurement with THz illumination. Instead of the tungsten-halogenlamp, however, an electronically-switched LED provides the NIR power only while theFEL is lasing.
absorption lies entirely in getting enough optical power through the sample while also
ensuring spatial and temporal overlap between the NIR and THz light.
A light-emitting diode (LED) can be pulsed very quickly at high currents, see Fig. 3.10.
The pulse-generating circuit is capable of driving about 200 mA in pulses with 10 ns rise
and fall times. The pinhole spatial filter ensures that the LED spot has a top hat profile
with 200 µm diameter, well within the 1 mm THz spot. The spatial filter and FEL duty
cycle severely limit the amount of light at the sample, so integrating enough signal is an
issue. The FEL has to be operated in hole coupler mode, capable of µs-long pulses, to
get enough signal, so the THz power is limited to a few kilowatts, or less than 5 kV/cm
in GaAs. Even that low a field is enough to substantially red-shift the exciton absorption
66
1 . 5 4 0 1 . 5 5 0 1 . 5 6 0 1 . 5 7 0
2
4
6
Abso
rption
(dB)
F r e q u e n c y ( e V )
U n d r i v e n 4 k V / c m
1 0 n m G a A s Q W s f T H z = 5 4 0 G H zT = 1 5 K
Figure 3.11: Example driven absorption measurement. Driving the QWs with a 4 kV/cmTHz field in the plane of the QWs redshifts the undriven QW absorption by about 2 meVand lowers the absorption coefficient. The FEL is lasing using the hole coupler, so theTHz pulse length is approximately 1 µs long. Even the large duty cycle of 10−6 does notprovide high-enough signal-to-noise to perfectly filter out the Fabry-Perot fringes as inFig. 3.3, but the peak is still well resolved.
lines, see Fig. 3.11.
3.2 FEL characterization
Rapid developments in the generation of strong THz light have motivated new under-
standing of the FEL to discover its comparative advantage. On the optical side, THz
time-domain setups have opened coherent THz spectroscopy [27]. New techniques in
optical rectification have increased peak fields in single-cycle picosecond pulses above
MV/cm [70, 51]. On the electronics side, new developments have pushed commercial
sources and detectors above even 1 THz. The advantages of the FEL lie in its high
power, narrow bandwidth, and long pulses.
67
3.2.1 Power measurement
Measuring the instantaneous power of the FEL at arbitrary frequencies is important for
nonlinear effects where the peak power dominates the response. A pyroelectric (pyro)
detector can be a very simple option. These use a ferroelectric material which has a large
temperature-dependent polarization response. By heating them, for instance by painting
them black and illuminating them, they output a voltage approximately proportional to
the intensity of the light on the element. They can be made very fast, with bandwidths
of approximately 100 MHz, but must then be very small. That speed is just fast enough
to measure the instantaneous FEL power, but the size means that they are relatively
sensitive to FEL alignment and spatial profiles that change in time, see Sect. 1.6. Large
area pyro detectors which are more robust to alignment changes are necessarily much
slower. The sensitivity of these detectors is very dependent on ambient temperature
as well, and so make poor long-term absolute power meters. A more robust large-area
detector is the Thomas-Keating (TK) detector. This photo-acoustic detector contains a
thin gas cell with one side coated with a thin film of metal with known THz absorption.
When the metal absorbs the THz power, it creates an acoustic wave in the cell. A
pressure transducer registers the pressure change, which is directly proportional to the
energy in the energy in the THz beam. The large size of the mylar sheet means that the
bandwidth is very small, less than a kHz. Furthermore, this detector is easily calibrated
by driving a DC voltage across the metal film and measuring the response.
Instantaneous power measurements can be made by combining the speed of a pyro
68
detector with the absolute precision of the TK detector, see Fig. 3.12. First, about 5%
of the THz beam is picked off by a mylar beamsplitter and focused on a horn-coupled
fast pyro. The horn makes the detector more insensitive to alignment or spatial profile
changes. The pyro detector is connected directly to an oscilloscope with a 1 MΩ input
impedance so that it operates as a THz power integrator. The height of the discontinuous
jump relative to the total height (not including the ringing) is the proportion of energy
of the FEL pulse in the 40 ns cavity dump refChapter1. The TK detector is placed
in the beam path behind a rotatable wire-grid-polarizer attenuator. With the polarizer
attenuator set to full transmission, the power in the FEL is measured simultaneously
by the pyro and the TK detectors, thereby calibrating the pyro detector for the day’s
measurements. By rotating the front polarizer, the THz power can be attenuated easily
and repeatably while the rear polarizer maintains the polarization at the sample. The
TK can then calibrate the attenuator as well.
After recent improvements inside the accelerator, the FEL is generating record cavity
dump powers. On an average day at 540 GHz, an FEL pulse contains about 2.9 mJ,
and 90% of that is in the cavity dump. For those 40 ns, the THz power is 65 kW. The
proportion of energy in the cavity dump depends on the quality of the electron beam
tuning. Poor tuning leads to slow lasing times, and the proportion of energy in the cavity
dump can be as low as 60%. Also, by calibrating the pyro at the beginning of the day,
any long-timescale drifts can be easily monitored and corrected for using the polarizing
attenuator.
69
0 1 2 3 4 5
10
20
Pyr
o si
gnal
(m
V)
Time (µs)
Cavitydump
Beginlasing
0 20 40 60-200
-100
100
0 45 90 13510-3
10-2
10-1
100
TK
sig
nal (
mV
)
Time (ms)
FEL trig.
∆V ∝Pulse energy
Datacos4 Fit
Tra
nsm
issi
on
Angle (°)
Polarizer attenuator
Integrating pyro
OAP
Mylar B
SRotating wiregrid
Fixed wiregrid
TKOAP
FEL outputport
Figure 3.12: Measuring FEL power. (a) Schematic for measuring FEL power. The THzbeam exits the port and travels along the optical table. A mylar BS picks off about5% of the power and focuses it onto a horn-coupled fast integrating pyro detector. Therest of the beam passes through the polarizer attenuator consisting of a rotating wire-grid polarizer and a fixed wire-grid polarizer. The THz beam is then focused on theTK detector. (b) Time trace of a cavity dumped (CD) FEL pulse measured by theintegrating pyro. Initially, the silicon coupler reflects a few percent of the THz lightout of the cavity, and the integrated signal rises approximately linearly. At 3 µs, anintense green pulsed laser creates an electron-hole plasma in the silicon, dumping allof the THz power out of the FEL cavity in 40 ns. The pyro signal then jumps nearlydiscontinuously. After 3 µs, the oscillations are due to impedance mismatch. (c) Timetrace of the TK detector after absorbing an FEL pulse. The peak-to-peak amplitude ofthe signal is directly proportional to the total power in the FEL pulse. Traces (b) and(c) combine to measure the power only in the cavity dump. (d) Results of calibratingthe polarizer attenuator using the TK. The FEL output horizontally polarized and thefixed polarizer transmits horizontally polarized light, so the transmission through thetwo polarizers is proportional to cos4 θ where θ is the angle between the two wire grids.Because of imperfections in the polarizers, the minimum attenuation is about 1%.
70
3.2.2 Frequency and linewidth
Because of the low duty cycle and low repetition rate of the FEL, measuring the frequency
is not trivial. The FEL electron paramagnetic resonance lab measures the FEL frequency
using superheterodyne detection [52]. This method can measure the frequency and phase
of the FEL very accurately, but the bandwidth is limited to a few GHz around a single
frequency that is not tunable. Another method is to mix the FEL output with a weak NIR
beam in a nonlinear optical crystal [72]. Using a material with a large χ(2) for efficient
sum-frequency generation, the output NIR light and first order sideband frequencies
can be measured, fFEL = fSB − fNIR. This technique has a large bandwidth, capable
of measuring from 0.4–2.5 THz with large signal-to-noise ratios. With only first order
mixing, the precision for measuring the FEL frequency is limited to the resolution of the
spectrometer, ∆FEL = ∆spec ∼ 3 GHz [72]. Increasing the order of the mixing makes
this measurement much more accurate, so that even lower resolution spectrometers can
leverage this effect to be extremely precise.
With dozens of orders of sidebands, HSG is an ideal mechanism to measure the FEL
frequency very accurately. Measuring the location of the nth order sideband increases
the resolution by a factor of n:
fFEL =fSBn − fNIR
n(3.2)
∆FEL =∆spec
n(3.3)
71
1 . 61 . 71 . 81 . 9
- 3 00
3 0
0 2 0 4 0 6 0 8 0 1 0 0 1 2 0 1 4 00
5 0
1 0 0
f F E L =6 3 3 . 7 G H z
6 3 3 . 2 G H z6 3 3 . 2 G H z
SB fre
q. (eV
)
f N I R = 1 . 5 5 3 2 e V
Resid
uals (
GHz) ( f r o m p i e c e w i s e l i n e a r f i t )
Γ SB (G
Hz)
S i d e b a n d o r d e r
Figure 3.13: Three spectrometer steps measuring the frequency characteristics of ageneric sideband spectrum. The FEL was set to fFEL = 640 GHz, and the THz fieldstrength in the QWs was FTHz = 57 kV/cm. (Top) The sideband frequencies increasealmost perfectly linearly, yielding the actual THz frequency, fTHz = 633 GHz. (Middle)Residuals from the linear fits to the sideband frequencies. The residuals increase withincreasing order, likely because the highest order sidebands are quite weak. Even thelargest residuals are less than 30 GHz, which means that the frequencies of the sidebandsdeviate from linearity by less than one part in 104. (Bottom) The linewidths of the side-bands from their Gaussian fits. These also increase approximately linearly with order,but they are several times larger than the residuals.
72
Because there are sidebands at every order, the frequency can be known very well, see
Fig. 3.13(top). With the FEL tuned near 640 GHz, sidebands up to order n = 126 are
observed, requiring three spectrometer steps to measure them all. The image at each
spectrometer step has about 25 individual sidebands (because n must be even), and so
a single spectrometer step can measure the FEL frequency quite well. For context, the
slight change in the FEL frequency between the three steps results from a voltage drift
of 4 kV on the 5 MV accelerator terminal voltage over about twenty minutes (assuming
fFEL ∝ Vterminal). Plotting the residuals shows that the sideband frequencies measured by
the spectrometer, which was home-calibrated using neon and argon lamps, Fig. 3.13(mid-
dle), deviates by less than 30 GHz from linear until the last measured sideband. These
residuals are far smaller than the linewidths of the sidebands measured by the spectrom-
eter, Fig. 3.13(bottom).
These sideband linewidths, ΓSB, provide a measurent of the FEL linewidth. They
increase approximately linearly with order n, although noisily. The FEL is broadened
nonhom*ogeneously by terminal voltage fluctuations, and so the sideband linewidths add
incoherently, ΓSB =√
∆2spec + n2Γ2
THz (The NIR laser has a linewidth of less than 5
MHz, so it is ignored). For small n, ΓSB ∼ ∆spec, and for large n, ΓSB ∼ nΓTHz. The
slope from a linear fit of ΓSB(n) represents ΓTHz and the y-intercept represents ∆spec.
Linewidth results from several FEL frequencies are shown in Fig. 3.14. All but the
247 GHz sideband spectra required multiple spectrometer steps. The linewidth of all
sidebands was calculated and were averaged where sidebands were measured twice (in
73
01 02 0
02 55 0
02 04 0
05 0
1 0 0
0 5 0 1 0 005 0
1 0 0
ΓS B d a t a L i n e a r f i t
f F E L = 2 4 7 G H zΓF E L = 3 3 0 M H z
ΓF E L = 2 7 0 M H zf F E L = 3 4 1 G H z
ΓF E L = 1 0 0 M H zf F E L = 4 3 6 G H z
Sideb
and l
inewid
th, Γ S
B (GHz
)
ΓF E L = 4 4 0 M H zf F E L = 5 3 9 G H z
ΓF E L = 5 6 0 M H zf F E L = 6 3 3 G H z
S i d e b a n d o r d e r
Figure 3.14: The fits of the linewidths for various FEL frequencies. The black squaresare the linewidths from fitting sideband peaks with Gaussian functions, and the reportedfFEL is measured from sideband spacings as discussed in Fig. 3.13. The red line is alinear fit to linewidths as a function of order. The y-intercept for all but the 247 GHzspectrum is consistently 15 GHz. The slope is the linewidth of the FEL, ΓFEL, reportedin each subfigure.
74
2 0 0 3 0 0 4 0 0 5 0 0 6 0 0 7 0 0
1 0 6
1 0 7
1 0 8
Spec
tral ra
dianc
e (W/
cm2 /G
Hz)
F E L F r e q u e n c y ( G H z )0
5 0
1 0 0
1 5 0 Peak electric field (kV/cm)
Figure 3.15: Spectral radiance and peak electric field in vacuum. The spectral radianceincreases with frequency until it saturates in the mid-107 range, solid line, left axis. Thepeak electric field (in air) increases linearly, dashed line, right axis. Diffraction at lowfrequencies (long wavelengths) causes substantial losses in the FEL. At higher frequencies,however, the FEL becomes less stable and the linewidth gets broader. The peak electricfields keep increasing but the spectral radiance plateaus.
overlapping regions).
These results allow the spectral radiance of the FEL to be measured accurately.
The spectral radiance (sometimes called spectral “brightness”) is the power flux of an
optical beam per solid angle per area per bandwidth. It is invariant along the beam
path (assuming propagation through a uniform medium) and is best understood as the
highest intensity possible at the best possible focus of that beam. For a moderately-
focused Gaussian beam (so that the paraxial approximation is valid),
R =P
πw20 πθ
2=P
λ2, (3.4)
where R is the radiance, P is the power in the beam, w0 is the beam radius at the
focus, θ is the divergence angle (θ = λ/(πw0)), and λ is the wavelength of the light. The
75
spectral radiance is therefore the radiance per bandwidth. The FEL, being a high-power,
quasi-continuous wave source, has very large spectral radiances. Large spectral radiances
are useful for sideband experiments were the highest orders might be washed out without
it.
3.3 Outlook
The sideband spectrometer can be expanded and extended to encompass other impor-
tant measurements to study nonlinear THz-NIR interactions. A complete set of optical
edge filters are required to fully make use of the Ti:sapphire laser tuning range for side-
band measurements. There are many materials that support excitons in this range, but
currently the filters only allow for excitons between about 750–850 nm. The pulsed
LED also does not span the full 700–1000 nm range, and not all materials will have
such well-understood excitonic responses to strong THz fields. A super-continuum laser
source for driven absorption measurements would be a big upgrade. Because the NIR
laser is highly tunable, photoluminescence excitation can be added to the repertoire to
characterize samples, and it could also be performed under FEL excitation. Also, for
more efficient polarization state measurements, including finely controlling the NIR laser
polarization, a commercial polarimeter is necessary.
76
Chapter 4
Terahertz-induced birefringence and
sideband polarimetry
A recent observation of high-order harmonic generation (HHG) in ZnO has set off a
flurry of new research on high-order nonlinear processes in solids [22, 51, 55, 56]. Us-
ing extremely strong pulsed mid-infrared (MIR) or terahertz (THz) laser pulses, these
studies create interband polarization by causing electrons to tunnel across the band gap
in semiconductors like ZnO and GaSe. The nonlinear response comes from two differ-
ent mechanisms, one interband and one intraband. Harmonics created by the interband
process result from electron-hole recollisions, much like like HHG in atoms. Harmonics
created by the intraband process result from the non-parabolic dispersion relation of the
bands. In the works by Ghimire, et al., Schubert, et al., and Vampa, et al., the electrons
(holes) accelerate up to nearly the top (bottom) of the conduction (valence) band, where
77
the bands are extremely non-parabolic. These papers suggest that the driving frequency
determines what mechanism dominates, MIR for interband and THz for intraband. These
mechanisms are important to differentiate, as the recollision physics could be a useful
tool for measuring band structure from electron and hole trajectories [41, 56].
High-order sideband generation (HSG) is only sensitive to recollision physics. In
HSG, the generation of interband polarization results from weak, resonant NIR excitation
instead of electric field-induced tunneling. Resonant excitation restricts the creation of
electron-hole pairs to a small region of momentum space with well-defined spin properties,
whereas tunneling generates electrons and holes throughout the Brillouin zone with little
regard to spin [73, 38]. The quasiparticles created by the NIR laser then evolve, driven
by the strong THz electric field. Because the state of these quasiparticles can controlled
independently by the NIR excitation, it is possible to separate the recollision physics from
the non-parabolic band physics. Langer, et al., showed that the behavior of the high-
order nonlinearities could be controlled by switching between HHG and HSG experiments
in the same sample. Rotating the bulk WSe2 crystal under only strong THz illumination
(keeping the THz polarization direction constant) leads to HHG spectra that reflect in
a six-fold rotational symmetry of the crystal. Under NIR and THz illumination, the
dependence on lattice angle was largely smeared out. In the HHG case, the number of
excited carriers depends on how steep the bands are around the band edge. Steeper bands
(lower masses) in certain k-space directions leads to more difficult tunneling and so lower
interband polarization. In HSG, however, the number of electron-hole pairs is roughly
78
constant as a function of lattice angle, since the bottom of the bands is approximately
spherically symmetric [38].
This chapter will show how changing the polarization of the NIR laser and the angle
of the [100] direction of the lattice with respect to the polarization of the THz laser
affects HSG. Linear NIR polarization along the THz polarization creates a superposition
of spin-up and spin-down electrons (and holes) that are in phase, but when the NIR
polarization is perpendicular to the THz polarization, the superposed spins are 180 out
of phase. This different initialization will lead to substantial changes in sideband intensity
and polarization state.
4.1 Sample details
The large increase in sideband order observed by moving to unstrained, clean GaAs/AlGaAs
quantum wells (QWs) and the much stronger THz field produced by the FEL inspired
the request for growth of a new sample and the development of a new, more robust
processing technique.
4.1.1 Sample growth
Three quantum well (QW) samples with varying degrees of quenched disorder were grown
by molecular beam epitaxy. The 5 nm GaAs sample consists of twenty 5 nm GaAs QWs
with 20 nm Al0.3Ga0.7As barriers. The 10 nm AlGaAs sample consists of twenty 10 nm
Al0.05Ga0.95As QWs with 20 nm Al0.3Ga0.7As barriers. Those two samples are the same
79
5 nm GaAs QWs 10 nm AlGaAs QWs 10 nm GaAs QWs50 nm GaAs 50 nm GaAs 150 nm GaAs
40 nm Al0.3Ga0.7As 40 nm Al0.3Ga0.7As 50 nm Al0.3Ga0.7As20 nm Al0.3Ga0.7As 20 nm Al0.3Ga0.7As 15 nm Al0.3Ga0.7As
NQW 5 nm GaAs 10 nm Al0.05Ga0.95As 10 nm GaAs
20 nm Al0.3Ga0.7As 20 nm Al0.3Ga0.7As 15 nm Al0.3Ga0.7As40 nm Al0.3Ga0.7As 40 nm Al0.3Ga0.7As 50 nm Al0.3Ga0.7As
50 nm GaAs 50 nm GaAs 150 nm GaAsES 300 nm Al0.7Ga0.3As 300 nm Al0.7Ga0.3As 300 nm Al0.735Ga0.265As
50 nm GaAs 50 nm GaAs 100 nm GaAsSubst. 500 µm GaAs 500 µm GaAs 500 µm GaAs
Table 4.1: The growth structure of all three samples. The left two samples were used inChap. 2, and the right sample is new. The QW region, etch stop (ES), and substrate(Subst.) layers are labeled. The layers labeled N are repeated, N = 20 for the 5 nmGaAs and the 10 nm AlGaAs samples, and N = 10 for the 10 nm GaAs sample. The onlydifference is the material in the quantum well region. Both samples are grown analogexcept for the Al0.05Ga0.95As layers in the 10 nm QW sample, which were grown digitally.The 5% aluminum alloying raises the band gap of the quantum well to cancel out thesmaller confinement energy of the wide quantum well.
used for the experiments in Chap. 2. The 10 nm GaAs sample is new and consists of ten
10 nm GaAs QWs with 15 nm Al0.3Ga0.7As barriers, grown with interrupts at AlGaAs-
on-GaAs interfaces for very smooth walls [21, 31, 35, 25]. All samples were grown on
(100) semi-insulating GaAs substrates.
4.1.2 Sample processing
The old membrane sample geometry, discussed in Ch. 2, could not withstand the intense
fields of the rebuilt FEL, so an epitaxial transfer process was developed so that the epitax-
ial layer could be supported by a NIR- and THz-transparent sapphire substrate1 [65, 64].
This process involves bringing two extremely smooth surfaces (A-scale roughness) into
1With significant instruction from Garrett Cole from Crystalline Mirror Solutions
80
contact, here a very clean c-plane, (0001), sapphire wafer and an initially-internal sur-
face of an epitaxial layer exposed by etching away the substrate and sacrificial layers,
see App. C. Because both surfaces are so clean and so smooth, Van der Waals forces
spread across millimeters are strong enough to bond the surfaces. No extra materials are
needed to create the bond, so there are no confounding problems like strong terahertz
absorption or stress-induced birefringence. Compared to the membrane geometry, the
sapphire substrate is much more robust and easier to handle. GaAs and sapphire also
have similar thermal expansion coefficients, so thermal cycling between 15 K and 294 K
does not harm the sample.
With a total sample thickness on the order of the THz wavelength, interference be-
tween the incident field and reflections off the back surface affect the THz field strength
at the front surface, in the QWs. This interference leads to a periodic dependence of the
THz field strength in the QWs on THz frequency which is usually undesirable. This inter-
ference could be reduced by depositing a THz anti-reflection coating to the back surface
of the sapphire [44]. Instead, we deposited a 300 nm, 12.6 Ω/ thin film of indium-tin
oxide (ITO). The large THz reflection off the film exaggerates the interference at the
front surface, enhancing the electric field at certain frequencies inside the QWs by more
than 50%, see Fig. 4.1, though other frequencies suffer almost complete destructive in-
terference. The THz source, the UCSB Free Electron Laser (FEL), is broadly tunable in
frequency with a very narrow linewidth, and HSG does not depend on resonant effects,
so the THz frequency can be chosen solely based on the results of processing. The ITO
81
film is very transparent at NIR wavelengths, so it does not attenuate the laser or the
sidebands.
The field enhancement as a function of frequency, κ(fTHz), was measured using a
frequency-extended vector network analyzer (VNA) as a quasi-optical vector reflectome-
ter [59, 3]. Schottky diode multipliers and dividers in the transmitter and receiver extend
the 0–40 GHz frequency domain of the VNA up to 70–700 GHz, and precision horns cou-
ple the light to and from free space. The reflectometer is set up so that a transmitter
illuminates a planar sample. The reflected light is then split off by a beam splitter and
directed towards a receiver. Because the reflectometer operates in the frequency domain,
Fabry-Perot interference develops that can confound the signal. By running a calibration
routine through the VNA on a metal mirror, the phase changes from the beam path can
be subtracted out. The reflection spectrum of the sapphire-ITO cavity then represents
the electric field at the sapphire surface, where the QWs will be [47]. This measurement
has been repeated both with and without QWs on the front surface, and there is little
difference. The THz field in the QWs is then Fin QWs = κ√εGaAs
Fappliedwhere εGaAs is the
dielectric constant of GaAs.
4.1.3 Sample optical properties
The thickness and composition of the QWs determine the locations of the exciton ab-
sorption peaks. Absorbance spectra were taken using the experimental setup described
in Sect. 3.1.1. The sample surfaces are so smooth that the sapphire substrate still func-
82
Epi layer
ITO filmSapphire
VNA path
(a) (b)
Figure 4.1: The THz field enhancement measurement geometry and results. (a) Thefocused quasi-optical light from the VNA illuminates the sample. The light transmitsthrough the epitaxially-transferred QWs (epi layer), through the sapphire, and reflectsoff of the ITO surface back at the QWs. Interference between the incident and ITO-reflected THz field creates the field enhancement κ(fTHz). (b) The field enhancementas a function of frequency. The enhancement is peaked at about 1.5 with a period ofapproximately 100 GHz in the measurement range, but there are also frequencies withalmost total destructive interference in the QWs.
83
1 . 5 4 5 1 . 5 6 0 1 . 5 7 5 1 . 6 2 0 1 . 6 3 5 1 . 6 5 002468
5 n m G a A s
1 0 n m A l G a A sAbso
rption
(dB)
F r e q u e n c y ( e V )
1 0 n m G a A s
Figure 4.2: Optical absorption spectra of the three samples measured by differentialtransmission. In all three samples, the heavy hole exciton line (largest peak) and thelight hole exciton line (slightly weaker, blue-shifted by 10 or 25 meV depending on wellwidth) are both clearly resolved. Absorption and all other experiments are performed at15 K.
tions as an etalon at optical wavelengths. As previously mentioned in Sect. 3.1.1, recall
that the short-period sinusoid with an amplitude approximately 30% of the transmission
signal can be removed in software using a short pass Fourier filter. The samples were
held at 15 K for all experiments. Absorbance spectra for all three samples are found in
Fig. 4.2. In each spectrum, two peaks are clearly visible, the heavy hole exciton (HHX)
and the light hole exciton (LHX) absorption lines. The 5 nm and 10 nm samples can be
differentiated by the size of the HHX-LHX splitting. Quantum confinement breaks the
HH-LH degeneracy at the top of the valence band such that narrower wells lead to larger
splittings. The two 10 nm samples can be differentiated by the aluminum content in the
well region. The 5% Al content increases the 2D band gap so that the HHX absorption
line is blue-shifted up to the 5 nm GaAs sample HHX absorption line. The 5 nm GaAs
and 10 nm AlGaAs samples are from the same wafers used in [4].
The linewidths of the exciton absorption lines reflect the quenched disorder of the
84
samples and so depend on the growth materials and conditions. In the two GaAs samples,
the linewidths of the HHX peaks are largely determined by well-width fluctuations [21,
31, 35, 25]. Because it is half as wide, the 5 nm GaAs QWs should exhibit four times
the nonhom*ogeneous broadening of the 10 nm GaAs QWs for the same size well-width
fluctuations. The 10 nm GaAs sample was grown, however, to have particularly smooth
well-barrier interfaces, so it should be even more narrow than the 5 nm GaAs sample from
well-width fluctuations alone. Indeed, the 5 nm GaAs sample has an HHX full width at
half max (FWHM) of 6.3 meV where the 10 nm GaAs sample has an HHX FWHM of
2.0 meV. From well-width fluctuations alone, the 10 nm AlGaAs sample should have a
narrower HHX peak than the 5 nm GaAs sample, but it also has a FWHM of 6.3 meV.
The additional nonhom*ogeneous broadening in the 10 nm AlGaAs sample is due to alloy
disorder in the well region. The effect of alloy disorder is modeled by Poisson-distributed
aluminum content fluctuations over the 10-nm-diameter exciton wavefunction leading to
local fluctuations in the band edge. At large length scales, these fluctuations lead to slight
blurring of the band separation. We assume that the electron and hole are spheres 10 nm
in diameter and that the aluminum atoms in the well region are randomly distributed.
85
The lattice constant of GaAs is 0.57 nm.
Ve =4π
3r3e ∼ 520 nm3
〈NAl〉 ∼ 140 atoms
σAl =√〈NAl〉 ∼ 12 atoms
nAl = 5.0± 0.4%
EΓ = 1.5777± 0.0048 eV,
where Ve is the volume of the quasiparticle wavefunction, 〈NAl〉 is the average number
of aluminum atoms in that volume, σAl is the standard deviation of the aluminum atom
number and nAl is the number density of the aluminum atoms. From this derivation,
we should expect alloy disorder to contribute about 5 meV of nonhom*ogeneous excitonic
linewidth broadening from 5% aluminum concentration.
4.2 Sideband measurements
4.2.1 Experimental preparation
Sidebands are generated by focusing the NIR and THz beams at the same spot on the
sample with collinear propagation normal to the sample, see Sect. 3.1.2. Recall that the
radii of the NIR and THz focal spots are approximately 100 µm and 0.5 mm, respectively.
The NIR polarization is switched between horizontal and vertical by a half-wave plate
86
(HWP), and the THz polarization is kept horizontal. First, the ideal NIR and THz
frequencies and powers were chosen to maximize signal and experimental convenience.
Then, the THz polarization state was measured to ensure it was highly linearly polarized.
NIR frequency dependence
With the new sample design with dramatically increased THz field in the quantum wells,
it was important to remeasure the HSG dependence on NIR frequency. In the 5 nm GaAs
sample, Fig. 4.3(left), the excitonic absorbance spectrum spans 40 meV (20 nm), from
the low energy wing of the HHX absorption line to the onset of the E1-LH1 continuum.
The sideband spectrum was measured in a 50 meV (25 nm) band encompassing these
absorbance features, using a THz frequency, fTHz = 540 GHz, total field in the QWs,
FTHz > 15 kV/cm, and horizontally polarized NIR and THz lasers2. Because it was an
early measurement in the development of the sideband spectrometer, only three orders
were measured, n = 10, 12 and 14. They are reported at the NIR excitation frequency
and wavelength, not those of the sideband. The sideband strengths of all three samples
are peaked at the HHX line, similar to the results in the membrane samples, see Chap. 2.
At lower frequencies, as fNIR shifts into the gap, the sideband signal dies quickly. The
sideband signal remains strong up to the high-frequency edge of the measurement, 25
2The rigorous THz power measurement routine from Sect. 3.2.1 had not been developed yet. Thisand other early experiments measured the energy in the FEL cavity dump by attempting to subtractaway the power output before the cavity dump by triggering the 532 nm green laser pulse exactly atthe end of the electron beam pulse or long after the electron beam pulse. The difference between anFEL pulse with and without a green pulse would be the energy in the cavity dump, if the ringdown ofthe FEL cavity was not a substantial amount of energy and it was easy to align the end of the electronbeam with the green pulse trigger. While this approach seems elegant, neither of these of assumptionsare valid.
87
meV higher in energy than the nearest exciton absorption peak, suggesting that even
free electron-hole pairs can generate HSG, although excitons seem to be more efficient.
Unfortunately, the sideband signal also seems to jump up and down by approximately
30-50% as fNIR is tuned through the experiment. This effect results from the sapphire
substrate forming an etalon for the NIR laser and the sideband, which has been observed
in absorbance spectra, discussed in Sect. 3.1.1. For continuous sources like white light,
the effect can be filtered out using software, but to do the same by stepping the NIR laser
more finely than the etalon fringes would require unreasonable stability of both lasers,
not to mention extreme patience. This effect only limits the precision of measurement of
changes smaller than 30% in sideband signal, not the measurement of a whole sideband
spectrum which can span ten orders of magnitude. Because this sample and the 10 nm
AlGaAs sample were both discussed in Chap. 2 and are very similar, similar data for the
10 nm AlGaAs sample was not taken.
A NIR wavelength scan of the 10 nm GaAs sample reveals similar qualitative physics,
see Fig. 4.3(right). The fNIR scan is much narrower here, 20 meV (10 nm), because the
excitonic features are also much narrower and the HHX-LHX splitting is much smaller,
but the THz frequency and field are the same. For this measurement, however, the NIR
laser is vertically polarized, perpendicular to the THz polarization, but the polarization
orientation has little effect on this measurement. The sideband signal is again peaked
near the HHX absorption line, but the peak is much broader than the line itself. The
exciton line is likely observably broadened by the strong THz field because the zero-field
88
1 . 6 2 0 1 . 6 4 0 1 . 6 6 00 . 0
0 . 5
1 . 0
p o l N I R = H
Norm
alized
signa
l
N I R e x c i t a t i o n f r e q u e n c y ( e V )
A b s . 1 0 1 2 1 4
5 n m G a A s Q W s
1 . 5 5 0 1 . 5 6 0 1 . 5 7 0
A b s . 2 0 2 4 3 0 3 2
p o l N I R = V
1 0 n m G a A s Q W s7 7 0 7 6 0 7 5 0
N I R e x c i t a t i o n w a v e l e n g t h ( n m )8 0 0 7 9 5 7 9 0
Figure 4.3: Measuring the NIR frequency dependence of HSG. (left) The signal of then = 10, 12, 14 sidebands plotted as a function of fNIR and normalized to the frequencywith the highest signal in the 5 nm GaAs QWs. The zero-field absorption spectrumis plotted as a reference. The NIR laser is horizontally polarized, parallel to the THzpolarization. The sideband signal peaks sharply at the zero-field HHX absorption line.Sideband signal remains strong over a very broad frequency band, however, suggestingthat HSG does not require excitons. (right) The signal of several sidebands between20 ≥ n ≥ 32 plotted as a function fNIR and normalized to the frequency with the highestsignal in the 10 nm GaAs QWs. The zero-field absorption spectrum is plotted as areference. The NIR laser is vertically polarized, perpendicular to the THz polarization.The sideband signal is not peaked as strongly around the HHX absorption line, andthere is a clear secondary peak near the LHX absorption line. For both plots, FTHz > 15kV/cm and fTHz = 540 GHz.
89
line is so much narrower in this sample. There is also a second peak in the sideband
signal around the LHX absorption line, which fits into the hypothesis that HSG signal is
more dependent on the optical density of states than on electron-hole pair character in
such strong THz fields.
NIR power dependence
It is also important to determine the best NIR power, PNIR, to use for measurements. For
all of these experiments, the NIR spot is not very focused; the diameter is about 220 µm.
In the 5 nm GaAs sample, Fig. 4.4, FTHz > 45 kV/cm (still using the difference technique
for measuring FEL power). Sideband signal for several different orders up to twentieth
is linear in PNIR up to 150 mW. In the 10 nm GaAs sample, Fig. 4.4(center), the NIR
power dependence is very complicated. Here, FTHz > 30 kV/cm. The sideband signal
actually peaks at about 125 mW. There are two possible hypotheses to explain this kind
of power dependence: exciton-exciton interactions or sample heating. Assuming that
the absorption per QW is about 10% and that the excitons live for approximately 1 ns
[18], which is most certainly a large overestimate for excitons in THz fields even above 10
kV/cm, the peak density of excitons is 1011 cm−2. The average distance between excitons
at that density is 26 nm, about two and a half times the diameter of the excitons (given
by the well width). At these (very underestimated) separations, there should not be any
exciton-exciton interactions. It is more likely that the interaction of the very strong FEL
pulse, featuring average powers of several kW, and the NIR laser, which creates carriers
that can absorb the THz radiation efficiently, heats up the sample and decreases sideband
90
0 1 0 00 . 0
0 . 5
1 . 0
0 1 0 0 2 0 0 0 1 0 0 2 0 0
p o l N I R = H
Sideb
and s
treng
th (no
rm.)
N I R p o w e r ( m W )
8 1 4 2 0
5 n m G a A s Q W sλN I R = 7 6 5 . 2 n m
p o l N I R = H
1 0 n m G a A s Q W sλN I R = 7 9 8 . 5 n m
1 8 2 4 3 0 3 6
2 4 . 0 3 2 . 0 4 0 . 0 4 8 . 0 5 6 . 0
1 0 n m A l G a A s Q W sλN I R = 7 6 5 . 2 n m
p o l N I R = V
Figure 4.4: The NIR power dependence of HSG in the three samples. (left) The sidebandsignal in the 5 nm GaAs sample is almost exactly proportional to the NIR power forsideband orders between 8 and 14. (center) The sideband signal in the 10 nm GaAssample, however, has a strong peak at 125 mW for orders between 18 and 36, but itdeparts from linear at around 50 mW. (right) The sideband signal in the 10 nm AlGaAssample actually plateaus at 50 mW.
intensity. A similar phenomenon is seen in the 10 nm AlGaAs sample, Fig. 4.4(right),
here with perpendicular NIR and THz polarizations and FTHz > 45 kV/cm. Above 50
mW of NIR power, the sideband signal flattens out as the PNIR is increased.
The NIR power and frequency are chosen for the polarization experiments from the
results of the preceding subsections. For all following experiments in this chapter, the
NIR laser will be tuned to the HHX absorption line, 1.6227 eV (764.0 nm) for the 5 nm
GaAs and the 10 nm AlGaAs samples, and 1.5530 eV (798.4 nm) for the 10 nm GaAs
sample. The NIR power will be 50 mW for all experiments so that the sideband signal is
strong but still approximately linear in PNIR. For ease of FEL electron beam tuning, the
FEL frequency will be set to 540 GHz. Also, to ensure that all experiments are directly
comparable, the THz field will be attenuated down to a total field of 35 kV/cm inside
91
the quantum wells, accounting for losses from the ITO beam combiner and the cryostat
window. The highest-order sidebands are measurable by averaging over one thousand
FEL pulses.
Terahertz polarization
Before final measurements of the THz-induced birefringence and sideband polarimetry,
we measured linearity of the polarization of the THz beam to ensure that there was
no significant THz electric field component in the vertical direction which could feasibly
confound the HSG results and interpretation. To do this, we inserted a rotatable wire grid
polarizer at the focus of the beam and a horn-coupled pyroelectric detector immediately
behind the polarizer, see Fig. 4.5(a). The angle is defined such that φ = 0 when the wires
are vertically oriented (they transmit horizontally polarized light). It should be noted
that this setup is only a linear polarimeter, as a wire grid polarizer cannot distinguish
between circularly polarized light and unpolarized light. The transmitted power as a
function of wire grid angle, Fig. 4.5(a), shows that the THz field is linearly polarized
to a high degree. The contrast ratio between the minima and maxima of the sinusoidal
fit indicate that the THz field is > 99% polarized. The locations of the extrema also
show that the polarization is horizontal. Deviations from perfect horizontal linear THz
polarization are likely due to wire grid quality or alignment in either the linear polarimeter
or the wiregrid polarizers in the THz attenuator discussed in Sect. 3.2.1
92
(b)
0 90 180 270 3600.0
0.5
1.0DataFit
TH
z po
wer
(no
rm.)
Polarizer angle ϕ (°)
Rotating polarizer
Pyroelectricdetector
FocusedTHz
(a)
ϕ
Figure 4.5: Measuring THz polarization. (a) A wire grid polarizer is mounted to a rotat-able stage and placed at the focus of the THz beam. A fast, horn-coupled pyroelectricdetector is placed immediately behind the wire grid. (b) The THz power measured be-hind the polarizer as it was rotated. The signal was normalized the the signal when thepolarizer was removed. The peaks at 0 and 180 and minima at 80 and 270 reveal thatTHz beam is > 99% horizontally polarized.
4.2.2 Results
NIR polarization dependence
The strong terahertz fields lead to the observation of sidebands of more than sixtieth
order, but the shape of the sideband spectra are very different between samples, see
Fig. 4.6. Because these samples are grown on (100) GaAs, a plane with inversion sym-
metry, no odd-order sidebands are observed. For the rest of the chapter, the sideband
spectra will be referred to by the relative polarization of the NIR and THz light. All
three samples show many more sidebands of positive order than negative orders, and
each spectrum spans more than 150 meV. In the classical three step model, the electron
and hole tunnel ionize from the exciton and begin to accelerate in their respective bands
93
with zero momentum and kinetic energy. As they accelerate in the THz field, therefore,
they can only gain energy, so that photons from recollisions must contribute to positive
sidebands. In the 5 nm GaAs sample, Fig. 4.6(a), the negative order sidebands fall off
quickly and purely exponentially. The positive order sidebands also fall off exponentially,
quickly at first but then much more slowly, such that the n = 66 order sideband is the
highest order clearly observed. In the 10 nm AlGaAs sample, Fig. 4.6(b), the sideband
spectrum is almost identical to spectrum in the 5 nm GaAs sample. The negative orders
again fall off quickly and purely exponentially, and the positive orders also fall off quickly
at first but then much more slowly. Despite the sideband signal being a little bit weaker
than the spectrum in 5 nm GaAs sample overall, the n = 66 order sideband is clearly
observed. In the 10 nm GaAs sample, Fig. 4.6(c), the sideband spectrum has a much
different shape. The negative order sidebands fall off exponentially at first, but between
−8 ≤ n ≤ −16, the sideband signal flattens out before falling away again. This nega-
tive order sideband “shoulder” is likely related to interference between QW HSG signal
and weak HSG signal generated by the relatively thick GaAs end caps, see Table 4.1 for
growth details. The positive order sidebands also shows three distinct regions. Up to
about the n = 12 order sideband, the sidebands fall off quickly and exponentially. At
this order, they flatten out significantly until about the n = 30 when they fall off at a
more moderate exponential rate. This very slow fall off looks suspiciously like the classic
three step model plateau, but it is not nearly large enough. The ponderomotive energy
for the THz field strength and frequency is 1.2 eV, meaning the plateau should extend
94
1 0 - 21 0 11 0 4 5 n m G a A s
p a r a l l e l
1 0 - 21 0 11 0 4
p a r a l l e l
( a )
Signa
l (arb.
u.)
1 0 n m A l G a A s( b )
7 0 0 7 2 5 7 5 0 7 7 5 8 0 0 8 2 51 0 - 21 0 11 0 4
p a r a l l e l
W a v e l e n g t h ( n m )
1 0 n m G a A s ( c )
1 . 7 0 0 1 . 6 0 0 1 . 5 0 0P h o t o n e n e r g y ( e V )
Figure 4.6: The sideband spectra created by parallel NIR and THz polarizations. TheNIR laser is represented by a dotted vertical line. The colors represent different detec-tors. Black discontinuous peaks are very close to the NIR laser and must be measuredwith a PMT in a high-rejection monochromator. Blue continuous spectra are measuredwith a CCD in an imaging spectrometer, darker having longer exposure times. Sidebandsmeasured by the CCD are scaled to match those measured by the PMT by ensuring thedetectors measure several sidebands of overlap. To match CCD images of the same side-band spectrum with different integration times, the signal is normalized to the number ofFEL “pulses.” The readout noise, which is the dominant noise source, is then diminishedby a factor of the number of pulses. The noise floor of the PMT is higher, but no extranegative sidebands are observed using the CCD to detect those sidebands.
for twelve hundred orders with a cutoff wavelength of 230 nm.
Changing the relative polarization of the NIR and THz lasers has a large effect on
the sideband spectra, leading to the observation of up to 86th order sidebands with total
spectrum bandwidth of more than 200 meV, see Fig. 4.7. The sideband conversion
efficiency is proportional to the area of a Gaussian function fit to a sideband peak.
More details are shown in Fig. 2.6 and discussed in Sect. 3.1.2. In all three samples,
the negative order sidebands and the positive low-order sidebands do not change with
95
relative polarization. These sidebands result from both perturbative sideband generation
and very low energy electron-hole recollisions which do not extend far into the bands [57].
Above about tenth order, however, sidebands from the perpendicular geometry fall off
much more slowly than sidebands from the parallel geometry for all samples. That
different NIR polarizations exhibit very different HSG response means the strong THz
field induces birefringence in the sample. In the 5 nm GaAs sample, Fig. 4.7(a), the two
polarization geometries separate in HSG intensity between twentieth and thirtieth order.
In the 10 nm AlGaAs sample, Fig. 4.7(b), the two spectra separate at much lower order
relative to the 5 nm GaAs sample, between tenth and twentieth order. In the 10 GaAs
sample, Fig. 4.7(c), the perpendicular HSG spectrum looks like the parallel spectrum has
been stretched horizontally. The two spectra separate at a similar order to the other 10
nm sample, suggesting that this phenomenon is connected to the well width.
Lattice angle dependence
The observation that the well width determines at what order the HSG spectra from the
two excitation geometries diverge suggests that subband separation could play a role in
this polarization effect. Although the subband energy differences at k = 0 cannot be
changed without growing new samples, the energies at finite k depend strongly on lattice
direction. According to the Luttinger model for bulk GaAs, the 〈100〉 family of directions
has a different effective mass than the 〈110〉 family. This non-spherical bulk effective mass
reduces to non-cylindrical effective masses in a quantum well with complicated effects on
hole subband avoided crossings. By rotating the sample relative to the THz driving field,
96
1 0 - 1 01 0 - 81 0 - 61 0 - 4
1 0 - 1 01 0 - 81 0 - 61 0 - 4
- 5 0 0 5 0 1 0 0 1 5 0 2 0 01 0 - 1 01 0 - 81 0 - 61 0 - 4
5 n m G a A s Q W s( a )p e r p e n d i c u l a rp a r a l l e l
N I R p o l a r i z a t i o n :Sid
eban
d con
versi
on ef
ficien
cy
1 0 n m A l G a A s Q W s
T H z p o l a r i z a t i o n :
p e r p e n d i c u l a rp a r a l l e l
( b )
S i d e b a n d o f f s e t e n e r g y ( m e V )
1 0 n m G a A s Q W sp e r p e n d i c u l a rp a r a l l e l
( c )
- 2 0 0 2 0 4 0 6 0 8 0S i d e b a n d o r d e r
Figure 4.7: Sideband conversion efficiencies for two THz-NIR relative polarizations. Po-larization definitions are drawn on the top. The perpendicular-excited sidebands aremuch stronger than the parallel-excited sidebands above a certain order threshold for allthree samples, nt ≈ 30 for the 5 nm GaAs sample and nt ≈ 10-20 for the two 10 nmsamples. This threshold is directly related to the subband spacing of the QWs. Abovethis threshold, the perpendicularly-excited sidebands fall off much slower. In both the5 nm GaAs and 10 nm AlGaAs samples, the sidebands above nt fall off almost exactlyexponentially with order. High order sidebands from the 10 nm GaAs sample, however,show a much more complicated relationship with increasing order.
97
it is possible to isolate how HSG spectra and the THz-induced birefringence depend on
the subbands at finite k.
Rotating the sample relative to the terahertz polarization reveals the effects of the
lattice on HSG, a strong sensitivity to the band structure and a robustness to quenched
disorder, Fig. 4.8. In the upper left of Fig. 4.8, the rotation of the lattice is defined as
the angle between the [100] axis of the GaAs lattice and the THz polarization by the
angle θ looking at the sample along the propagation direction of the excitation light.
In the 5 nm GaAs sample, the upper right of Fig. 4.8, two angles are measured and
their dispersion relations are calculated, 85 and 55. At 85, as has been shown in
Fig. 4.7, sideband spectra from the two excitation geometries diverge at a relatively high
sideband order, about n = 30. At 55, the two excitation geometries actually do not
diverge significantly at all, up to the highest order observed sidebands. The small THz-
induced birefringence in this sample can be traced to the large avoided crossing between
the HH1 and HH2 subbands, almost 10 meV wide at relatively high momentum, k ∼ 0.1
1/a. In the 10 nm GaAs sample, bottom right of Fig. 4.8, two angles are measured
and their dispersion relations are calculated, 91 and 63. At 91, as has been shown
in Fig. 4.7, sideband spectra from both excitation geometries diverge at a relatively low
sideband order, n < 20. At 63, sideband spectra from both excitation geometries change
significantly, the sideband conversion efficiencies become stronger and the shapes change.
Compared to the 5 nm QW dispersion relations, the HH1-HH2 avoided crossing is much
smaller, only 5 meV, and at lower momentum, k ∼ 0.08 1/a. In the 10 nm AlGaAs
98
[100]GaAs lattice
THzpolarization
Figure 4.8: The ratio I⊥/I‖ in all three samples for different lattice orientations, andcalculated valence subband dispersion relations. The THz polarization was kept hori-zontal and the sample was rotated clockwise, shown in the upper left. The empty blacksquare scatter points are replotted from Fig. 4.7. The valence subband dispersion rela-tions are plotted for two lattice directions to demonstrate the relationship between holedispersion and sideband spectrum. For comparison, the hole kinetic energy contributesapproximately 30-40% of the sideband offset energy (SB off. en.). 5 nm GaAs: Thelarge order threshold nt for parallel and perpendicular sideband spectra to diverge is dueto the HH1-HH2 avoided crossing being at much larger momentum with a much largersplitting than in the 10 nm QWs. At 55 orientation, so the THz is nearly along the[110] direction, the splitting is even larger, and so there very little difference between theparallel and perpendicular HSG spectra. 10 nm AlGaAs: The SB efficiencies show lit-tle dependence on lattice orientation. Alloy disorder, which blurs the dispersion relationat moderate length scales, smooths out the cubic symmetry of GaAs and is representedin the dispersion relation as a broadening of the curves. The ensemble of recollisionstherefore average out to have the same response at different angles despite the local lackof cylindrical symmetry. 10 nm GaAs: The sideband spectra depends substantially onangle because the avoided crossings in the dispersion relation are at relatively low energy,and there is little macroscopic disorder to average out hole state interference.
99
sample, bottom left of Fig. 4.8, three angles are measured and two of their dispersion
relations are calculated, 93, 77, and 47. Unlike the other two samples, the sideband
spectra from this sample are almost independent of orientation. Related to how alloy
disorder broadens the exciton absorption lines, it also blurs the hole subbands at large
length scales. At the length scale of a recollision, a few tens of nanometers, alloy disorder
does not mask the local avoided crossings, so the THz-induced birefringence is similar to
that in the 10 nm GaAs sample. At the length scale of the 200 µm NIR spot, however,
alloy disorder smears out the hole subbands. The ensemble of recollisions averages out
the four-fold symmetry to nearly cylindrical symmetry.
Sideband polarimetry
So far, these experiments have been performed assuming the NIR laser is perfectly linearly
polarized and that the sidebands have identical polarization states to the laser. Using
the polarimeter and techniques discussed in Sect. 3.1.2, the actual polarization states of
the NIR laser and sidebands were measured, see Fig. 4.9. In all of the measurements, the
NIR laser and all of the sidebands were measured to be perfectly polarized. They can
then be described by two angles, α, the angle that the major axis of the ellipse makes
with the THz polarization, and γ, the “degree” of ellipticity, see the center of Fig. 4.9.
In the 10 nm AlGaAs sample, the left half of Fig. 4.9, the polarization states of the
sidebands were measured for three NIR polarization states, one in the parallel excitation
geometry and two in the perpendicular excitation geometry. In the 10 nm GaAs sample,
the right half of Fig. 4.9, the polarization states of the sidebands were measured for four
100
γ
α
THzPol.
γ > 0
γ < 0
Figure 4.9: The measured polarization states of sidebands for various NIR laser polar-ization states. The angles of the NIR polarization ellipse, αNIR and γNIR, are circled forclarity. (center) The definition of the angles in the polarization ellipse, α is the anglethe major axis makes with THz polarization, and γ is the arctangent of the ratio ofthe semiminor and semimajor axes. The top row of the graphs plot α and the bottomrow plots the associated γ for the particular measurement. (left) Sideband polarizationstates for three different NIR polarization states, one with the NIR laser approximatelyparallel to the THz polarization and two with the NIR laser approximately perpendic-ular. The sideband polarization states are systematically different from the NIR laser.(right) Sideband polarization states for four different NIR polarization states, two withthe NIR laser approximately parallel to the THz polarization and two with the NIR laserapproximately perpendicular. The sideband polarization states again are systematicallydifferent from the NIR laser.
NIR polarization states, two for each polarization excitation geometry. For all data sets,
the sideband polarization states are measurably different, sometimes substantially so,
from the NIR polarization state. Furthermore, the sideband polarization states depend
sensitively on the NIR polarization state. Small changes in either angle of the polarization
ellipse of the NIR laser lead to large changes in the polarization ellipses of the sidebands.
It is difficult to construct a simple model to explain these observations, as these results
show that two linearly polarized lasers are combining to make elliptically polarized light
with seemingly random angles.
101
4.3 Theoretical model
In collaboration with theorists Qile Wu and Renbao Liu, full quantum mechanical and
semiclassical simulations of the HSG process were developed3. These simulations show
that, under experimental conditions, electron and hole trajectories explore roughly 10%
of the Brillouin zone around the Γ point. The trajectories of the spin 1/2 electrons are
entirely contained in the lowest electron subband, which is nearly parabolic. However,
the spin 3/2 holes travel through highly non-parabolic subbands that are strongly coupled
away from the Γ point. We are unable to account for the experimental results within the
simplest semiclassical model of electron and hole motion, in which ~k = eE(t) and the
phase velocity at any instant is determined solely by the dispersion relation E(k), where
~ is the reduced Planck constant, k is the crystal momentum, e is the electric charge,
E is the electric field, and E is the energy of the state [2]. Within the last 20 years,
it has become generally appreciated that such a description is incomplete and requires
the consideration of Berry physics [11, 62]. The Berry curvature Fmn(q) is defined by
calculating the Berry connection Rmn(q), which is analogous to the Berry phase in one-
dimension:
Fmn(k) = ∇k ×Rmn − iRmn ×Rmn (4.1)
Rmn(k) = i 〈umk|∇k |unk〉 , (4.2)
3As the theory was developed by Qile Wu and Renbao Liu, and the former performed all of calcu-lations, the following paragraphs will be an overview to explain the relevant physics rather than a fullderivation of the theory.
102
where |umk〉 is the non-plane wave or cellular part of the Bloch wavefunction of an
electron (eik·x |uk〉), m and n are the band indices spanning the bands included in the
calculation, and k is the crystal momentum [11]. The Berry curvature is zero in the
conduction subband but nonzero in the hole subbands. For this calculation, the Berry
connection is a 2×2 matrix, and therefore non-Abelian. The Berry connection rotates the
spinor that combines the HH1 and HH2 subbands, ~η = [|HH1〉 , |HH2〉]T , and describes
the hole state. Inclusion of the non-Abelian Berry curvature and connection in the
semiclassical equations of motion captures the crucial effects of spin-orbit coupling and
how the hole state mixes with other valence subbands during its trajectory. When the
electron and hole recombine to emit a sideband, its polarization state is determined by the
angular momentum of the electron and hole just before recombination. This calculation
reproduces all the main features observed in experiments, supporting the role of Berry
curvature in the high-order sideband generation.
4.3.1 Separation into circular components
To better consider how non-Abelian Berry connection in the valence subbands affects
HSG, it is useful to think of the electron-hole pair excitation process in the circular po-
larization basis to isolate individual electron and hole species, see Fig. 4.10. An incoming
NIR photon is separated into left and right circularly polarized components, σ−NIR and
σ+NIR, with complex coefficients given by the initial NIR photon. This conceit allows the
use of Berry physics. In “traditional” manifestations of Berry phase, only closed paths
103
create detectable a Berry phase because the dynamical phase of two interfering particles
otherwise is random. The HSG process is at its core not a closed path. The non-parabolic
hole bands mean the recollision will not necessarily be at the same real-space location as
creation, and the electron and hole will be at much different momenta. By initializing
every HSG event with the same relative phase, however, the Berry connection becomes
measurable. Once the electron and hole recollide, the electron state remains the same,
but hole state has changed dramatically.
Consider an exciton excited by a σ−NIR photon resonant with the HHX state, the top
of Fig. 4.10. Recall from Ch. 1 that subbands are defined by an envelope function fn of
the nth QW bound state and that, at k = 0, the electrons have spin-1/2, the heavy holes
have j = 3/2, mj = ±3/2, and the light holes have j = 3/2, mj ± 1/2. The electron
state is in the first electron subband with spin-up, |Ei〉 = f1 |S, ↑〉, and the hole state
begins as a heavy hole in the first subband |HHi〉 = f1 |3/2,+3/2〉. The electron remains
in the same subband with the same spin throughout its trajectory, |Ef〉 = |Ei〉. As the
hole accelerates, however, it mixes with nearby states in the valence subband manifold
to become a superposition |HHf〉 =∑
n,j ηn,jfn |uj〉 where ηn,j are coefficients calculated
from the non-Abelian Berry connection and |uj〉 is the spin state label. Not all of the
terms of this sum can recombine with the electron state, however. For a dipole-allowed
transition, the two states must be in the same subband and have an angular momentum
difference of ±1. In |HHf〉 there are components with |3/2,−1/2〉 spin, meaning that
there is nonzero probability for a σ−NIR to create a σ+HSG photon. The bottom of Fig. 4.10
104
NIRexcitation
Acceleration HSGemission
Figure 4.10: Mechanism for the creation of a single HSG photon including the Berryconnection of the valence band. There are three steps to this mechanism: NIR excita-tion, tunneling and acceleration, and HSG emission. First, the NIR excitation light isdecomposed into left and right circularly polarized light to isolate its coupling to a singleelectron species species, each with complex coefficients determined by the NIR polariza-tion. A σ−NIR (σ+
NIR) photon resonant with HHX creates a f1 |S, ↑〉 (f1 |S, ↓〉) electron anda f1 |3/2,+3/2〉 (f1 |3/2,−3/2〉) heavy hole. These particles then accelerate, driven bythe THz. During acceleration, the electron wavefunction does not change character. Theheavy hole wavefunction, however, mixes with the nearby hole subbands. Upon recolli-sion, the hole state is no longer simply f1 |3/2,+3/2〉 (f1 |3/2,−3/2〉), but a superpositionof several different hole subbands
∑n,j ηn,jfn |uj〉 with ηn,j coefficients determined by the
non-Abelian Berry curvature and the initial phase of the photon that created the hole,fn the subband label, and |uj〉 the spin state label. Some of these states have allowedelectric dipole moments with the electron, and so create photons σ±HSG that then interfereto create the HSG emission. Controlling the relative phase of the σ+
NIR and σ−NIR lightinfluences the output by changing how the emitted photon states interfere.
105
demonstrates the same physics but starting with excitation by a σ+NIR photon.
4.3.2 Semiclassical trajectories
To demonstrate how the non-Abelian Berry connection relates to HSG trajectories, two
semiclassical trajectories that are influenced very differently by relative NIR and THz
polarization are shown in Fig. 4.11. As can be seen in Fig. 4.7(c), the relative strength of
the parallel- and perpendicular-excited sidebands of twentieth and sixtieth orders are very
different. The sixtieth order is not even observed in the parallel-excited HSG spectrum,
for example. The phase of the THz field for ionization and recollision for the two orders
is labeled in the upper right of Fig. 4.11. The trajectory resulting in the sixtieth order
sideband is over 100 fs longer (more than 30% longer) than the twentieth order trajectory,
so it has more time to accelerate in the THz field.
The details of the twentieth order trajectory are plotted in the lower left two graphs in
Fig. 4.11. Over the course of the acceleration step, the electron and hole paths separate
by almost 30 nm at their farthest. The spinor state of the two particles is plotted at
seven different times (the spinor direction is chosen for excitation by a σ+NIR photon). The
electron spinor direction does not change, but the hole spinor rotates by a large amount
in the last 50 fs. The same seven instants for the hole state are shown on the dispersion
relation below, where the area of the circles represents the norm of the spinor coefficient.
At instant #1, the hole is entirely in the HH1 subband. As the hole accelerates to the
left, instant #2, it only slightly mixes with the HH2 subband. As the hole accelerates
106
12 3 4
5 6
71
23 4 56
7
8
z
xpseudo-spin
Figure 4.11: Semiclassical trajectories for the n = 20 and n = 60 sidebands in the 10 nmAlGaAs sample. (upper left) Full perpendicular HSG spectrum from the 10 nm AlGaAssample oriented at 93. (upper right) Time trace of the THz electric field. The arrowspoint to the time of ionization and recollision for the two sidebands modeled here, inand rn for the time of ionization and the time of recollision. (middle left) The real-spacetrajectories of the electron and hole for the 20th order sideband. The arrow representthe spinor direction in x-z plane for seven instants throughout the trajectory. The holespinor state is an approximately equal superposition of HH1 and HH2 at recollision.(lower left) The location of the hole in the valence subbands at those seven instants withthe area of the maroon circle representing the relative weight in either subband. (middleright) The real-space trajectories of the electron and hole for the 20th order sideband,with the spinor direction drawn for eight instants. The hole spinor almost entirely flipsto the HH2 state. (bottom right) The location of the hole in the valence subbands atthose eight instants.
107
to the right, instants #3–7, the hole spinor shifts significantly as it accelerates through
the avoided crossing. The majority of the spinor weight is then in the HH2 subband
where it cannot recombine with the electron via an electric dipole transition. Also, in
the position plot again, notice that the position at recollision is not at exactly zero, it
is slightly positive. This translation results from the hole becoming “more” massive as
it enters the region of the HH1 subband with less curvature. If the electron and hole
remained the same mass the entire time, the recollision would occur at exactly zero.
The details of the sixtieth order trajectory are plotted in the lower right two graphs in
Fig. 4.11, and they tell a far more exaggerated story. The electron and hole travel much
farther apart for this trajectory, but, more significantly, the hole spinor state rotates
substantially twice, shown now at eight instants in the trajectory. At t = 120 fs, instant
#3, the hole sits right in the avoided crossing, and some weight is transferred to the HH2
subband. As it accelerates to the right, instant #4, some of the weight remains in HH2.
As it continues to accelerate to the right in the last 100 fs, the spinor weight almost
completely shifts to the HH2 subband. Also, notice that the location of recollision has
shifted almost 10 nm.
Understanding how the two polarization geometries can result in such different side-
band strengths requires looking into how the trajectories excited by σ+NIR and σ−NIR pho-
tons interfere. The semiclassical results above only demonstrate the large scale of the
mixing calculated by the non-Abelian Berry curvature. The entirety of the HSG process
including non-Abelian Berry curvature can be written in one simple matrix for every
108
sideband order n, Tij,n in the equation:(σ+
HSG
σ−HSG
)n
=
(T++ T+−T−+ T−−
)n
(σ+
NIR
σ−NIR
). (4.3)
The elements of Tij,n contain all of the terms that create the i = ±-polarized HSG photon
from the j = ±-polarized NIR photon for a given THz polarization and lattice orientation.
The NIR polarization inputs different complex coefficients on the σ±NIR components and
causes different terms in the matrix product to constructively or destructively interfere.
The full calculation of Tij,n is very complicated. In addition to including the HSG
physics, it includes a full calculation of the dispersion relations and non-Abelian Berry
curvature for the six QW subbands for all four hole species and scattering as a simple
dephasing rate that does not couple subbands. It is an open question as to what kind of
scattering to include and how to actually include it. By taking the ratio of sidebands in
the two polarization geometries, however, some of the possible scattering mechanisms can
be divided out. Theoretical ratios and experimental ratios are compared in Fig. 4.12. The
top row of Fig. 4.12 compares theory and experiment for the 5 nm GaAs sample at both
lattice angles. The calculated and measured ratios do not agree very well, particularly at
55, likely due to the hard wall approximation which breaks down for narrow QWs and
leads to an overestimate of the Berry connection. The middle row of Fig. 4.12 compares
theory and experiment for the 10 nm GaAs sample. Theory and experiment agree much
better here. The calculated ratio predicts a large peak at a sideband offset energy of
about 125 meV, which is partially observed in the measurement. The source of this
peak, however, is not understood. The bottom row of Fig. 4.12 compares theory and
109
1
1 0
1 0 0
1
1 0
1 0 0
0 1 0 01
1 0
1 0 0
0 1 0 0
E x p e r i m e n t C a l c u l a t i o n
5 n m G a A s
8 5 ° 5 5 °
9 1 °
1 0 n m G a A sSid
eban
d rati
o (I ⊥ /
I //)
6 3 °
9 3 °
1 0 n m A l G a A s
S i d e b a n d o f f s e t e n e r g y ( m e V )
4 7 °
Figure 4.12: Comparison of experiment and theory. The ratio I⊥/I‖ is compared inall three quantum wells at different lattice angles. The non-Abelian Berry connection isresponsible for the qualitative agreement. The quantum theory of sidebands only includesmacroscopic dephasing, no k- or t-dependent scattering. Comparing intensity ratiosinstead of absolute intensities cancels out some of the complicated scattering behavior.
experiment for the 10 nm AlGaAs sample. Theory and experiment agree here as well.
The calculated ratio has a much smaller peak than in the other 10 nm sample, and the
measured ratio is follows this trend. Overall, this model matches the measured data quite
well.
A dramatic, and surprising, prediction of this non-Abelian Berry connection model
is the rotation of sideband polarization states relative to the incident NIR polarization
state. What initially began as a simple (if technically difficult) exploration of the source of
the stark difference between HSG spectra for different polarization excitation geometries
became the strongest evidence in support of this model, see Fig. 4.13 for a comparison of
110
experiment and theory (the experimental data is replotted from Fig. 4.9). The detailed
interference of different electron and hole trajectories results in a sensitive dependence of
the final sideband polarization state on the incident NIR polarization state. Calculations
in the Tij,n formalism, when the NIR polarization state as the input variable and no free
parameters, show surprising agreement with experiment in the 10 nm AlGaAs sample at
93, Fig. 4.13(left) It should be noted that the models contained in Tij,n are incomplete
and that quantitative agreement should not be expected, recall that scattering is not
perfectly accounted for and the QW properties are calculated assuming hard walls. The
qualitative agreement that individual sideband polarization states change with order
and depend sensitively on NIR polarization state is significant, though, and shows the
importance of considering Berry physics when thinking about HSG.
4.4 Conclusion and remaining questions
In conclusion, we have studied how the interplay between the relative orientations of
the NIR polarization, THz polarization, and lattice affect HSG. We have measured HSG
spectra of up to 86th order and spanning over 200 meV, a bandwidth of over 12% of
the NIR laser frequency, by manipulating those relative orientations. We have shown
conclusively that electrons and hole accelerate coherently through the the lattice before
recolliding. This coherence allows for interference between different electron and hole
states, initialized by the NIR polarization and caused by non-Abelian Berry curvature
in the hole subbands, and leads to large changes in sideband strength and sideband
111
γ
α
THzPol.
γ > 0
γ < 0
Figure 4.13: Polarization state of the sidebands. Experimental measurements are filledscatter points (replotted from Fig. 4.9, theory are empty scatter points. Polarization statemeasurements were performed on both 10 nm samples for both parallel and perpendiculargeometries, and all the measurements were duplicated with approximately the same NIRlaser polarization except for one. The NIR laser polarization state is plotted as theorder zero sideband and circled in black for each measurement. Overall, both theory andexperiment agree that he polarization state of a given sideband is extremely sensitive tothe NIR laser polarization state. In the 10 nm AlGaAs sample, theory and experimentagree almost quantitatively.
112
polarization state.
The results presented here point to several interesting future experiments. The rela-
tionships between HSG and either the NIR power or frequency are poorly understood.
The nonlinear dependence of HSG on the NIR power presented in Fig. 4.4 is likely due
to heating. If heating could be removed, however, then there are potentially interesting
effects that occur at high exciton densities. Furthermore, the injection of non-excitonic
electron-hole pairs by tuning the NIR frequency could also result in interesting phenom-
ena such as the observation of a large number of negative sidebands predicted in [63].
In the next experiments, the observations discussed here should lead to a new genera-
tion of complete band structure measurement because HSG is inherently sensitive to both
elements of the Bloch wavefunction, the plane-wave dispersion relation, eik·x, and the cel-
lular function, |uk〉. By clever control of the NIR polarization, the THz frequency and
field, and lattice orientation, a self-consistent algorithm can be developed for the direct
measurement of the electron and hole dispersion relations, non-Abelian Berry curvature,
and even k- and t-dependent scattering rates in any material. For example, exciting with
purely circularly polarized light will isolate any polarization changes to just one electron
and hole species, measuring the non-Abelian Berry curvature more directly by eliminat-
ing the complex interference generated by linearly polarized light. Then, by tuning the
THz frequency and field strength, the timescales for sidebands of the same order or offset
energy can be changed to measure different scattering rates. Because the THz polariza-
tion must be linear for high-probability recollisions, all the proposed experiments would
113
have to be repeated for several different THz polarization-lattice orientations.
114
Appendix A
Equipment details
This appendix includes descriptions of most of the equipment used in the experiments
for this dissertation. Many descriptions of the equipment used in this appendix comes
from the theses of Ben Zaks and Sam Carter.
A note on software In redesigning the experimental setup, a decision was made
to move from using LabVIEW to Python and PyQT. Although development time in
LabVIEW from new instrument to functional graphical user interface (GUI) can be
quite fast, development of more advanced interfaces and even modest number crunching
routines is much simpler in Python. Furthermore, Python works on all major operating
systems and is free as in speech and as in beer, so installing on new and home computers
is trivial. Programs are kept up to date and consistent using Github, a cloud-based
git repository service, under a collaboration named SherwinGroup. It is a little bit like
Dropbox but aimed at collaborative programming. It allows for editing the same file
115
by multiple people at the same time and handles merging all of the changes. The old
LabVIEW code is still on the computer in 1380D, called 1380D-PC.
A.1 Cryostat
The cryostat is a CTI Cryogenics Model 22 closed-cycle helium refrigerator that can
cool to a minimum temperature of 10 K in about an hour. The cryostat is connected
to a CTI Cryogenics Model 8200 helium compressor through two high pressure helium
lines. The cold head is mounted on a linear bearing to allow tight optical access as
well as simple sample access. Bungie cords and a micrometer provide fine left-right
control of the sample in the optical path. The cryostat currently has two electrical
connections, one to the compressor and one to a Lakeshore 330 temperature controller.
The samples are glued to a copper bar with rubber cement. For quick sample changes,
the sample bar is bolted onto a copper mount that is semi-permanently attached to
the cryostat. A Lakeshore temperature diode is attached to this mount to monitor the
temperature of the samples. The copper mount is attached to a three-axis dove-tail
translation stage, an Elliot Scientific ultra-small micropositioner stage MDE269, with
3 mm of travel in x, y, and z. This stage can be translated when the sample is cold
using Allen wrenches fed through the vacuum shroud. The temperature of the sample
will spike to 30 K or so because the wrenches are room temperature, but it should settle
back down. A molybdenum-based spray dry lubricant prevents sticking and eases sliding
at all temperatures. The stage is a poor conductor of heat, so the outer conductor of
116
a BNC cable provides thermal contact between the copper sample mount to the copper
cold finger mount. The cold finger mount has a 22 Ω nichrome wire heater attached
at the base with stycast. The copper cold finger mount bolts to the cold finger of the
cryostat. All interfaces have Apiezon N grease for improved thermal contact. The cold
finger itself has wires for electrical contact, but these are not currently set up.
Several pieces make up the thermal radiation shield and the vacuum shroud. Both
the shield and the vacuum shroud are made up of two main pieces, a cylindrical part
and a head part. The cylindrical parts are semi-permanent and attach to the actual
cryostat. The head parts attach to the cylindrical parts and allow for easy sample access.
The vacuum shroud head has three rotary feed-throughs for cryogenic control of the
three-axis translation stage, as mentioned earlier. There are also two large windows for
optical access. The current front window is most likely made of suprasil quartz, which has
decent THz transmission and is not birefringent nor optically active at either THz or NIR
frequencies. The transmission has been measured by the VNA, see Fig. A.1. It is only
attached using vacuum, and so it must be held in place during pump down. The back
window is glass with a NIR anti-reflection coating. It is clamped in place using plastic
straps. With all of these o-ring seals and large optical windows in the thermal radiation
shield, the temperature can only reach about 14.5 K. The o-rings of the feed-throughs
and the windows need to be checked and re-greased every few months.
The current vacuum shroud and heat shield are designed to allow for THz-NIR beam
alignment using an indium-tin oxide (ITO)-coated glass slide as a dichroic beam com-
117
5 0 0 6 0 0 7 0 00 . 0
0 . 2
0 . 4
0 . 6
0 . 8
1 . 0
Powe
r tran
smiss
ion
F r e q u e n c y ( G H z )
C u r r e n t c r y o s t a t w i n d o w E x t r a p o l a t i o n u s i n g f i t
T = 0 . 0 9 6 9 * s i n ( π( f + 4 . 3 4 ) / 1 1 . 2 5 )+ 0 . 8 4 6
Figure A.1: The power transmission of the cryostat window as measured by the vectorreflectometer. The fit only includes one band from the VNA, a more careful measurementmay be required for frequencies too far away.
biner. The ITO slide is 50 mm by 75 mm with a sheet resistance of about 10 Ω/ and
is mounted to a tip-tilt stage for independent NIR and THz alignment.
Cool down procedure
1. While holding the suprasil on the cryostat, rough pump the cryostat. Once the
window is fixed, pump the cryostat to 10−6 torr or so for at least 2 hours, overnight
is best.
2. Turn on building cooling water to the compressor.
3. Turn on compressor and wait until the the sample is cold, approximately 1.5 hours.
Close pump valve when thermometer goes below 77 K. This protects turbo pump
from being contaminated with rough pump oil, at least theoretically.
4. Once stabilized at low temperature, the temperature controller can hold the tem-
118
perature anywhere up to 150 K of so.
5. When finished or maintaining temperatures above 60 K, it is important open the
pump valve to a rough pump because vacuum leaks will bring the pressure up to
about 10−2 torr when the air boils off.
The cryostat will generally stay below 17 K for at least a day. It is possible with more
vacuum-tight shrouds for the cryostat to remain cold for two or three days. Manipulating
the translation stage makes the leaks more severe, but the cryostat will hold below 20 K.
Maintenance
If the temperature seems to be creeping up more quickly than usual, regreasing or re-
placing the o-rings on the vacuum shroud head should make it more stable, though the
feed-throughs will be more slippery and prone to being sucked into unintentional contact
with the cold parts inside. Every six months, the helium pressure in the compressor
needs to be checked, and the manual has instructions for filling it and the parts for doing
this should be in the cabinet nearby. If the cryostat needs to be moved, there are special
wrenches to remove the high-pressure lines.
A.2 Sample imaging camera
A major quality-of-life improvement the experiment is the sample imaging camera, in-
troduced to us by Kyle Seyler of the X. Xu Lab at the University of Washington, Seattle.
119
This camera allows for simple sample-NIR alignment as well as convenient sample mon-
itoring during experiments. For excitation sources with long focal lengths, the basic
camera schematic is shown in Fig. A.2. For short focal length experiments, the exci-
tation lens can be used as Lens B, but this arrangement is less common for large-area
quantum well samples. The optics are tucked away in between and behind the important
THz and NIR optics, so that the camera can be used without needing realignment or
disrupting a measurement. Lens A, the “eyepiece” lens, is mounted on a lens tube to
the camera to block out unwanted light and so the camera is fixed to be at the focus of
Lens A, fA. Lens B, the “objective” lens, is mounted so the sample is at its focus, fB,
when reflected off a 50:50 pellicle beam splitter (BS) (1” diameter). There is deliberately
nothing at the focus of the transmission path to avoid double-imaging. A scale bar of
known size could feasibly be mounted at the focus, if more quantitative imaging was
desired. The ratio of the focal lengths fB/fA is the magnification. The actual sizes of
fA and fB determine the amount of light collected. For broad illumination, an bright
lamp (likely halogen) is coupled into the beam path between Lenses A and B using a
2” diameter 50:50 pellicle BS. The indium-tin oxide beam combiner (ITO BC) overlays
some small interference fringes, but they generally aren’t a problem.
Alignment for long excitation focal lengths starts by fixing the location of the camera,
then Lens A, then the lamp and illumination pellicle, then Lens B and the imaging
pellicle. To start, mount Lens A on a lens tube to the camera that is approximately
the focal length of the lens. An adjustable lens tube is recommended for peaking up
120
Camera
Lens tube
Lamp50:50pellicle
50:50pellicle
Lens A
Lens B
Excitation source
(for alignment)
Sample incryostat
ITO BC
Figure A.2: The optical setup for the sample imaging camera.
the image focusing later. Then, align Lens B to be collinear with the lens tube. Try to
mount Lens B close to where it will be in the final setup, but this is unnecessary. Turn
the camera on. Using a beam block or a magnetic ruler with some lines or words written
on it, find the location in front of Lens B that has a vaguely sharp image on the camera
(do not use a pellicle BS at this point, just keep the optical path straight). A business
card taped to a beam block works really well for this. Peak up the locations of both
lenses to get the sharpest image on the camera. Use the lamp to illuminate the beam
block from an angle if more light is needed (it usually is). Once the image is sharp, lock
down the position of the camera and Lens A as much as possible, place posts to defend
from accidental bumps, and measure the distance between Lens B and the beam block.
Now, mount the illumination pellicle on a tip-tilt stage between Lenses A and B, and
couple the lamp into the beam path. Peak up the brightness of the image on the camera
by tweaking the pellicle and translating the lamp (and maybe the pellicle) around. Once
121
the image is good, lock both down. At this point, the camera is at the focus of Lens
A, and the axis of the lens tube and the lamp are collinear. Now, using the distance
measured between Lens B and the beam block, mount Lens B and the imaging pellicle
so that the sample is that distance from Lens B. Be careful to ensure that the excitation
source (the Ti:sapphire laser or the differential transmission source) are not blocked by
the pellicle, and put the imaging pellicle on a magnetic clip or flip mount so it can be
removed reliably. With a sample in place, peak up the location and angle of Lens B
and the imaging pellicle to get the sharpest image on the camera. Only then should the
alignment of the lamp or the position of Lens A be tweaked to ensure the highest-quality
image.
This setup makes sample-NIR alignment extremely easy. The Ti:sapphire laser at
low power and 50% duty cycle should be readily apparent on the sample now. The FEL
alignment laser can technically be seen, but its spot is usually less than perfect. Both
the tungsten-halogen lamp and the LED (from the differential transmission experiments)
can be seen with this camera, even with a 30 µm spot size.
A.3 M2 SolsTiS Ti:sapphire laser
The M2 SolsTiS Ti:sapphire laser is a continuous-wave NIR laser pumped by a 7 W,
532 nm Sprout laser from Lighthouse Photonics. It outputs more than 1 W of power
across its prodigious tuning range of 700–1000 nm, and it is a single mode laser with a
linewidth of less than 5 MHz. The nicest feature, however, is that it cannot be opened by
122
the user. Occasionally the pump alignment needs to be peaked up, but otherwise there
is no maintenance that even can be done to the laser itself beyond cleaning the windows
(which hopefully never has to happen).
The laser is controlled via ethernet and a local network IP address. The laser is tuned
using software. The birefringent filter provide the coarsest tuning. It works using a look
up table to associate wavelength and filter angle, and can get to within 0.5 nm of the
desired wavelength. The etalon yields slightly finer tuning. It can move the laser about
0.5 nm total in steps of about 0.05 nm. The laser can technically lock itself using the
etalon, but this mode is prone to mode hopping. Finally, one of the resonator mirrors
can be tuned with two settings, coarse and fine. These can theoretically move the laser
wavelength about 0.5 pm and 0.005 pm, respectively, but we don’t have a detector precise
enough to actually measure that kind of tuning. This level of tuning should give mode-
hop free tuning.
The software is also capable of diagnosing certain electrical problems within the laser,
but only with M2 support close at hand.
For safety and the elimination of detectable stray light, both laser heads and much
of the optics are kept in a laser housing made of anodized aluminum sheets and 80-20
posts. There is a hole cut into the top sheets covered by square of laser curtain material
to allow easy but safe access to the half-wave plate attenuator.
123
Alignment
About once a month, the pump alignment should be peaked up to ensure that it does not
drift too far. There is one mirror accessible externally with three knobs. The diagonal
knobs are tip-tilt, and only they should be touched. The middle knob, the point of
the “L,” should never be touched. The alignment is extremely sensitive, so be careful
when peaking it up. To measure the laser power easily, there is a flip mirror about two
inches in front of the laser head. It redirects the laser into a power meter head with a
1000:1 attenuator on it, and the power meter readout is kept on top of the SPEX. It is
highly recommended never to move this power meter as alignment of the attenuator is
not trivial.
Maintenance
The only other maintenance that needs to be done on this laser and its pump laser is to
change the fluid in the chillers every six months. The SolsTiS chiller is a dedicated unit.
The Sprout chiller is contained in the power supply. The plumbing needed to drain and
refill the chillers are found on top of the Sprout power supply. Both chillers use distilled
water with 10–20% isopropanol? Distilled (not de-ionized) water can be bought at the
grocery store if it’s not available in the chemistry storeroom.
124
A.3.1 Polarizer attenuator
The Ti:sapphire laser outputs as much as 1.8 W continuous-wave power. That much
power is certainly too much for most experiments, but it is also too much for some
optics. Any optic with even a small amount of absorption will heat up wherever the
beam contacts it and expand. This expansion is called thermal lensing, and it causes
Gaussian beams to become donut-shaped. One might think that expanding the beam
diameter would limit thermal lensing, but the increased surface area of the thermal lens
roughly cancels out the decrease in thermal expansion. The easiest way to reduce thermal
lensing is to attenuate the beam using optics designed for high powers. In order to keep
the power controllable, the polarized Ti:sapphire light is sent through a half-wave plate
and then through a high-power polarizer. The component of polarization that passes
through the polarizer can be controlled by rotating the half-wave plate. Be aware that
rotating polarizers causes slight lateral beam shifts based on the prism spacing, but
rotating rotating a half-wave plate does not. The unwanted power is dumped into a
razor blade stack.
A.4 PhotonControl spectrometer
The PhotonControl SPM-001-N spectrometer is a compact, fiber-coupled spectrometer
capable sensing light from 680-920 nm. It is used to quickly measure the wavelength
of the Ti:sapphire laser. The manual claims the resolution is 0.24 nm, but experience
reveals that the Gaussian fit algorithm can tell when the laser moves by 1 1/cm (about
125
0.06 nm at 800 nm). The M2 Ti:sapphire laser can theoretically be sent to any wavelength
between 700 and 1000 nm, but it only works using by rotating a birefringent filter based
on a lookup table. The actual wavelength can differ by as much as 0.5 nm based on
etalon and resonator settings. A very small amount of laser power is picked off and sent
to the PhotonControl for precision laser tuning. By “very small amount of laser power,”
a pellicle picks off part of the beam and sends it at the side of a fiber coupler. The diffuse
reflection off the pellicle is enough to measure the laser wavelength.
Software
The PhotonControl spectrometer is connected to the computer via USB, and it is con-
trolled with a proprietary LabVIEW-esque interface that we cannot seem to talk to with
external software.
A.5 Acousto-optic modulator
An IntraAction AOM-602N acousto-optic modulator (AOM) is used to as a fast electronic
shutter for the NIR beam. The IntraAction Model ME signal processor drives a piezo
in the AOM with an RF signal, creating a density Bragg grating that is capable of
diffracting approximately 65% of the input power into a first-order diffracted beam, also
sometimes called a sideband. The first-order beam has a contrast ratio (Pon/Poff) of at
most 3000, which is enough for the SPEX but often not enough for imaging spectrometers
like the CCD and the ICCD. The driver is controlled by a TTL pulse. The switching
126
time depends on the beam size, but it is approximately 100 ns.
Alignment
1. Ensure that the AOM is level and the aperture is 5” above the table. For simpler
alignment peaking up in the future, it is useful to mount the AOM on a transverse
horizontal translation stage and a z-axis rotation stage.
2. Without driving the AOM, send the laser straight through the AOM and peak up
the power. Ensure the laser is 5” above the table before and after the AOM.
3. Apply a 50% duty cycle TTL signal to the AOM and turn the carrier level to 10.
4. Rotate and translate the AOM around until a sideband appears. The AOM deflects
the sideband by about 2. Depending on the direction of rotation, this could be to
the left or the right of the zeroth-order beam.
5. Block all sidebands (I’ve seen as many as three total sidebands at any given time)
except the first-order one. Measure the power and peak up the laser power going
into teh sideband the rotation and translation stages. It should be possible to get
about 65% total power into the first order beam (legends claim some AOMs are
capable of > 85% efficiency).
6. Be sure to block the zeroth-order beam using a razor stack or other absorber. The
higher-order beams can be blocked with an iris if necessary.
127
A.6 Mechanical shutter
A Uniblitz shutter driven by a Uniblitz model SD-10 shutter drive timer controls the
output of the laser out of the laser enclosure. The contrast ratio of the AOM alone is
not high enough to block the laser given the duty cycle of the FEL. With a duty cycle of
10−6 and a contrast ratio of 103, approximately one thousand times more laser light hits
the sample, possibly causing photoluminescence, and enters the CCD, possibly damaging
the CCD or at least raising the noise floor substantially, when the AOM is “off” than
when it is “on.” Although the shutter has an opening time of about 8 ms, it completely
blocks the laser when closed, bringing that factor of a thousand to less than ten. The
shutter driver is controlled with a TTL pulse.
A.7 Absorption LED
Driven absorption measurements require fast pulses of broadband light. A Thorlabs LED
on a metal-core printed circuit board can be driven with the HP 214B pulse generator to
make pulses as fast as 25 ns with rise and fall times of about 10 ns. The pulse generator
can drive a 50 Ω load up to 10 V. Above 10 V, it can only drive “high Z” loads, but up
to 200 V. The LED impedance drops very quickly at about 1.5 V, so the pulse generator
acts more as a current source above this voltage.
128
A.8 SPEX monochromator
Any light within about 10 nm of the laser has to be detected using the SPEX 1403 0.85 m
double monochromator. The SPEX has two 1800 line/mm gratings, making its efficiency
very low, about 10%, but its rejection extremely high. The detector won’t see any laser
power at all when the SPEX is about 10 wavenumbers away (this number has not be
tested). The entrance slits are almost always set to 400 µm, giving a resolution of 1
cm−1. Incident light must be horizontally polarized or the efficiency will be much lower
but not zero, so polarization-sensitive measurements need external polarizers.
Besides measuring the low-order sidebands and the laser itself, the SPEX is extremely
useful for ensuring good alignment between the NIR and THz spots for sideband measure-
ments. The ITO beam combiner can move the THz spot around easily without affecting
the NIR spot or alignment. First, peak up the power in the second order sideband, then
move to a higher order one, the sixth or the eighth, and peak up again. This alignment
process needs to be done first thing in the morning before any experiment, and may need
to be done several times depending on the day’s experiments. If more careful alignment
is necessary, check with the CCD as well. The feedback with the CCD is much slower,
though.
Alignment
1. Find the white mirror cover, a 6” diameter metal plate with five small holes, which
should be under the SPEX whenever not in use. This plate isn’t necessary, but it
129
will make alignment much easier.
2. Ensure the laser beam is collimated going towards the SPEX, but block it from
entering the SPEX. Put a lens on a three-axis stage with a focal length to match
the f/7.8 of the SPEX.
3. Turn the detector off, and set the SPEX wavenumber to far, far away from the
alignment laser line.
4. Open the hatches on top of the SPEX and put the mirror cover over the first mirror.
5. Unblock the laser and send through the 400 µm slit, don’t try too hard.
6. Adjust the beam to be centered horizontally and vertically on the slits and on the
mirror cover.
7. Place the focusing lens at the appropriate distance from the slits, and repeat the
alignment, centering the beam on the slits and the mirror cover.
8. For best alignment, which may not be necessary, choke the slits down to the calcu-
lated spot size and repeat the alignment.
9. With the detector blocked, or even removed, set the SPEX to pass the laser and
ensure that the light is hitting the detector slits. Hope that it is.
10. Remove the mirror cover and put back under the SPEX, and put everything back
together.
130
11. Using either the laser (attenuate by. 60 dB) or even photoluminescence/sidebands,
peak up the signal on the detector by translating the last lens or the last mirrors.
12. Set up irises so that you don’t have to do this process again any time soon.
Software
The SPEX gratings can only be controlled with a computer. Being so old that GPIB was
new, the SPEX is controlled by a SPEX MiniStepDriver which itself plugs into another
box which can then connect to the computer via GPIB. These are left on all the time.
Occasionally they must be power-cycled, or the power goes off. They must be turned on
in the correct order, so if they don’t work the first time, turn them off and turn them on
in the other order.
To minimize backlash when rotating the gratings, the SPEX is designed to always
step down in wavenumber when moving. If a higher wavenumber is needed, the SPEX
will go to twenty cm−1 above the desired location and then rotate back. The physical
readout on the side of the SPEX is higher than the actual wavenumber setting of the
optics path by about 4 cm−1, so be aware if things look wrong.
There are two sets of software to control the SPEX. Ben Zaks modified the Optilab
code, Vinh Nguyen’s LabVIEW code for the VNA, to operate it and the oscilloscope.
This software has been replaced with Python code in SherwinGroup/SPEXSidebands-
fuzzy. They both work in the same way, but the Python code is much more advanced.
This software also controls reading from the oscilloscope.
131
A.9 Hamamatsu photomultiplier tube
A Hamamatsu R7400U-20 photomultiplier tube (PMT) detector was mounted on the
SPEX. The response time is very fast, approximately 1 ns, and the noise-free gain is
large, allowing for fast single photon detection. The efficiency of the PMT is low, around
10%, and the response dies off dramatically above 830 nm. The full response curve is
taped to the detector. A voltage of −700 to −1000 V is applied using an SRS high
voltage source. The gain ratio between 1 kV and 700 V is almost exactly 10. The output
current is sent to an SRS 445 preamp set to 50 Ω sent through two channels of 5 gain.
Single photon peaks are approximately 50 mV tall.
A.10 Acton spectrometer
For broadband optical measurements, an Acton SpectraPro 2750 spectrometer is used.
The total focal length is 275 mm, and there are two input ports and two exit ports.
Currently, only the side entrance slit and the end exit port are used. The mirror that flips
in for the side exit port does not block the end port entirely, so detectors mounted there
could be damaged by errant bright light. There are currently three gratings mounted in
the spectrometer, the bandwidth and resolution are approximations:
132
Grating Groove density Blaze Bandwidth Resolution
no. (g/mm) (nm) (nm) (GHz)
1. 1800 500 13 8
2. 1200 1000 24 16
3. 600 1000 53 30
Table A.1: Gratings currently installed in the Acton SP-2750. Resolution is approximated
by the spacing of three pixels in the data array, the approximate linewidth of a low-order
sideband (which itself should have a linewidth of approximately 1 GHz, far smaller than
the resolution of the spectrometer).
All of the gratings generally prefer vertically polarized light, the efficiency is higher and
flatter throughout the center wavelength range. Acton/Princeton Instruments have a
website with this information for lots of gratings.
The grating turret inside the spectrometer has two “stable” positions, one that has
been calibrated, and one that is approximately 1 nm off. When measuring sidebands,
it is generally easy to tell when the spectrometer has “jumped” as long as the laser
wavelength is well known externally. It is less obvious when measuring PL or absorption.
This jumping seems to be entirely random, but possibly more common at certain center
wavelengths. Getting the spectrometer to jump back to the calibrated position requires
shifting it around a half a nanometer or so at a time until it is where it says it is based
on the measurement of a known wavelength. This bug is very, very frustrating.
133
Software
Software to control the Acton spectrometer in SherwinGroup/InstrumentLibrary-yolo.
The CCD camera software more directly controls the spectrometer.
A.11 Andor electron-multiplying CCD camera
The Andor NewtonEM electron-multiplying (EM) CCD camera is the primary HSG
detector because it is very low-noise and can measure many sidebands in parallel. The
primary advantages of the camera are the Peltier cooling down to -90 C requiring only
minor water cooling, an EM amplifier, and hardware binning. Cooling to -90 C allows for
5 minute integration times with minimal dark current noise. The EM amplifier allows for
low light signal, signal on the order of single photons, to be amplified “noise-free” above
the readout noise. For sideband measurements, readout noise is the limiting factor of the
detector. Hardware binning allows for binning before amplification or readout, so small
signals can be summed before the multiplication and readout. Current limits of detection
are about 0.5 sideband photons per cavity dump pulse, or about 0.5 photons per second
when integrating for twenty minutes (four 5 minute images). Achieving this sensitivity
requires 5 minute integration times, the chip to be at -90 C, and a gain setting of 110,
which corresponds to real gain of about 6-7 (it is extremely sensitive to temperature).
In addition, the 1380D ceiling lights, the hallway lights, and the FEL vault lights must
be off, and the transport system windows in room 1380C must be covered in aluminum
foil. It is likely that a shutter before the spectrometer blackout box triggered to open
134
only during the FEL pulse would relax these requirements.
The background of the CCD is complicated. Increasing the EM gain It is not truly
noise free, as increasing the gain increases the amplitude of the background noise. The
cause of this background is not well understood, as it is most apparent when binning a
large number of pixels and integrating for a substantial time. It is likely from individual
electrons of dark current summed over tens of pixels. Unfortunately, this background
noise is substantially larger, though usually still small, on the left side of the chip, which
is the shorter-wavelength side. It is as if the bottom left side, which does not seem to
be the readout corner, is warmer than the right three quarters of the chip. It is unlikely
that it is actually warmer, but the noise behaves similar to dark current. It is currently
unknown what the longest integration time is to maximize the signal-to-noise ratio when
the camera is at -90 C.
Optical edge filters
Safe and effective use of the CCD for HSG measurements requires decent long pass filters
and very good short pass filters. To measure positive sidebands, the long pass filter
blocks amplified spontaneous emission from the laser and the short pass filter blocks the
laser from the CCD. It is difficult to say with certainty, but the short pass filter must be
able to attenuate the laser by at least 60 dB, likely more 70 dB to suppress laser scatter
in the spectrometer. In general, Semrock makes the best filters. Rotating filters away
from normal decreases the transition edge, and good filters can be shifted almost 100
nm, although at that point the filter mount physically blocks the light more than the
135
filter fails. A more reasonable tuning range is 40 nm. There is currently one very good
long pass filter with an edge of 808 nm. It can be angle-tuned down to 765 nm easily.
There are two very good short pass filters, 847 nm and 785 nm. These both have very
sharp transitions from transmission to reflection. The 847 nm short pass filter is good
enough to allow CCD measurement of sidebands just 6 nm from the laser line. The 785
nm short pass filter is sharp enough to allow measurement of sidebands 4 nm from the
laser line. When changing filters, never pull by the optic holder, only by the post. The
optic slips easily and spins if pulled by the frame, often resulting in huge fingerprints on
either face of the filter. Cleaning Semrock filters is annoying because of their shape.
Alignment
1. Block the spectrometer.
2. Send the laser with 50% duty cycle through the laser polarizer and through the
short pass filter. Place a power meter behind the filter and note the power, usually
tens of milliwatts.
3. With the filter reflecting light towards the edge of the table (not at the laser box
wall), rotate the filter so that only some of the light is transmitting through.
4. Turn off the room lights and have a flash light (or cell phone) ready.
5. Rotate the filter with the rotation stage until about 100 µW are transmitting
through the filter.
136
6. Using the fine-tuning screw of the polarizer, minimize the signal transmitting
through the attenuator. The filter edge for horizontal (p-polarized) light moves
much faster as a function of rotation than vertical (s-polarized) light. Any hori-
zontal component of the laser light will then not be removed by the filter, so the
polarizer must be tuned along with the filter.
7. Iterate between rotating the filter and the polarizer, minimize the light transmitting
through the filter. Be careful not to rotate the filter farther than necessary, as too
much rotation will start to block sidebands you may want to measure.
At this point, it may be useful to ensure quality alignment to the spectrometer and
CCD. If the spectrometer lens never moves, the two irises in the spectrometer path
should get the alignment very close. The best data, however, comes when the laser spot
is aligned to the CCD that day. Be very careful doing this, it is better to assume the
CCD is very fragile. I don’t know how much laser power it can handle, so err on the safe
side and block the laser when in doubt.
1. Put one of the ND20A filters, called “blue” or “white,” in the laser path. In front
of the polarizer is a good place.
2. Set the center wavelength of the spectrometer to the laser wavelength, and open
the spectrometer slits a lot, as large as 1 mm.
3. With the laser blocked and AOM at zero, take a quick (say 0.2 s, no gain, no
binning, fastest readout) picture with the CCD.
137
4. Unblock the laser, keep the AOM at zero, and take another picture. No laser spot
should visible on the screen. If there is light, check to make sure everything is in
place and that the slits are open enough.
5. Set the CCD shutter to “always open” and change CCD to alignment mode. Slowly
increase the AOM setting until a laser spot is visible on the CCD.
6. Using the last two mirrors before the spectrometer, put the laser at the right place
vertically. Traditionally, it has been around 210–250, but that may change.
7. For horizontal alignment, close the slits until the laser is just barely visible. Scan the
laser left or right to find the other edge of the slits. Move the laser to approximately
the center and repeat this process iteratively until the slits are about 50–65 µm wide
and the laser is dead center.
The NIR laser alignment should now be rock solid. The usual culprits responsible for
misaligning the laser to the spectrometer are changing/rotating the long or short pass
filters. Rotating the half-wave plates shouldn’t affect the alignment at all, and tilting the
ITO beam combiner doesn’t seem to, either.
Software
There are two methods to use this camera, a LabVIEW software development kit (SDK)
and a C++ SDK. A set of Python wrappers using the ctypes library were written for the
C++ SDK for easy inclusion into a PyQT GUI in SherwinGroup/EMCCD-psychic. This
138
GUI handles all CCD-based experiments, including photoluminescence, absorbance, and
sideband generation. It is extremely simple to use and is capable of running complex
experiments rather simply, but the code itself is very complex. It is not recommended
to edit the code without a very complete understanding of the software and PyQT in
general.
A.12 Thomas-Keating energy meter
The Thomas-Keating (TK) energy meter is used to measure the absolute energy of an
FEL pulse. A more detailed description can be found in Sect. 3.2.1. The main points to
worry about when using the TK are not to focus the FEL directly onto it and to use AC
coupling on the oscilloscope for accurate measurements. There is a lot of high-frequency
noise on the TK output, averaging a few FEL shots is usually necessary. This thesis used
the peak-to-peak of the output, but other sources have just used the positive or negative
amplitude.
Software
There is software for automatically measuring and calibrating the integrating pyroelectric
detector using the TK is found in SherwinGroup/InstrumentLibrary-yolo.
139
Appendix B
FEL transport details
Bringing the intense light from the FEL to the sample requires transmitting long-wavelength,
therefore fast-diffracting, light from the vault to the optical table of the experiment, tens
of meters and several walls away. The FEL transport system was developed to accomplish
this task without significant attenuation and while maintaining a high quality spatial pro-
file. The FEL transport system is a series of 6”-diameter tubes pumped down to about
200 mtorr to minimize absorption by atmospheric water. It features several output ports
and computer-controlled retractable mirrors and lenses. The lenses are placed with a
regular spacing throughout the transport system with a focal length of 3 m to maintain
a collimated beam for tens of meters for wavelengths between about 60 µm – 1 mm.
Depending on the frequency, some mirrors are curved and others are flat. The lenses are
made of the THz-transparent plastic TPX and the mirrors are coated with a thick layer
of gold to minimize losses due to optics. The particular transport system setup for the
140
FEL output port,TPX window
TPX lens,f = 3 m
Flat mirror,gold
MB1CS:Curved mirror,
goldTPX lens,f = 3 m
THz from FEL
1380C1380DVault
Figure B.1: The setup of the transport system for bringing the FEL to the optical table.The THz light is generated in the vault behind a very thick concrete wall. That lightis brought through a transport system tube into room 1380C. The light reflects off of acurved mirror, named MB1 Curved South (MB1CS). The light is then focused/collimatedby a TPX lens, brought through the wall into room 1380D, reflected off a flat mirror andfocused/collimated again before exiting the transport system through an FEL outputport, made of TPX plastic.
work in this dissertation is shown in Fig. B.1.
141
Appendix C
Cleanroom processing
The cleanroom on campus has a large collection of tools and well-maintained wet benches
available for sample preparation and making/modifying the occasional optical compo-
nent. This appendix contains several recipes and procedures for common processes and
tools. The sample processing recipes are for making epitaxial growth membranes and
performing epitaxial transfer. It is fairly straightforward to modify these recipes to fit the
needs of a given process. For newer or more difficult processes, however, the cleanroom
staff are extremely helpful and can often point you to another user who is experienced in
the process. There are also procedures for depositing high-quality indium-tin oxide and
for using the dicing saw.
142
C.1 Quantum well membrane process
For experiments where the highest terahertz fields are not necessary and where an optical
etalon would be a problem, it is useful to etch a small window in the GaAs substrate
and expose a thin membrane of epitaxially-grown quantum wells. The membrane is not
very robust to high terahertz fields, but, with a thickness on the order of 1 µm, it does
not form a Fabry-Perot cavity for NIR wavelengths.
Instructions
The etching steps of this process can take a long time. If making multiple samples,
put them on different slides because some may etch faster than others. The Physics
Department storeroom sells 1×3” glass slides, and the cleanroom has 2×3” glass slides.
Glass slides are robust to all of the etches except HF. The glass will etch noticeably
during the HF dip, but dip is not long enough to substantially weaken the slide.
• Spin acid-resistant photoresist (PR) like AZ4210 on growth side of the sample for
protection, 3 krpm for 30 seconds, followed by a 60 second, 100 post exposure
bake (even though no exposure). If worried, test that it’s hard enough that it
won’t scratch off and bake longer if necessary.
• Spin the same acid-resistant PR on substrate side with the same recipe.
• Use a real mask, or tape four pieces of foil together to make a 1.5–2 mm square
hole, a make-shift mask. Taping pieces is better than cutting a hole in aluminum
143
foil because the corners will be sharper. Time spent on this step saves much more
time later. Smaller is better, but 1 mm is too small because of sloping etch walls.
• Expose the substrate side of the sample with the preferred mask. The rectangle
needs to be at least 1.5 mm from the edge of the sample, and it’s best if everything
makes right angles.
• Do the post exposure bake and develop according to the PR instructions.
• Put one drop of PR on a glass slide, and lay the sample growth-side down on the
drop. Bake for a while so that the sample is firmly attached to the glass slide.
• Drip some PR around the now-stuck on sample so that the PR covers everything
but the hole in the middle that will be etched. Double check under microscope
for pinholes in the PR, especially the corners and edges. Bake slowly for several
minutes, until PR is dry and rigid. Bake too quickly and the PR will bubble and
expose pinholes. Check under microscope again.
• This is a stopping point. The sample can be exposed to ambient lights and stored.
• Prepare to etch the sample. There are two options here. Use either the dilute
piranha etch, 1:8:8 H2SO4:H2O2:H2Ol, or Garrett Cole’s epi transfer etch, 30:1
H2O2:NH4OH. Dilute piranha is very fast, about 4 µm/min. It is not at all selective,
though. As for any wet etching procedure, always have sample facing down while
etching, this geometry keeps etch by-products from interfering with the etching
process.
144
• Dektak at least twice while etching, especially if using a non-selective etch. The
dilute piranha etch can undercut the PR layer substantially, so only drag the Dektak
needle from high to low to prevent it catching on an overhang. Any post-Dektak
measurement is a stopping point, so long as no AlGaAs with over 80% Al is exposed.
• If using dilute piranha to etch the substrate, then need to switch to a selective etch
with about 50 µm remaining. Garrett Cole’s etch, 30:1 H2O2:NH4OH, works, as
does citric acid and H2O2. Check progress with a microscope.
• Do quick buffered HF dip to remove the high-aluminum-content etch stop layer.
Again, check progress with a microscope. 70% AlGaAs can take about two minutes
to etch through 200 nm, but 73.5% AlGaAs etches much faster in buffered HF.
AlGaAs with less than 60% Al does not etch at all.
• Submerge entire slide in acetone, sample side up. Cover to prevent evaporation and
wait several hours. Do not try to rush things by agitating the solution or pulling
on the sample. The membrane is very fragile and will tear extremely easily
It may be safer for particularly thin epi-layers, such as the 10 nm GaAs QWs grown
by Loren Pfeiffer which are only about 500 nm thick, to have the window region overhang
the glass slide. The sample will be a little more fragile during etching, but removing the
PR should be much easier. When removing the PR, it will dissolve quickly and evenly,
hopefully eliminating any shearing that occurs when the PR is dissolved slowly between
the sample and the glass slide.
145
C.2 Epitaxial transfer process
For experiments where very strong or even enhanced terahertz fields are useful and Fabry-
Perot interference is acceptable, the epitaxial transfer process is extremely useful. This
process was introduced to us by Garrett Cole of Crystalline Mirror Solutions.
Instructions
Before going into the clean room:
• Sketch the wafer geometry, including which direction the flat was. Lapping will
remove all reference points, and the sample may break in two.
• Lap down all samples to 150 microns thick.
• Dice sapphire to appropriate size(s).
Once in the clean room:
• Clean 1165-Acetone-Isopropanol-DI water-dry N2.
• Do an O2 descum, 100 W, 1 minute to finish cleaning the sample.
• Take a full set of pictures of the growth side of the sample and record the size if
any pieces broke off during lapping.
• Heat up on hot plate to 130 deg C to finish drying. Silicon makes a good wafer
carrier for this process because thermal contact is occasionally important.
• Remove the sample and let cool to room temperature.
146
• Put large wax piece on the growth side of the sample, then put the whole wafer
carrier on the hot plate, still at 130 C. The wax takes a few minutes to melt. If
more is needed, remove the wafer to cool so that if the wax falls of the tweezer at
the wrong time, it wont mess up the sample. Take this step slowly as removing
black wax is difficult and very dirty.
• After melting fully, drop the temperature to 100 C in 5 C increments. Once the
temperature gets to 100 C, cover with an aluminum boat and let anneal for an
hour.
• Decrease temperature to 80 C in 5 C increments.
• Lay/balance the silicon carrier wafer or glass slide over the wax. If it is off to one
side, do your best to right it while the wax is still soft. Needs to be > 11/4” to use
with vacuum evaporator.
• Drop the temperature to 50-60 C in 5 C increments, then remove from the hot
plate and let cool completely. The samples should be glued to the wafer carrier
with black wax with the lapped sides exposed
• Prepare 30-100:1 H20:NH4OH and 30:1 H2O2:NH4OH solutions (200:5 and 300:10
mL, respectively, works nicely). We call the latter Garrett Cole’s etch.
• Etch in water-ammonium hydroxide solution for 30 seconds, then immediately place
in peroxide-ammonium hydroxide solution.
147
• Etch until the substrate is removed. It will be obvious as the entire sample will be
very smooth. This takes about 90 minutes for 150 microns. Occasionally a white
scum builds up on the substrate, just swab it off in situ.
• While etching, clean a sapphire substrate on a solvent spinner if possible (500-1000
rpm, but substrate may be too small) following this order: (tergitol+swab)-DI-IPA-
DI, Nomarsky microscopy inspect and pictures, repeat until very clean, no dust at
all. Some dots may remain that (often) get annihilated in the descum. Then 200 W
300 mtorr O2 descum the sapphire wafer, 2 minutes. Double check with Nomarsky
microscopy to ensure cleanliness.
• Once peroxide-ammonium hydroxide etch is complete, take pictures, etch in buffered
HF until the color is very uniform, no interference fringes, 3 minutes. Every 30-45
seconds, rinse in DI water. Then take pictures of the entire sample.
• After HF, check if any weird dust/filaments are there. If so, wash in AZ400K in
a clean beaker, could possibly take 5 minutes. Be patient here, this dust prevents
good Van der Waals contact.
• Add a very small drop of water to the exposed epi, then place the sapphire on top.
It works best if the sapphire is appropriately sized. Be very careful that it doesn’t
slide off at first. Wick away any drops of water that are squeezed out with a clean
wipe. Press firmly with the wooden end of a qtip, and you should see Fabry-Perots
disappear. Wick away as much moisture as possible without touching the sample
148
with the wipe.
• IF IF IF there is an obvious bubble (of water, not air), it is maybe possible to
squeeze the bubble out by covering with a Berkshire wipe and using a qtip like a
rolling pin.
• Dont touch them after that, let the water evaporate. Takes about 2 hours when
pulling rough vacuum, or a day or two in air.
• In aluminum boat, dissolve black wax in chloroform. May take a few fresh chlo-
roform baths to get it all. Make sure to dispose of the used chloroform solution
safely.
• Once no more black wax is visible, IPA-DI-chloroform-IPA-Ace-IPA-DI- gentle blow
dry to clean
There are several improvements that could be made to this process. Using etching to
define the transferred region would be very useful. Dicing introduces far too many micro-
cracks at the edges to be useful, and cleaving is imprecise and can induce microcracks at
the initiating corner. By defining a sample size with a photolithography mask, samples
will be much more uniform and less likely to be severely damaged at the edges. Also, it
may be possible to skip lapping entirely by simply etching the back surface faster. Also,
to ensure an extremely smooth epitaxial bonding interface, more etch stop layers could be
added in growth. A triple layered structure, 73.5% AlGaAs-GaAs-73.5% AlGaAs would
take advantage of the extreme selectivity of HF.
149
C.3 Indium-tin oxide deposition
Indium-tin oxide (ITO) is a terrific material for terahertz-optical experiments. It is highly
conducting below about 10 THz, but is practically lossless all the way up to 450 nm. It
can be grown in the electron-beam vacuum deposition chamber for dielectrics in the
cleanroom, named E-Beam 2. The ITO grown here should have a DC resistivity of about
400 µΩcm and a Im(n) = k ∼ 10−5 at 800 nm. Significant optical absorption should
not exist until about 400 nm. Thanks to John Leonard from the GaN lasers group for
teaching us this recipe.
Before signing up for time, prepare the sample and a test piece of undoped silicon
wafer, about 1 cm square. Silicon has very high index contrast with ITO, nSi ∼ 4 versus
nITO ∼ 1.7, so it can be used to test the optical quality of the layer with the ellipsometer
in the cleanroom, but be sure to ask for help with using that machine, analyzing the
results can be complicated.
Depositing a high-quality film of ITO requires careful control of the wafer tempera-
ture, the O2 pressure, and the deposition rate. All of these details are, again, courtesy of
John Leonard, and meeting these requirements results in films that are extremely smooth
(supposedly better than nanometer roughness, although that has not been measured by
us), highly transparent, and very conducting. The best films grow when the sample
is held at 290 C. Because of the complexities of ITO conduction, which are somehow
related to the oxygen content of the material, the deposition chamber must be pumped
down to about 6 · 10−6 torr, then 30 sccm of O2 is bled into the chamber, resulting in a
150
pressure of about 3 · 10−4 torr. The ITO then needs to deposited ery slowly, 0.01 nm/s
for the first 10 nm, then 0.05 nm/s until 50 total nm, then 0.1 nm/s until the film is
complete. Once the film is deposited, the temperature must be allowed to drop slowly in
the oxygenated atmosphere until it drops below the ITO glass transition temperature of
about 140 C.
C.4 Dicing samples
The dicing saw run by the cleanroom staff in Engineering II is an extremely useful tool
for making samples the exact size desired. Sapphire substrates, which have proven to be
a very effective support material for GaAs epitaxial films, cannot be cleaved in rectangles.
To cut 500 µm c-plane sapphire, we use 2.187-8C-54RU-3 diamond dicing blades. The
width of the cuts are about 300 µm thick. We have also cut 1 mm sapphire wafers and
3 mm thick glass with this blade, but expect very colorful sparks and the blades to wear
very quickly. The cleanroom staff can give specific advice for any given cut or material,
including best practices to protect fragile processing. It is not recommended to dice
samples post-epitaxial-transfer. The UV-release tape bonds stronger to the GaAs than
the sapphire.
151
Bibliography
[1] S. Arlt, U. Siegner, J. Kunde, F. Morier-Genoud, and U. Keller. Ultrafast dephas-ing of continuum transitions in GaAs and AlGaAs. Technical Digest. Summariesof Papers Presented at the Quantum Electronics and Laser Science Conference,59(23):860–863, 1999.
[2] Neil W. Ashcroft and N. David Mermin. Solid State Physics. Brooks/Cole, Belmont,CA, first edition, 1976.
[3] Mary Lou P Bailey, Andrew T Pierce, Aaron J Simon, Devin T Edwards, Gerald JRamian, Nikolay I Agladze, and Mark S Sherwin. Narrow-Band Water-Based Ab-sorber With High Return Loss for Terahertz Spectroscopy. IEEE Transactions onTerahertz Science and Technology, 5(6):961–966, 2015.
[4] Hunter B. Banks, Benjamin Zaks, Fan Yang, Shawn Mack, Arthur C. Gossard,Renbao Liu, and Mark S. Sherwin. Terahertz Electron-Hole Recollisions in GaAsQuantum Wells: Robustness to Scattering by Optical Phonons and Thermal Fluc-tuations. Physical Review Letters, 111(26):267402, dec 2013.
[5] Robert W. Boyd. Nonlinear Optics. Academic Press, San Diego, CA, 1992.
[6] S. G. Carter, V. Ciulin, Mark S. Sherwin, M. Hanson, A. Huntington, L. a. Coldren,and Arthur C. Gossard. Terahertz electro-optic wavelength conversion in GaAs quan-tum wells: Improved efficiency and room-temperature operation. Applied PhysicsLetters, 84(6):840, 2004.
[7] P. Cavalie, J. Freeman, K. Maussang, E. Strupiechonski, G. Xu, R. Colombelli, L. Li,a. G. Davies, E. H. Linfield, J. Tignon, and S. S. Dhillon. High order sideband gener-ation in terahertz quantum cascade lasers. Applied Physics Letters, 102(22):221101,2013.
[8] J. Cerne, J. Kono, Mark S. Sherwin, M. Sundaram, Arthur C. Gossard, andGe Bauer. Terahertz Dynamics of Excitons in GaAs/AlGaAs Quantum Wells. Phys-ical review letters, 77(6):1131–1134, aug 1996.
152
[9] J. Cerne, A. G. Markelz, M. S. Sherwin, S. J. Allen, M. Sundaram, A. C. Gossard,P. C. Van Son, and D. Bimberg. Quenching of excitonic quantum-well photolumi-nescence by intense far-infrared radiation: Free-carrier heating. Physical Review B,51(8):5253–5262, 1995.
[10] E. Y. Chang, Yeong-Lin Lai, Y. S. Lee, and S. H. Chen. A GaAs/AlAs Wet SelectiveEtch Process for the Gate Recess of GaAs Power Metal-Semiconductor Field-EffectTransistors. Journal of The Electrochemical Society, 148(1):G4, 2001.
[11] Ming-Che Chang and Qian Niu. Berry curvature, orbital moment, and effectivequantum theory of electrons in electromagnetic fields. Journal of Physics: CondensedMatter, 20(19):193202, 2008.
[12] Ming-Chang Chen, Christopher Mancuso, Carlos Hernandez-Garcıa, Franklin Dol-lar, Ben Galloway, Dimitar Popmintchev, Pei-Chi Huang, Barry Walker, LuisPlaja, Agnieszka A Jaron-Becker, Andreas Becker, Margaret M Murnane, Henry CKapteyn, and Tenio Popmintchev. Generation of bright isolated attosecond softX-ray pulses driven by multicycle midinfrared lasers. Proceedings of the NationalAcademy of Sciences of the United States of America, 111(23):E2361–7, 2014.
[13] NC Constantinou and BK Ridley. Electron energy relaxation via LO-phonon emis-sion in free-standing GaAs wafers. Journal of Physics: Condensed Matter, 2(7465),1990.
[14] Paul B. Corkum. Plasma Perspective on Strong-Field Multiphoton Ionization. Phys-ical Review Letters, 71(13):1994–1997, 1993.
[15] Paul B. Corkum. Recollision physics. Physics Today, (March), 2011.
[16] J. A. Crosse and Ren Bao Liu. Quantum-coherence-induced second plateau in high-sideband generation. Physical Review B - Condensed Matter and Materials Physics,89(12):1–5, 2014.
[17] John H. Davies. The Physics of Low-dimensional Semiconductors: An Introduction.Cambridge University Press, 1997.
[18] J Feldmann, G Peter, EO Gobel, P. Dawson, K. Moore, C. T. Foxon, and R. J.Elliott. Dependence of Radiative Exciton Lifetimes in Quantum Wells. PhysicalReview Letters, 59(20):2337–2340, 1987.
[19] M Ferray, Anne L’Hullier, XF Li, G. Mainfray, and C. Manus. Multiple-harmonicconversion of 1064 nm radiation in rare gases. Journal of Physics B: Atomic, Molec-ular and Optical Physics, 21, 1988.
[20] P. A. Franken, A. E. Hill, C. W. Peters, and G. Weinreich. Generation of opticalharmonics. Physical Review Letters, 7(4):118–119, 1961.
153
[21] D. Gammon, B. V. Shanabrook, and D. S. Katzer. Excitons, phonons, and interfacesin GaAs/AlAs quantum-well structures. Physical Review Letters, 67(12):1547–1550,1991.
[22] Shambhu Ghimire, Anthony D. DiChiara, Emily Sistrunk, Pierre Agostini, Louis F.DiMauro, and David A. Reis. Observation of high-order harmonic generation in abulk crystal. Nature Physics, 7(2):138–141, 2010.
[23] Dennis H. Goldstein. Polarized Light. CRC Press, Boca Raton, FL, third edition,2011.
[24] Ronald L Greene, Krishan K. Bajaj, and Dwight E. Phelps. Energy levels of Wannierexcitons in GaAs/GaAlAs quantum-well structures. Physical Review B, 29(4):1807,1984.
[25] M. A. Herman, D. Bimberg, and J. Christen. Heterointerfaces in quantum wellsand epitaxial growth processes: Evaluation by luminescence techniques. Journal ofApplied Physics, 70(2), 1991.
[26] FA Ilkov, JE Decker, and SL Chin. Ionization of atoms in the tunnelling regime withexperimental evidence using Hg atoms. Journal of Physics B: Atomic, Molecularand Optical Physics, 25:4005–4020, 1992.
[27] P. U. Jepsen, D. G. Cooke, and M. Koch. Terahertz spectroscopy and imaging -Modern techniques and applications. Laser and Photonics Reviews, 5(1):124–166,2011.
[28] C. J. Joachain, N. J. Kylstra, and R. M. Potvliege. Atoms in Intense Laser Fields.Cambridge University Press, Cambridge, UK, 2012.
[29] R A Kaindl, M A Carnahan, D. Hagele, R. Lovenich, and Daniel S. Chemla. Ultrafastterahertz probes of transient conducting and insulating phases in an electron-holegas. Nature, 423:734–738, 2003.
[30] J. P. Kaminski, J. S. Spector, C. L. Felix, D. P. Enyeart, D. T. White, andG. Ramian. Far-infrared cavity dump coupling of the UC Santa Barbara free-electronlaser. Applied Physics Letters, 57(26):2770–2772, 1990.
[31] D. S. Katzer, D. Gammon, and B. V. Shanabrook. Modification of the microrough-ness of molecular-beam epitaxially grown GaAs/AlAs interfaces through changes inthe growth temperature. Journal of Vacuum Science & Technology B: Microelec-tronics and Nanometer Structures, 10(2):800, mar 1991.
[32] LV Keldysh. Ionization in the field of a strong electromagnetic wave. Sov. Phys.JETP, 20(5):1307, 1965.
154
[33] Dai Sik Kim, Jagdeep Shah, J. E. Cunningham, T. C. Damen, Wilfried Schafer,Michael Hartmann, and Stefan Schmitt-Rink. Giant excitonic resonance in time-resolved four-wave mixing in quantum wells. Physical Review Letters, 68(7):1006–1009, 1992.
[34] J. Kono, M.Y Su, J. Cerne, Mark S. Sherwin, S. J. Allen, Jr., T. Inosh*ta, T Noda,and H. Sakaki. Terahertz dynamics in confined magnetoexcitons. Physica B: Con-densed Matter, 249-251:527–533, jun 1998.
[35] R. F. Kopf, E. F. Schubert, T. D. Harris, and R. S. Becker. Photoluminescence ofGaAs quantum wells grown by molecular beam epitaxy with growth interruptions.Applied Physics Letters, 58(6):631, 1991.
[36] JL Krause, Kenneth Schafer, and KC Kulander. High-order harmonic genera-tion from atoms and ions in the high intensity regime. Physical Review Letters,68(24):3535–3538, 1992.
[37] G. Lambert, T. Hara, D. Garzella, T. Tanikawa, M. Labat, B. Carre, H. Kitamura,T. Shintake, M. Bougeard, S. Inoue, Y. Tanaka, P. Salieres, H. Merdji, O. Chubar,O. Gobert, K. Tahara, and M.-E. Couprie. Injection of harmonics generated in gas ina free-electron laser providing intense and coherent extreme-ultraviolet light. NaturePhysics, 4(4):296–300, 2008.
[38] F. Langer, M. Hohenleutner, C. P. Schmid, C. Poellmann, P. Nagler, T. Korn,C. Schuller, M. S. Sherwin, U. Huttner, J. T. Steiner, S. W. Koch, M. Kira, andR. Huber. Lightwave-driven quasiparticle collisions on a subcycle timescale. Nature,533(7602):225–229, 2016.
[39] M. Lewenstein, Ph. Balcou, Misha Yu Ivanov, Anne L’Huillier, and Paul B. Corkum.Theory of high-harmonic generation by low-frequency laser fields. Physical ReviewA, 49(3):2117, 1994.
[40] Anne L’Huillier and Ph Balcou. High-order harmonic generation in rare gases witha 1-ps 1053-nm laser. Physical Review Letters, 70(6):774–777, 1993.
[41] Renbao Liu and Bang-Fen Zhu. High-order THz-sideband generation in semicon-ductors. AIP Conference Proceedings, 893:1455–1456, 2007.
[42] JM Luttinger and W. Kohn. Motion of Electron and Holes in Perturbed PeriodicFields. Physical Review, 173(4):869, 1955.
[43] JJ Macklin, JD Kmetec, and CL Gordon III. High-order harmonic generation usingintense femtosecond pulses. Physical Review Letters, 70(6):766–769, mar 1993.
[44] S.W. McKnight, K.P. Stewart, H. D. Drew, and K. Moorjani. Wavelength-independent anti-interference coating for the far-infrared. Infrared Physics,27(5):327–333, sep 1987.
155
[45] Eun-A Moon, Jong-Lam Lee, and Hyung Mo Yoo. Selective wet etching of GaAs on[Al xGa 1x]As for AlGaAs/InGaAs/AlGaAs pseudomorphic high electron mobilitytransistor. Journal of Applied Physics, 84(7):3933, 1998.
[46] KB Nordstrom, K Johnsen, S. J. Allen, Jr., AP Jauho, B. Birnir, J. Kono, T. Noda,Hidefumi Akiyama, and H. Sakaki. Excitonic dynamical franz-keldysh effect. Phys-ical Review Letters, 81(2):457–460, 1998.
[47] Andrew T Pierce. Development of a Frequency Domain THz Vector Reflectometer.Bachelor honor’s thesis, University of California, Santa Barbara, 2015.
[48] Gerald Ramian. The new UCSB free-electron lasers. Nuclear Instruments andMethods in Physics Research Section A: Accelerators, Spectrometers, Detectors andAssociated Equipment, 318(1-3):225–229, 1992.
[49] S. Rudin and T. Reinecke. Exciton-acoustic-phonon linewidths in GaAs bulk andquantum wells. Physical Review B, 65(12):9–12, 2002.
[50] K Miyazaki Sakai and H. High-order harmonic generation in rare gases with intensesubpicosecond dye laser pulses. Journal of Physics B: Atomic, Molecular and OpticalPhysics, 25(3):L83–L83, 1992.
[51] O. Schubert, M. Hohenleutner, F. Langer, B. Urbanek, C. Lange, U. Huttner,D. Golde, T. Meier, M. Kira, S. W. Koch, and R. Huber. Sub-cycle control ofterahertz high-harmonic generation by dynamical Bloch oscillations. Nature Pho-tonics, 8(2):119–123, 2014.
[52] S Takahashi, L C Brunel, D T Edwards, J van Tol, G Ramian, S Han, and M SSherwin. Pulsed electron paramagnetic resonance spectroscopy powered by a free-electron laser. Nature, 489(7416):409–413, 2012.
[53] Susumu Takahashi, Gerald Ramian, and Mark S. Sherwin. Cavity dumping of aninjection-locked free-electron laser. Applied Physics Letters, 95(23):2007–2010, 2009.
[54] K L Teo, J S Colton, P Y Yu, E R Weber, and M F Li. An analysis of temper-ature dependent photoluminescence line shapes in InGaN An analysis of tempera-ture dependent photoluminescence line shapes in InGaN. Applied Physics Letters,1697(1998):10–13, 2012.
[55] G Vampa, T J Hammond, N Thire, B E Schmidt, F Legare, C R McDonald,T Brabec, and P B Corkum. Linking high harmonics from gases and solids. Nature,522(7557):462–464, 2015.
[56] G. Vampa, T. J. Hammond, N. Thire, B. E. Schmidt, F. Legare, C. R. McDonald,T. Brabec, D. D. Klug, and P. B. Corkum. All-Optical Reconstruction of CrystalBand Structure. Physical Review Letters, 115(19):1–5, 2015.
156
[57] G. Vampa, C. R. McDonald, G. Orlando, D. D. Klug, P. B. Corkum, and T. Brabec.Theoretical Analysis of High-Harmonic Generation in Solids. Physical Review Let-ters, 113(7):073901, 2014.
[58] Carlo Vicario, Balazs Monoszlai, and Christoph P. Hauri. GV/ m single-cycle tera-hertz fields from a laser-driven large-size partitioned organic crystal. Physical ReviewLetters, 112(21):1–5, 2014.
[59] N. Q. Vinh, S. James Allen, and Kevin W. Plaxco. Dielectric spectroscopy of proteinsas a quantitative experimental test of computational models of their low-frequencyharmonic motions. Journal of the American Chemical Society, 133(23):8942–8947,2011.
[60] Martin Wagner, Harald Schneider, Dominik Stehr, Stephan Winnerl, Aaron M. An-drews, Stephan Schartner, Gottfried Strasser, and Manfred Helm. Observation of theintraexciton Autler-Townes effect in GaAs/AlGaAs semiconductor quantum wells.Physical Review Letters, 105(October):1–4, 2010.
[61] Martin Wagner, Harald Schneider, Stephan Winnerl, Manfred Helm, T. Roch,Aaron M. Andrews, S. Schartner, and Gottfried Strasser. Resonant enhancementof second order sideband generation for intraexcitonic transitions in GaAs/AlGaAsmultiple quantum wells. Applied Physics Letters, 94(24):241105, 2009.
[62] Di Xiao, Ming Che Chang, and Qian Niu. Berry phase effects on electronic proper-ties. Reviews of Modern Physics, 82(3):1959–2007, 2010.
[63] Xiao Tao Xie, Bang Fen Zhu, and Ren Bao Liu. Effects of excitation frequencyon high-order terahertz sideband generation in semiconductors. New Journal ofPhysics, 15, 2013.
[64] E. Yablonovitch, D. M. Hwang, T. J. Gmitter, L. T. Florez, and J. P. Harbison. Vander Waals bonding of GaAs epitaxial liftoff films onto arbitrary substrates. AppliedPhysics Letters, 56(1990):2419–2421, 1990.
[65] Eli Yablonovitch, T. Gmitter, J. P. Harbison, and R. Bhat. Extreme selectivity inthe lift-off of epitaxial GaAs films. Applied Physics Letters, 51(1987):2222–2224,1987.
[66] Jie-Yun Yan. Theory of excitonic high-order sideband generation in semiconductorsunder a strong terahertz field. Physical Review B, 78(7):075204, aug 2008.
[67] Fan Yang and Rb Liu. Nonlinear optical response induced by non-Abelian Berrycurvature in time-reversal-invariant insulators. Physical Review B, 90:245205, 2014.
[68] Fan Yang and Renbao Liu. Berry phases of quantum trajectories of optically excitedelectronhole pairs in semiconductors under strong terahertz fields. New Journal ofPhysics, 15(11):115005, nov 2013.
157
[69] Fan Yang, Xiaodong Xu, and Ren Bao Liu. Giant Faraday rotation induced bythe Berry phase in bilayer graphene under strong terahertz fields. New Journal ofPhysics, 16, 2014.
[70] K. L. Yeh, M. C. Hoffmann, J. Hebling, and Keith A. Nelson. Generation of 10 µJultrashort terahertz pulses by optical rectification. Applied Physics Letters, 90(17):1–3, 2007.
[71] Benjamin Zaks, Hunter B. Banks, and Mark S. Sherwin. High-order sideband gen-eration in bulk GaAs. Applied Physics Letters, 102(1):012104, 2013.
[72] Benjamin Zaks, James Heyman, Dominik Stehr, Dan Allen, Nelson Coates, andMark Sherwin. Single shot high resolution THz upconversion spectrometer. 33rdInternational Conference on Infrared and Millimeter Waves and the 16th Interna-tional Conference on Terahertz Electronics, 2008, IRMMW-THz 2008, pages 1–2,2008.
[73] Benjamin Zaks, Renbao Liu, and Mark S. Sherwin. Experimental observation ofelectron-hole recollisions. Nature, 483(7391):580–3, mar 2012.
[74] Benjamin Zaks, Dominik Stehr, Tuan-Anh Truong, Pierre M Petroff, StephenHughes, and Mark S. Sherwin. THz-driven quantum wells: Coulomb interac-tions and Stark shifts in the ultrastrong coupling regime. New Journal of Physics,13(8):083009, aug 2011.
[75] Ph. Zeitoun, G. Faivre, S. Sebban, T. Mocek, A. Hallou, M. Fajardo, D. Aubert,Ph. Balcou, F. Burgy, D. Douillet, S. Kazamias, G. de Lache‘ze-Murel, T. Lefrou,S. le Pape, P. Merce‘re, H. Merdji, A. S. Morlens, J. P. Rousseau, and C. Valentin.A high-intensity highly coherent soft X-ray femtosecond laser seeded by a high har-monic beam. Nature, 431(1):426, 2004.
158