Plasmon enhanced photoelectron spectroscopy and the generation of isolated attosecond XUV pulses for use with condensed matter targets

Transcription

1 Plasmon enhanced photoelectron spectroscopy and the generation of isolated attosecond XUV pulses for use with condensed matter targets by Phillip Michael Nagel A dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy in Chemistry in the Graduate Division of the University of California, Berkeley Committee in charge: Professor Stephen R. Leone, Chair Professor Daniel M. Neumark Professor Roger W. Falcone Fall 2011

2 Plasmon enhanced photoelectron spectroscopy and the generation of isolated attosecond XUV pulses for use with condensed matter targets Copyright 2011 by Phillip Michael Nagel

3 1 Abstract Plasmon enhanced photoelectron spectroscopy and the generation of isolated attosecond XUV pulses for use with condensed matter targets by Phillip Michael Nagel Doctor of Philosophy in Chemistry University of California, Berkeley Professor Stephen R. Leone, Chair Surface plasmon resonances (SPRs), collective oscillations of quasi-free electrons in metals, can produce strong electric field enhancements at the surface of nanoparticles. These oscillations typically occur at optical frequencies (thus having a period on the order of one to a few femtoseconds) and only remain coherent for a few to tens of femtoseconds. Because of their increasing importance in various applications, it is important to understand SPRs at a fundamental level. The ultrafast nature of the processes involved with SPRs make time-resolved spectroscopy an important tool for probing their dynamics. Recently developed light sources capable of producing isolated attosecond (10 18 s) pulses of light can provide snapshots of electron dynamics on a sub-femtosecond timescale. Fewer than a dozen laboratories in the world currently have the ability to produce such light pulses. In this dissertation I discuss the development and construction of an experimental apparatus capable of producing and utilizing isolated attosecond pulses to study condensed matter, including surface plasmon dynamics. The ultimate goal of the experiments presented here is to laser-excite plasmonic resonances in metallic nanostructures and to detect the field enhancement at the surface of the nanostructures by measuring photoelectron spectra. In the first experiment presented, electron photoemission from lithographically prepared gold nanopillars using nominally few-cycle, 800 nm laser pulses is described. Electron kinetic energies are observed that are higher by up to tens of ev compared to photoemission from a flat gold surface at the same laser intensities. A classical electron acceleration model consisting of multiphoton ionization followed by field acceleration qualitatively reproduces the electron kinetic energy data and suggests average enhanced electric fields due to the nanopillars that are between 25 and 39 times greater than the experimentally used laser fields. In the second experiment presented, attosecond streaking from a W(110) single crystal and from an amorphous Cr thin film is demonstrated. In addition, a novel concept for SPR enhanced attosecond streaking is proposed and evaluated with the aid of a numerical model.

4 To Aisling, for always being by my side. i

5 ii Contents List of Figures iv 1 Introduction Overview Attosecond Spectroscopy History and mechanisms of high-harmonic generation Attosecond pulse trains from HHG Isolated attosecond pulses from HHG Experiments using isolated attosecond pulses Surface plasmon resonance Definition of the surface plasmon resonance Propagating and localized surface plasmons Plasmon lifetimes Overview of photoelectron spectroscopy from surface plasmon systems Summary Experimental Apparatus Overview Laser System Oscillator Amplifier Prism Compressor Carrier-Envelope Phase Stabilization f-2f Mach-Zehnder Interferometer CEP Locking Electronics Spectral Broadening in Hollow-Core Fiber Chirped Mirror Compressor Vacuum System High Harmonic Generation Metal filters XUV Optics

6 iii 2.10 Time-of-Flight Electron Spectrometry TOF Spectrometer Signal analysis Sample TOF data XUV Spectrometer Surface plasmon electron acceleration Introduction Experimental Apparatus Nanopillar Sample Results and Discussion Photoelectron Spectra Total Electron Emission Scaling with Laser Intensity Classical acceleration model Conclusions Condensed Matter Attosecond Streaking Overview Previous experiments in the literature Streaking results from a W(110) single crystal Photoelectron background emission Demonstration of attosecond streaking Streaking results from amorphous Cr thin film Surface plasmon enhanced attosecond spectroscopy Conclusions Bibliography 75

7 iv List of Figures 1.1 Illustration of the three step model for high-harmonic generation put forth by Corkum [16]. Step 1: The intense laser field is strong enough to distort the binding potential of the atom s valence electron and allow tunnel ionization. Step 2: The free electron is accelerated in the laser field back toward the parent ion. Step 3: Recombination of the free electron with the parent ion releases a photon with energy up to the cutoff at I p U p A 5 fs FWHM, 800 nm Gaussian laser pulse with CEP = -π/2. The CEP is defined as the phase difference between the peak of the Gaussian pulse envelope and the maximum of the carrier wave Demonstration of the carrier-envelope offset frequency (CEO) taken from Ref. [25]. The CEO results from the gap between zero frequency (ν = 0) and the lowest peak (ν CEO )when the laser frequency comb is extended to zero frequency. f rep is the inverse of the repetition rate of the laser. The absolute frequency of any spectral line, ν(m) can be determined by ν(m) = ν CEO +mf rep Comparison between a 5 fs FWHM, 800 nm Gaussian laser pulse and a 25 fs FWHM pulse. The difference in electric field strength of adjacent half cycles of the laser pulse becomes much less for longer pulse durations and causes the value of the CEP to lose effect Schematic of the effect of CEP on the ability to isolate a single attosecond pulse. a) A cosine pulse (CEP = 0) creates a situation in which one half cycle of the driving field is of higher intensity than all other half cycles. b) A sine pulse (CEP = π/2) where there are two half-cycles of equivalent intensity. When the spectral filter is applied an isolated attosecond pulse can be obtained from the cosine pulse but not from the sine pulse Principle of attosecond streaking taken from Ref. [36]. Electrons are released by the near-intantaneous attosecond pulse and receive an ultrafast sub-cycle modulation of their momentum from the few-cycle streak field A typical streak trace taken in sulfur hexafluoride gas. The streaking of electrons ionized from the outer valence orbitals appears around 76 ev while the faint signal of inner valence streaking can be seen around 50 ev

8 v 1.8 Diagram of the plasmon dispersion relation for bulk plasmons and surface plasmons, modified from Ref. [54]. ω p is the bulk plasmon frequency, ω s is the surface plasmon frequency, and the light line is k = ω/c where ω is the angular frequency of light and c is the speed of light in vacuum. Coupling of the light line to the plasmon modes via momentum transfer by surface roughness, k r, is shown for both plasmon modes A schematic of the Kretschmann configuration for exciting propagating surface plasmon waves in a flat metal surface. Total internal reflection of the laser beam at the ɛ 0 /ɛ 1 interface launches an evanescent wave which can excite a SPR at the ɛ 1 /ɛ 2 interface Schematic depicting localized SPRs oscillating in gold nanospheres. In a localized plasmon the electron cloud oscillates back and forth across the particle, concentrating the incident electromagnetic energy into a small physical volume. In the schematic the incident field is traveling in the direction of the arrow and driving the plasmon oscillations in the particles. The shaded area represents the electron density as the particles are macroscopically polarized An example of persistent spectral hole burning from Ref [52]. As the oblate gold-nanoparticles on a sapphire substrate are irradiated, an increasingly large hole forms in the measured spectrum. From this hole the homogenous plasmon linewidth can be determined. In this case the homogenous linewidth is 94 mev, which corresponds to a dephasing time of 14 fs An example of second-harmonic generation from gold nanoparticles in Reference [48]. The bold line is the autocorrelation measured using a standard BBO crystal and representing the laser pulse duration. The thin line is the autocorrelation measured with second-harmonic light generated from a gold nanostructured surface. The broadening is from the plasmon dephasing lifetime and corresponds to a lifetime of 6 ± 1 fs Surface plasmon-based electron acceleration demonstrated by Irvine, et al. [60]. In this experiment a 27 fs laser oscillator pulse is used to both excite a SPR in a gold film in the Kretschmann geometry and ionize photoelectrons into the enhanced field. The electrons are then classically accelerated to high kinetic energies An overview of the experimental apparatus Schematic of the Femtolasers Femtosource Scientific Pro oscillator. PL - pump laser, L - lens, Ti:S - titanium sapphire crystal, CM - chirped mirror, WP - wedge pair, OC - output coupler, CP - compensating plate, BS1-50:50 beamsplitter, BS2-30:70 beamsplitter, FI - Faraday isolator, TOD mirrors - chirped mirrors for third order disperson (TOD) compensation, GB - 10cm long SF10 glass block, RR - retro-reflector, PO - pick-off mirror, PD - photodiode. Modified from Ref. [70]

9 vi 2.3 Schematic of the Femtolasers Femtopower Compact Pro amplifier. L1,2 - telescope for pump beam, L3 - lens for focusing of pump beam, PBFM - pump beam focusing mirror, P1-4 - periscopes, IRFM1,2 - infrared focusing mirrors, RR1,2 - retro-reflectors, PBS1,2 - polarizing beam-splitters, PC - Pockel s cell, BC - Berek polarization compensator, PO1,2 - pick-off mirror, VC - vacuum chamber, BW - Brewster window, Ti:S - titanium sapphire crystal, C - Peltier cooling, PD - photodiode. Courtesy of Ref. [70] Schematic of the Femtolasers Femtopower Compact Pro prism compressor. Courtesy of Ref. [70] Spectral fringes measured in the slow loop. The first 8 minutes show spectra collected while varying the offset voltage that is sent to the phase locking electrons. The changing voltage varies the CEP and thus the positions of the spectral fringes. From 8 minutes onward, the offset voltage is only controlled by the slow loop error signal and long-term locking stability of the CEP is demonstrated. Inset: Fast loop beat signal measured on a spectrum analyzer A schematic of the Mach-Zehnder interferometer used in the CEP fast loop. MO - microscope objective, PCF - photonic crystal fiber, L - lens, DC - dichroic mirror, P - polarizer, BBO - beta barium borate crystal, BPF - band pass filter, PD - photo diode A schematic of the beam stabilization system for input into the hollow-core fiber (not drawn to scale). PZ1, PZ2 - Piezo actuated mirrors, L - focusing lens, QPD1, QPD2 - Quadrant photodiodes, BS - Beamsplitter, HCF - Hollowcore fiber Laser pulse spectrum measured after spectral broadening in a hollow-core fiber filled with 1.9 Bar of Ne gas but before temporal recompression A measured autocorrelation trace of the few-cycle laser beam after spectral broadening in the Ne filled hollow-core fiber and temporal recompression in the chirped mirror compressor. The black circles are experimentally measured points while the red line is the calculated fit to the measured data. The corresponding FWHM laser pulse duration is 6.5 fs. The deviation from the fit in the wings of the pulse results from higher order phase terms that manifest as pre- or post-pulses. These are not well characterized by autocorrelation but could be characterized using a pulse-reconstruction technique such as SPIDER or FROG Schematic of the vacuum chamber Typical HHG cutoff spectrum generated by an 800 nm laser pulse in Ne gas. This spectrum was generated by laser pulses with an unlocked carrier-envelope phase

10 2.12 Generation of a continuum in the HHG cutoff region by using a CEP locked 800 nm laser pulse. The two spectra are taken at zero and π relative CEP. The continuum at π relative phase is indicative of an isolated attosecond pulse in that energy region Calculated transmission of a 200 nm thick Zr foil filter. For use with the XUV spectrometer, two such filters are used in-line nm thick Zr filter mounted on an aluminum coated pellicle. This filter serves to separate the copropagating XUV and IR light into an inner beam and an outer beam so that a time delay can be introduced between the two Multilayer Mo/Si XUV mirror reflectivity shown compared to the HHG continuum. The mirror reflectivity is designed to spectrally select only the continuum region of HHG, leaving an isolated attosecond pulse after reflection Schematic of the cored-mirror used in this apparatus. The central portion of the mirror has a multilayer XUV coating and can be moved in the beam axis independently from the outer mirror by a piezo-translation stage. The outer mirror (not shown here for clarity) is gold coated and can be moved in the x- and y-directions by picomotors to allow for precise spatial overlap between the inner and outer beams Circuit diagram of the MCP apparatus used in TOF detection. Provided courtesy of Jordan TOF Products, Inc Kinetic energy resolution of the TOF electron spectrometer as a function of electron kinetic energy. For valence electrons emitted directly by the HHG produced XUV pulse, around 90 ev, the energy resolution is 0.9 ev. The plot only accounts for the instrument resolution and does not include the bandwidth of the XUV pulse or any other contributions to the final experimental resolution A sample TOF photoelectron spectrum collected from ionization of a gold nanopillar sample by the few-cycle IR laser and integrated over laser pulses Raw data for a TOF photoelectron spectrum collected from ionization of a W(110) crystal by HHG produced 93 ev photons and integrated over laser pulses. The peak at 104 ns is caused by scattered photons and can be used for calibrating the spectrometer time zero. To the right of the photon peak a sharp peak from XUV emitted photoelectrons can be seen, followed by a large low energy electron background The TOF data from Figure 2.20 after conversion to a kinetic energy scale and correction for the Jacobian. The peak centered at 86 ev is from XUVinduced valence band photoemission while the large low energy signal is the result of inelastically scattered electrons within the metal Schematic of the XUV spectrometer vii

11 viii 2.23 Measured zero order transmission of the HHG radiation generated in Ne gas, averaged over 10 scans with 1 s integration each scan. 1/e 2 diameters of the x- and y-cross sections are measured as 6.86 mm and 5.4 mm, respectively. The white dashed lines show the positions at which the line-outs were measured First order dispersion spectrum of HHG XUV radiation generated in Ne gas, averaged over 10 scans with 1 s integration each scan Schematic of the experimental apparatus. 30 fs FWHM, 800 µj laser pulses are spectrally broadened in a gas-filled hollow-core fiber (HCF) and temporally compressed to 7 fs FWHM with a series of multilayer chirped mirrors (CM). The laser is focused onto the sample surface and photoelectrons are detected using a linear time-of-flight spectrometer (TOF) (a) Scanning electron microscope (SEM) image of the gold nanopillar array. (b) Dark-field scattering measurement of a single nanopillar from an identically prepared sample with a larger pitch to allow for measurement of a single particle (a) Photoelectron kinetic energy spectra taken from a flat gold surface as a function of excitation intensity. (b) Photoelectron spectra taken from the gold nanopillars at the same intensities as (a). Strong acceleration of photoelectrons to high kinetic energies is indicative of photoelectron emission in the presence of plasmon-enhanced electric fields. Because of the inability of photons to directly excite a SPR in flat gold, a minimal increase in kinetic energy is present in (a) Log-log plot showing the total number of detected photoelectrons as a function of excitation intensity, I, for (a) the flat gold surface and (b) the gold nanopillar sample. While the flat gold surface demonstrates the expected third order multiphoton dependence, only a second-order dependence is observed in emission from the nanopillars (a) Spectra modeled from classical electron trajectory calculations (black lines) compared to the experimental data (symbols). Each trace is offset by one order of magnitude from the previous trace for clarity. In the model, multiphoton emission is followed by classical acceleration in an enhanced field. An average field enhancement of 32 brings the model in close agreement with the experimental data. The intensities shown in the legend are the enhanced intensity values, (I E 2 ), used for the calculation. The experimental data is the same as shown in Figure 3.3a. (b) The experimental data (symbols) compared to a range of modeled spectra calculated for average field enhancement factors from (shaded areas). Each trace and shaded area is offset by two orders of magnitude from the previous trace for clarity

12 ix 4.1 (a) Photoelectron spectra collected at different time delays between the XUV attosecond pulses and the IR laser pulses. The positions of the delays are shown as dashed lines in (b). Peaks from the tungsten valence band (83 ev) and 4f state (56 ev) can be seen in a spectrum taken far from zero time delay (blue line), and the Fermi level is denoted by E f. This same spectrum is also shown after subtraction of the large multiphoton background emission and numerical smoothing (red line). These peaks broaden out from streaking by the IR laser field in the spectrum measured at zero time-delay (black line). (b) The full streaking spectrogram after subtraction of the multiphoton background emission Streaking data from a W(110) crystal from Ref. [40]. (a) Streak traces following cubic-spline interpolation of photoemission from the valence band (upper panel) and the 4f state (lower panel) from a W(110) single crystal. A very small time delay between the two streak traces is highlighted by the dashed white lines. (b) Center-of-mass plots for the valence and 4f streak traces in (a). The resulting delay is 110 ± 70 as, where the error results from the calculation of the center-of-mass The universal curve for electron inelastic mean free path (IMFP) taken from Ref. [81]. For electrons with kinetic energy around 90 ev, the IMFP is only 0.4 nm. Because this length is shorter than the penetration depth of XUV radiation into the sample, a background of inelastically scattered electrons results A schematic illustrating the source of the large inelastically scattered photoelectron background resulting from XUV photoemission. The 93 ev light penetrates 3 nm normal to the surface (z-axis), releasing electrons all along the path of the light, while the IMFP is only 0.4 nm for 90 ev electron kinetic energy. Electrons released deeper than this have very little probability of escaping the surface without inelastically scattering

13 x 4.5 (a) Photoelectron kinetic energy spectrum from a W(110) single crystal surface ionized by HHG generated XUV pulses centered at 93 ev. The peak centered at 86 ev is from electrons that escape from the surface without scattering (indicated approximately by grey shaded area), while the large low energy background results from electrons that are inelastically scattered within the metal. (b) Comparison of XUV only photoelectron emission from W(110) (black line) to photoemission from the XUV plus the few-cycle, 800 nm laser pulses (red line). Instability in the HHG flux has led to a slight decrease in overall signal between the two measurements. The two pulses are positioned at zero time overlap and the intensity of the 800 nm laser pulses is typical for a streaking experiment. The broad peak centered at 34 ev is the result of multiphoton emission by the 800 nm laser pulses and is saturating the detector below 34 ev (indicated by the arrow). In addition, increased amplitude above 95 ev shows streaking of electrons from the valence band peak to higher kinetic energies First demonstration of attosecond streaking from the apparatus developed in this dissertation. The spectrogram is constructed from a series of photoelectron spectra collected from a W(110) single crystal at varying time delays between the attosecond XUV and the IR laser pulses. Negative time delays represent the time at which the attosecond XUV pulses arrive before the IR laser pulses. Each time step (200 as) is integrated over laser pulses Spectral centroid analysis of the streaking trace presented in Figure 4.6. The centroid is calculated from data between 70 ev and 110 ev and clearly demonstrates the sub-optical-cycle resolution of the streaking spectrogram Comparison of attosecond streaking from (a) a W(110) single crystal and (b) a 10 nm thick amorphous chromium thin film. The 800 nm streak field intensity was not the same in both measurements, which accounts for the different amounts of streaking in the kinetic energy domain Summation of the photoelectron yield between 93 ev and 100 ev for (a) a W(110) single crystal surface and (b) a 10 nm thick amorphous chromium thin film. Sub-optical-cycle resolution is clearly visible in both spectrograms. 71

14 4.10 Simulation of a proposed measurement of field-induced attosecond time-resolved dipole potentials. a) Schematic of the dipolar charge distribution induced in a metal nanoparticle by the exciting laser field. Electrons are freed by the attosecond pulse with an initial velocity v 0 and sample the induced electric field, experiencing a force F surface, before they are detected by a time-of-flight spectrometer with a small collection angle of 15. b) Simulated time-dependent photoelectron kinetic energy spectrum as a function of time-delay between the attosecond and the 530-nm laser pulse, where E 0 = v 2 0/2 is the kinetic energy in the absence of the plasmon excitation. A 6 fs laser pulse (white line) at W/cm 2 intensity excites the plasmon which is then probed by a time-delayed 500 as pulse. A temporal broadening of the dipole potential response function mapped out by the intensity maxima of the photoelectron spectral distributions compared to the driving pulse shows the finite lifetime of the plasmon resonance (sustained dipole oscillations at late times after the driving laser pulse is over). In addition to resolving the decay time of the plasmon resonance, individual plasmon oscillations are observed with sub-cycle resolution, permitting the possibility to unravel nonlinear dynamics. 73 xi

15 xii Acknowledgments After six years of grad school I can t count the number of people who have touched my life in ways that I will never forget. Thank you first to my amazing and beautiful very-soon-to-be wife Aisling. She has been there beside me in good times and bad, helping me celebrate or keeping me going. Her faith in me is what led me to complete this work. I owe her all of my love. I also especially want to thank my family for all the support that they have given me, not only in my graduate work, but throughout my entire life. They gave me the opportunity to be where I am today, and I hope that I ve made them proud. Thank you to my advisor, Steve Leone, for all of the opportunities he has provided me over the years. From him I learned the importance of being a rigorous scientist and to never stop asking questions of myself. I also never once had to worry about getting my paycheck on time or about not having enough funding to fix my broken laser, and there is a lot to be said for those things. Dan Neumark, an unofficial advisor to me in many ways, has provided me with a lot of guidance and a few good days of skiing as well. Robert Kaindl has been kind enough to let me collaborate with his group and work in his lab for a number of years now. I owe him thanks for giving me space to work and for providing expertise in condensed matter materials, something that none of the rest of us were very familiar with. I would also really like to thank my undergraduate research advisor, Alex Kandel at the University of Notre Dame. Without the opportunities that he gave me, I would never have known about the joys of working in a laboratory and I certainly would not have ended up here. I am greatly indebted to my two biggest influences in all things related to the laboratory, Thomas Pfeifer and Joe Robinson. I m pretty sure that no postdocs in history have ever been more patient with stupid questions from grad students or more willing to take time out of their schedule to help with whatever I needed. It has been an honor (honour for Joe) and a pleasure to work with both of them. I have had many great friends that have always supported me and have made my life quite enjoyable. Ben Doughty and I started at Berkeley together and were officemates for over four years. I can t think of a better person to have spent so much time with. Our discovery of Lucky House and Pad Ke Mao (well done, of course) was probably the discovery of the century at Berkeley. Mark Abel and I learned about attoseconds together, but we also managed to drink (and make) some great beer and listen to some great music. I also have to thank my roommate, Noah Bell, for putting up with me for all of these years and for being a great roommate. Thanks also to my newest officemates, Justine Bell and Annelise Beck. You guys may listen to some insane music, but we do laugh an awful lot and that makes it worth coming to work every day. I would like to thank Adam Bradford and Kathleen Fowler for their amazing help in navigating the absurd bureaucracies that are LBL and UC Berkeley. Without them I d probably still be waiting on laboratory equipment to arrive from the vendors. Lastly, I would like to thank Professor Roger Falcone for taking time out of his very busy schedule to serve on my thesis committee and help me finish what I came here to do.

16 1 Chapter 1 Introduction 1.1 Overview Surface plasmon resonances (SPRs), collective oscillations of quasi-free electrons in metals, have received a significant amount of attention over the last half century [1, 2, 3, 4]. These oscillations typically occur at optical frequencies (thus having a period on the order of one to a few femtoseconds) and only remain coherent for a few to tens of femtoseconds [5]. SPRs have enormous potential for applications in medicine, communications, and electronics [6, 7]. Most of these applications take advantage of the strongly enhanced electric field created by the plasmon at the metal surface. Techniques such as surface-enhanced Raman spectroscopy (SERS) exploit this near field enhancement and even enable spectroscopic detection of single molecules [8]. Because of the increasing importance of this phenomenon, it is important to understand at a fundamental level. The ultrafast nature of the processes involved with SPRs make time-resolved spectroscopy an important tool for probing SPR dynamics. Time-resolved spectroscopy, the use of short bursts of light to illuminate processes as a function of time, has been used to shed new light on scientific questions since the invention of flash lamps and eventually pulsed lasers [9]. From milliseconds to picoseconds, new regimes of study have been opened up with every new generation of technology. The 1980s and 1990s saw the widespread growth of femtosecond (10 15 s) laser systems. With this technology came unprecedented access to many previously mysterious ultrafast processes, including an isomerization that triggers the first steps of human vision in less than 200 fs [10] and the dynamics of molecules and transition states during chemical reactions [11], for which Ahmed Zewail won the Nobel prize in The turn of this century saw the development of the very first light sources to produce isolated attosecond (10 18 s) pulses [12]. The attsecond timescale is the natural timescale of electron motion inside of atoms and molecules; for instance, the time for an electron to complete one orbit of a Bohr hydrogen atom is 150 as [13]. Electron dynamics are the driving force behind much of chemistry and these new light sources offer an opportunity to understand the basic processes of the world around us at the

17 CHAPTER 1. INTRODUCTION 2 most detailed level yet possible. From gas phase to condensed matter, important processes not accessible through the frequency domain are now being uncovered with unprecedented temporal resolution. By ionizing electrons with isolated attosecond XUV pulses in the presence of enhanced SPR fields, potential nonlinearities in the oscillation of the plasmon electron cloud may be uncovered with sub-cycle resolution. With sub-cycle resolution the mechanisms that lead to the decoherence of the plasmon oscillation can be investigated more clearly. Of particular interest is what happens when various molecules are adsorbed onto the surface of the particle. Through a process known as chemical interface damping, electrons from the metal can tunnel into and out of surface states created by the adsorbed molecules [14]. When electrons that have been trapped in surface states re-enter the metal particle, they do not necessarily do so at the same phase as the plasmon electrons. This addition of this dephasing channel to the dephasing channels of the bare particle decreases the overall plasmon lifetime. If the plasmon field is monitored with sub-cycle resolution using attosecond pulses, the loss of electrons to surface states might be observed directly. Additionally, the amount of enhancement that can be achieved in a technique such as SERS has been shown to be proportional to the plasmon coherence lifetime [15]. It is thus of great interest to understand what controls this lifetime so that substrates for SERS can be designed to achieve maximum enhancement. These are just a few of the reasons that surface plasmon resonances present a highly interesting target for study with isolated attosecond pulses. The ultimate goal of probing plasmon resonances using isolated attosecond pulses could provide new insights into the basic physical properties of SPRs such as the dephasing rate of the oscillating electrons or the effects of surface conditions on the plasmon electron population. The overview goal of the experiments presented here is to laser-excite plasmonic resonances in metallic nanostructures and to detect the field enhancement at the surface of the nanostructures by electron ejection. Three kinds of experiments are anticipated: visible laser electron ejection by multiphoton processes, attosecond streaking at the surface of metals using an extreme ultraviolet (XUV) attosecond pulse coupled with a visible laser field, and pump-probe experiments with visible laser excitation of the plasmon and attosecond electron ejection. Results for the first two goals are achieved in this dissertation. The work consists primarily of two parts: the construction of an apparatus capable of producing and utilizing isolated attosecond light pulses in the XUV regime and the subsequent application of this apparatus to the study of nanostructured plasmonic systems and metallic surfaces. 1.2 Attosecond Spectroscopy History and mechanisms of high-harmonic generation In order to understand attosecond pulse generation, one must first understand the process of high-harmonic generation (HHG). HHG in the gas phase, its most commonly used form

18 CHAPTER 1. INTRODUCTION 3 1 E x1 2 x x 3 t field-/tunnel-ionization 2 hν 3 E c ~ I p U p acceleration in laser field ATI or recombination and photo-emission Figure 1.1: Illustration of the three step model for high-harmonic generation put forth by Corkum [16]. Step 1: The intense laser field is strong enough to distort the binding potential of the atom s valence electron and allow tunnel ionization. Step 2: The free electron is accelerated in the laser field back toward the parent ion. Step 3: Recombination of the free electron with the parent ion releases a photon with energy up to the cutoff at I p U p. today, was first pioneered in 1988 [17] and is a relatively simple process in principle. By focusing an intense femtosecond laser pulse into a gas medium, a highly nonlinear process can emit higher energy photons at multiples of the input photon energy. The well-known three-step model describing the HHG process semi-classically was introduced by Corkum in 1993 [16]. It is shown schematically in Figure 1.1. In the first step a strong laser electric field distorts the binding potential of the valence electrons in the atoms in the gas. The potential is so distorted that electrons can tunnel through the ionization barrier and become free electrons in the continuum. In the second step the free electron is accelerated first away from the parent ion, but then back towards it as the sinusoidal laser field changes direction. As the electron is being accelerated through the continuum it can gain a large amount of kinetic energy. The third step is the recombination of the free electron with the parent ion. This final step does not always occur, and when it does a burst of higher energy radiation can be released by the recombined atom. The photon energy is determined by the binding energy of the electron that was tunnel ionized and the amount of energy it gained in the laser field: ω max = I P U p (1.1)

19 CHAPTER 1. INTRODUCTION 4 where I p is the ionization potential of the harmonic generation medium and U p is the ponderomotive potential, defined as: U p = e2 E 2 o 4m e ω 2 0 (1.2) where e is the elementary electric charge, E 0 is the laser electric field strength in V/m, m e is the electron mass, and ω 0 is the angular frequency of the laser electric field. The HHG process has also been described in a fully-quantum-mechanical model [18]. While the three-step model describes the single atom response of HHG, emission from single atoms can build up coherently to form a macroscopic response under the correct phase matching conditions [19]. Proper phase matching is the condition such that the radiation emitted by individual atoms or molecules adds coherently to the radiation emitted by other atoms or molecules in the ensemble. If appropriate phase matching conditions are not met, the emission from individual atoms will destructively interfere and macroscopic HHG radiation will not be produced. Phase matching in HHG depends on a number of experimental parameters, including the gas density, the laser intensity and the free-electron density [20, 21] Attosecond pulse trains from HHG The discovery of HHG almost immediately led to the speculation that the sharp spectral spikes may be pulses of attosecond duration in the time domain [22]. In principle, this is easy to think about. The HHG spectrum is a series of equally spaced lines that when Fourier transformed, if they are phase-locked with respect to each other, will result in a series of pulses with attosecond time structure in the time-domain [23]. This is also easy to think about in terms of the three-step model, in which the recombination event occurs once every half cycle of the driving laser pulse and produces a burst of high energy radiation every half cycle. Because the recombination of the free electron and the parent ion occurs on an attosecond timescale, the burst of radiation has an attosecond temporal structure. The biggest question after the successful generation of HHG but before the realization of attosecond pulses was whether the harmonic orders were in fact phase-locked to each other. This was conclusively shown by Paul, et al. in 2001 when they were able to measure the relative phase of harmonic orders and determine that the time structure of their HHG emission consisted of attosecond bursts of light spaced by one half of an optical cycle [24]. In order to do this they used a quantum interference technique, which came to be known as RABITT, or reconstruction of attosecond beating by interference of two-photon transitions. The idea is again quite simple. Normally photoionization of an atomic gas such as argon by the HHG pulse would produce a photoelectron spectrum with peaks at the harmonic frequencies minus the ionization potential. By overlapping the HHG pulse with the driving few-cycle laser field, additional quantum pathways (the subtraction or addition of a laser photon) become possible. Because the harmonic peaks are evenly spaced at odd harmonic orders, the addition of a laser photon to one harmonic order creates a sideband that overlaps

20 CHAPTER 1. INTRODUCTION 5 1 CEP Normalized Electric Field Time (fs) Figure 1.2: A 5 fs FWHM, 800 nm Gaussian laser pulse with CEP = -π/2. The CEP is defined as the phase difference between the peak of the Gaussian pulse envelope and the maximum of the carrier wave. with the sideband from subtraction of a laser photon from the next higher energy harmonic order. By measuring the intensity of the sidebands as function of the delay between the HHG pulse and the few-cycle laser pulse, the relative phase of neighboring harmonic orders can be deduced and the temporal structure of the pulse determined. The result in this case was a train of attosecond bursts of radiation with an average duration of 250 as and the first ever measurement of a sub-fs light pulse Isolated attosecond pulses from HHG The carrier-envelope phase The biggest factor in spanning the bridge from attosecond pulse trains to isolated attosecond pulses is control of the carrier-envelope phase (CEP). The CEP is defined as the phase difference between the peak of the laser carrier wave (the oscillating electric field) and the peak of the laser pulse envelope [25], assumed to be approximately Gaussian in the case of the laser used in this dissertation. A diagram of the CEP principle is showing in Figure 1.2. The CEP results from a frequency domain property called the carrier-envelope offset (CEO) frequency, which can be described as follows. When a train of laser pulses in the time domain, such as that produced by a mode locked laser, is Fourier transformed to the frequency domain, it appears as sharp, discrete frequency peaks separated by the repetition rate of the laser cavity (the inverse of the roundtrip cavity time). This series of sharp peaks is called a frequency comb [25]. If this frequency comb is extrapolated to zero frequency,

21 CHAPTER 1. INTRODUCTION 6 Figure 1.3: Demonstration of the carrier-envelope offset frequency (CEO) taken from Ref. [25]. The CEO results from the gap between zero frequency (ν = 0) and the lowest peak (ν CEO )when the laser frequency comb is extended to zero frequency. f rep is the inverse of the repetition rate of the laser. The absolute frequency of any spectral line, ν(m) can be determined by ν(m) = ν CEO + mf rep. there exists an offset between zero frequency and the lowest frequency peak of the comb. This offset is the CEO frequency; Figure 1.3 demonstrates the concept. In the time domain, this CEO frequency causes the CEP to slip on a pulse-to-pulse basis. Isolated attosecond pulse generation depends on careful stabilization of the CEP. In most femtosecond laser systems the CEP will vary from pulse to pulse and measurements will be averaged over all values of the CEP. In most older femtosecond lasers there would have been very little effect on the measured results even if the CEP had been stabilized because the CEP does not typically start to become an important parameter until laser pulses reach the few optical cycle regime. In this regime, where the duration of the laser pulse is on the same order as the period of the laser carrier wave, the difference in electric field intensity between adjacent half-cycles of the carrier wave starts to become significant, leading to a dependence on the exact shape of the waveform. In longer laser pulses of equivalent peak intensity, the peak of the Gaussian pulse envelope flattens out the intensity difference between adjacent half-cycles of the carrier wave is small enough to cause little effect. This is illustrated in Figure 1.4. Isolated attosecond pulses by intensity gating The principle of generating an isolated attosecond pulse from HHG is the same as generating an attosecond pulse train but with two additional constraints, active stabilization of the CEP and a driving laser pulse duration on the order of 5 fs. Isolated attosecond pulse generation with CEP stabilized pulses was first achieved in the laboratory of Ferenc Krausz in 2003 [13]. The method for isolating one pulse from the train of pulses generated by HHG relies on the

22 CHAPTER 1. INTRODUCTION 7 Normalized Electric Field fs 25 fs Time (fs) Time (fs) Figure 1.4: Comparison between a 5 fs FWHM, 800 nm Gaussian laser pulse and a 25 fs FWHM pulse. The difference in electric field strength of adjacent half cycles of the laser pulse becomes much less for longer pulse durations and causes the value of the CEP to lose effect. Normalized Intensity a) b) Spectral filter Time (fs) Time (fs) Figure 1.5: Schematic of the effect of CEP on the ability to isolate a single attosecond pulse. a) A cosine pulse (CEP = 0) creates a situation in which one half cycle of the driving field is of higher intensity than all other half cycles. b) A sine pulse (CEP = π/2) where there are two half-cycles of equivalent intensity. When the spectral filter is applied an isolated attosecond pulse can be obtained from the cosine pulse but not from the sine pulse.

23 CHAPTER 1. INTRODUCTION 8 significant difference in intensity between adjacent half cycles of the laser pulse. Because the photon energy generated in the HHG process is partially determined by the laser intensity (see Equations 1.1 and 1.2), the difference in half-cycle intensities means that the HHG cutoff energy varies depending on which half-cycle produces the electrons. By stabilizing the CEP a waveform can be used for HHG that has one half-cycle that is higher in intensity than all other half-cycles (a cosine pulse). By using a CEP locked cosine pulse, one attosecond pulse of higher photon energy than the other pulses in the train will be generated every laser pulse. A spectral filter can then be used to reject all lower photon energy pulses and leave a single, isolated attosecond pulse. This is illustrated schematically in Figure 1.5a. The importance of the CEP can be visualized by shifting it π/2 radians to create a sine pulse. In this case there are two equivalent maxima of the carrier wave per laser pulse. This will result in two attosecond pulses of equivalent photon energy being generated within each laser pulse. Spectral filtering is then not possible and an isolated attosecond pulse cannot be achieved. This situation can be seen in Figure 1.5b. The second requirement, a driving pulse duration on the order of 5 fs, is related to the amount of spectral bandwidth in the high harmonics necessary to support an attosecond pulse duration. According to Equation 1.2, the bandwidth is determined by the difference in intensity in adjacent half cycles of the driving laser pulse. The larger the difference in intensities is, the larger the difference in photon energies generated by each half cycle is. This allows for the spectral filter to be placed at a lower energy to increase the bandwidth of the highest energy attosecond pulse to be isolated. According to the time-bandwidth relationship for a Gaussian pulse, (1/2π) ω t = 0.44 for a transform limited pulse [26]. This means that to support a pulse of less than 1 fs FWHM duration in the resulting high harmonic spectra, a bandwidth of at least 1.82 ev is necessary. The laser parameters necessary to generate this bandwidth depend on the HHG generation gas and the desired cutoff energy. A special case of intensity gating called two-color gating has also been investigated both theoretically and experimentally by several groups [27, 28, 29, 30, 31]. Two-color gating relies on the same principles as intensity gating but uses a HHG driving pulse that is a mixture of a fundamental frequency plus a harmonic or sub-harmonic of that frequency. This heterodyne technique breaks the symmetry of the HHG driving pulse and allows for fewer recombination trajectories. For example, adding a 400 nm laser field to an 800 nm laser field reduces HHG recombination to once every optical cycle, instead of twice for an 800 nm laser field alone. This reduces the frequency of pulses in the attosecond pulse train by a factor of two and thus relaxes the requirements on the pulse duration of the driving pulse by a factor of two. Isolated attosecond pulses by polarization gating A second technique for the generation of isolated attosecond pulses, called polarization gating, was first demonstrated by Sansone and coworkers in 2006 [32]. This technique takes advantage of the fact that a macroscopic HHG signal is only observed for linearly polarized light due to the fact that the free electron is driven away from the parent ion by a circularly

24 CHAPTER 1. INTRODUCTION 9 Figure 1.6: Principle of attosecond streaking taken from Ref. [36]. Electrons are released by the near-intantaneous attosecond pulse and receive an ultrafast sub-cycle modulation of their momentum from the few-cycle streak field. polarized field and recombination is unlikely to occur. In polarization gating the driving laser pulse is synthesized such that it is circularly polarized except for a brief moment (typically around half of an optical cycle) during which it is linearly polarized. This allows for HHG during only one half cycle and therefore only a single isolated attosecond pulse. Because this pulse does not have to be spectrally filtered from other attosecond pulses in a pulse train, it is capable of supporting a much larger bandwidth and thus a shorter attosecond pulse duration. Isolated attosecond pulses by double optical gating One of the latest techniques in isolated attosecond pulse generation is double optical gating (DOG) [33, 34, 35]. DOG combines the techniques of two-color gating and polarization gating to result in broadband supercontinuum generation supporting very short isolated attosecond pulses over a wide range of frequencies, all while relaxing the requirements on driving pulse duration Experiments using isolated attosecond pulses Attosecond streaking By far the largest number of experiments using isolated attosecond pulses to date take advantage of the technique known as attosecond streaking [36, 37, 38, 39, 40]. The principle of attosecond streaking is based on the long existing technique of the streak camera. In a streak camera a pulse of light impinges on a photocathode to produce a flow of electrons with intensity proportional to the light intensity. The electrons are then passed between a

25 CHAPTER 1. INTRODUCTION 10 Photoelectron Kinetic Energy (ev) Time Delay (fs) Figure 1.7: A typical streak trace taken in sulfur hexafluoride gas. The streaking of electrons ionized from the outer valence orbitals appears around 76 ev while the faint signal of inner valence streaking can be seen around 50 ev. pair of electrodes with a sweeping potential so that the photoelectron signal as a function of time is mapped into a spatial dimension. In attosecond streaking experiments the electrons are produced by ionization with the isolated attosecond pulse and the sweep electrodes are replaced with the electric field of few-cycle laser pulse. A schematic of the attosecond streaking principle is shown in Figure 1.6 [36]. As electrons are ionized from the target material they are accelerated by the few-cycle streak field, with the magnitude and direction of the acceleration dependent on the exact phase at which the electron is born. The electron velocity as a function of time can be described by: v(t) = e m e A(t) + [v 0 + e m e A(t i )] (1.3) where e is the elementary charge, m e is the electron mass, A(t) is the vector potential of the laser field where E L = A/ t, and t i is the time of electron ionization [41]. Because of the dependence on A(t i ), electrons ionized at different optical phases will achieve different final velocities; this is the principle which allows for time measurement on a sub-cycle timescale. By scanning the time-delay between the attosecond pump pulse and the few-cycle streak field, different values of t i can be sampled and can then be combined to form an attosecond streak trace that maps out the vector potential of laser field with sub-cycle resolution. A typical streak trace taken in sulfur hexafluoride gas is shown in Figure 1.7.

26 CHAPTER 1. INTRODUCTION 11 An early example of the use of streaking to measure electron dynamics was performed by Drescher and coworkers in 2002 [39]. In this experiment, they used an attosecond pulse to measure the lifetime of M-shell core-holes in krypton atoms. In this case the XUV attosecond pulse is used to create a core hole that is known from linewidth measurements to decay by Auger decay in a few femtoseconds. In Auger decay, ionization of an inner-shell electron creates an energetically unfavorable hole. To relax, a higher energy electron can fill the corehole, and in the process transfer enough energy to another outer shell electron to ionize it and leave a doubly charged ion behind. This secondary electron is released at a characteristic kinetic energy and can be streaked separately from the primary electron. If the electron were released promptly (i.e. a very short core-hole lifetime), a well resolved streak trace is expected. However, if the the core-hole lifetime is significantly long compared to the halfcycle of the streaking laser field, the sub-cycle oscillations will be blurred out as electrons are released at all optical phases of the streak field over the lifetime of the core-hole state. For long times, this results in sideband formation at ±1 photon energy. By analyzing the temporal behavior of the sidebands, the authors were able to determine a core-hole lifetime in Kr of fs. Attosecond streaking has also been performed from a condensed matter surface [40] and will be discussed in Chapter 4. Attosecond tunneling spectroscopy Another method of attosecond spectroscopy was introduced by Uiberacker and coworkers in 2007 [42]. In this method the authors take advantage of the fact that ionization with attosecond pulses not only liberates electrons but also leaves behind positively charged ions. Several experiments are presented in the paper, but here one is highlighted. By tracking the yield of Ne 2+ ions as a function of the delay between an XUV attosecond pump pulse and an 800 nm few-cycle probe pulse, direct observation of the timescale of light-induced electron tunneling was made possible. A mixture of Ne 1+ and Ne 2+ was initially produced by the attosecond pulse, then a strong enhancement of the Ne 2+ signal was observed when the attosecond pulse was brought into temporal overlap with the 800 nm laser field. Subfemtosecond steps in the ion yield, spaced by the half-cycle 800 nm laser period, were observed around the peaks of the laser electric field and matched well with the theory of electron tunneling put forth by Keldysh in 1965 [43]. Attosecond transient absorption The most recent measurement technique in attosecond spectroscopy is attosecond transient absorption [44, 45]. Transient absorption takes advantage not only of the extremely short pulse duration of attosecond XUV pulses, but also of the extremely large bandwidth that comes with such a short pulse duration. In this technique the few-cycle IR laser pulse is used as a pump pulse while the attosecond pulse is used as the probe pulse. By measuring

27 CHAPTER 1. INTRODUCTION 12 the dispersed spectrum of the attosecond pulse after interaction with the target gas, a large number of bound-bound or bound-free transitions can be probed simultaneously. When transitions occur, some peaks will decrease (bleaches) and some new peaks will appear (transient absorptions) in the transmitted attosecond spectrum. In this study, the few-cycle laser field is used to tunnel ionize krypton atoms and launch coherent electronic wavepackets. The attosecond XUV pulse can promote core level electrons into the hole left by tunnel ionization and thus probe populations and coherences on a sub-femtosecond timescale. 1.3 Surface plasmon resonance Definition of the surface plasmon resonance One of the most exciting phenomena in condensed matter physics today is the coherent electronic excitation in metals known as the surface plasmon resonance (SPR). The SPR is a collective oscillation of conduction band electrons that typically oscillate at optical frequencies in noble metals [46]. For a short amount of time, these electrons oscillate in phase and create a strongly enhanced electric field at the surface of the metal/vacuum or metal/dielectric interface. This field decays exponentially from the surface of the conductor, penetrating only tens to hundreds of nanometers into space [1], and it typically decays in a few to tens of femtoseconds [5, 47, 48, 15, 14, 49, 50, 51, 52]. After this time the coherence is destroyed by various mechanisms (discussed below) and the oscillation will eventually dissipate to lattice vibrations on the order of picoseconds[53]. The SPR is an extension of the concept of bulk plasmons. Bulk plasmons result when the free electrons of a metal are considered as an electron liquid that can undergo longitudinal density fluctuations. The oscillations have an energy described by: ω p = 4πne 2 m e (1.4) where ω p is the bulk plasmon frequency, n is the electron density, e is the elementary charge and m 0 is the electron mass [2, 1]. The solution to Maxwell s equations also shows that density fluctuations can be confined to the surface of the metal as a propagating or localized surface plasmon wave. This surface plasmon resonance has an energy of: where ω p is described by Equation 1.4. ω s = ω p 2 (1.5) Propagating and localized surface plasmons Typically surface plasmons will propagate along the surface of a metal just like a light wave [1]. One major drawback to applications involving propagating plasmons on flat surfaces

28 CHAPTER 1. INTRODUCTION 13 Bulk plasmon dispersion Light line Surface plasmon dispersion Figure 1.8: Diagram of the plasmon dispersion relation for bulk plasmons and surface plasmons, modified from Ref. [54]. ωp is the bulk plasmon frequency, ωs is the surface plasmon frequency, and the light line is k = ω/c where ω is the angular frequency of light and c is the speed of light in vacuum. Coupling of the light line to the plasmon modes via momentum transfer by surface roughness, kr, is shown for both plasmon modes.

29 CHAPTER 1. INTRODUCTION 14 Є 2 Є 1 Є 0 Figure 1.9: A schematic of the Kretschmann configuration for exciting propagating surface plasmon waves in a flat metal surface. Total internal reflection of the laser beam at the ɛ 0 /ɛ 1 interface launches an evanescent wave which can excite a SPR at the ɛ 1 /ɛ 2 interface. is that they cannot be directly excited by photons because the dispersion relation of SPRs falls below the dispersion relation of light propagating in vacuum (the light line, k = ω/c where ω is the angular frequency of the light and c is the speed of light in vacuum) [1, 54]. This is illustrated in Figure 1.8. Here ω p is the bulk plasmon frequency and ω s is the surface plasmon frequency. This means that a plasmon wave travels with greater momentum than a light wave of equivalent energy. In order to excite propagating plasmon waves on flat surfaces using photons, special geometries such as the Kretschmann configuration must be used [55]. In the Kretschmann geometry, illustrated in Figure 1.9, a thin metal film is deposited on the surface of a prism. The exciting laser beam is directed through one of the uncoated sides of the prism and angled such that it undergoes total internal reflection from the gold coated side. At the point of internal reflection at the ɛ o /ɛ 1 interface an evanescent light wave is launched in the interface with a wave vector k = ɛ 0 (ω/c) sin θ, where c is the speed of light in vacuum and θ is the angle of incidence from normal to the surface. If ɛ 0 sin θ > 1, the wave vector of the evanescent wave will lie to the right of the light line and can excite a SPR at the ɛ 1 /ɛ 2 interface because the wave vector is now commensurate with the SPR dispersion curve. A special case of SPRs, called localized surface plasmons, exists when a plasmon oscillation is confined to a small volume such as a nanoparticle. Localized plasmons in gold nanospheres are illustrated in Figure In these particles the free electron cloud oscillates back and forth across the particle, confining the electromagnetic energy to a small physical volume and potentially leading to a much greater enhancement of the incident electric field strength than in a propagating plasmon[2, 1]. Localized plasmons have the additional benefit that they can be excited directly by photons because the particle edges and surface roughness allow for the necessary exchange of momentum to couple the light line to the plasmon

30 CHAPTER 1. INTRODUCTION Figure 1.10: Schematic depicting localized SPRs oscillating in gold nanospheres. In a localized plasmon the electron cloud oscillates back and forth across the particle, concentrating the incident electromagnetic energy into a small physical volume. In the schematic the incident field is traveling in the direction of the arrow and driving the plasmon oscillations in the particles. The shaded area represents the electron density as the particles are macroscopically polarized. dispersion curve. This is illustrated in Figure 1.8 by process 2 1, where the plasmon dispersion curve is coupled to the light line by a momentum transfer, k r. In order to directly excited plasmons with laser light, the experiments presented in this dissertation are focused on localized SPRs Plasmon lifetimes One of the main motivating factors for studying surface plasmon resonances with attosecond pulses is the extremely short lifetime of the coherent electron oscillation. Typical plasmon dephasing times for gold and silver surfaces and nanoparticles are on the order of a few to tens of femtoseconds [5, 51]. This is the time in which the electrons lose phase coherence but are still oscillating in an excited state. After the loss of phase coherence it takes approximately 1 ps for the plasmon energy to thermalize to the ion lattice [53]. The dephasing time is an important factor because the large electric field enhancement only occurs while the electrons are oscillating in-phase with each other. If they have random phases the individual contributions will cancel out and there will be no macroscopic field effect. This means that any application that relies on the plasmon field enhancement is limited by the plasmon dephasing time. Because of this there is interest in accurately measuring plasmon dephasing times so as to understand the factors that go into it and potentially improve the custom fabrication of plasmonic systems tailored to specific applications.

31 CHAPTER 1. INTRODUCTION 16 Figure 1.11: An example of persistent spectral hole burning from Ref [52]. As the oblate goldnanoparticles on a sapphire substrate are irradiated, an increasingly large hole forms in the measured spectrum. From this hole the homogenous plasmon linewidth can be determined. In this case the homogenous linewidth is 94 mev, which corresponds to a dephasing time of 14 fs. A number of experiments have been performed in both the frequency domain [51, 15, 52, 14, 56] and the time domain [5, 47] to measure plasmon dephasing times. In the frequency domain, the plasmon dephasing time is measured by taking the inverse of the linewidth of the plasmon resonance. The commonly used tool to do this is persistent spectral hole burning. In this method a narrow-bandwidth light source is used to excite the plasmon resonance for a subset of an ensemble of particles. Because of inhomogeneous broadening, the bandwidth for the ensemble will be very large but individual particles will have more narrow bandwidths corresponding to their homogenous linewidths. The narrow-bandwidth light source is energetic enough to physically change the structure of the excited particles and thus change their plasmon resonance frequency. This leaves a hole in the spectrum when the ensemble is measured again. After accounting for power-broadening, the linewidth of this hole can be considered the homogeneous linewidth of the removed particles. An example of this method is shown in Figure In this experiment, a homogenous linewidth of 94 mev, corresponding to a dephasing time of 14 fs, is determined for oblate gold nanoparticles. In the time domain studies, the standard measurement of dephasing time is done through second- or third-harmonic generation at a nanostructured surface. By measuring an autocorrelation of the upconverted light from the nanostructures, a trace that is broadened by the plasmon dephasing time is obtained. By comparing this broadened autocorrelation trace to the non-plasmon-broadened autocorrelation obtained by using a standard frequency conversion crystal (such as a BBO for the case of second-harmonic generation), the plasmon

32 CHAPTER 1. INTRODUCTION 17 Figure 1.12: An example of second-harmonic generation from gold nanoparticles in Reference [48]. The bold line is the autocorrelation measured using a standard BBO crystal and representing the laser pulse duration. The thin line is the autocorrelation measured with second-harmonic light generated from a gold nanostructured surface. The broadening is from the plasmon dephasing lifetime and corresponds to a lifetime of 6 ± 1 fs. decay time can be extracted. A measurement made using this method is shown in Figure 1.12 [48]. In this experiment a dephasing time of 6 ± 1 fs is determined for lithographically prepared gold nanostructures. The primary mechanisms through which the plasmon electrons lose phase coherence are surface scattering, chemical interface damping (in which plasmon electrons become trapped in surface states of adsorbed molecules), electron-electron scattering and inter-band damping [56, 15, 14, 49, 57]. While these measurements have proven extremely useful, direct probing of the surface plasmon field by attosecond light pulses will be able to provide far more detailed information on plasmon dephasing processes. The primary advantage of this new technique is that it is sensitive to the phase and intensity of the plasmon electric field on an unprecedented timescale, whereas the previously used techniques have only provided intensity information on a much longer timescale. The addition of phase information and the improved time resolution will allow for the direct observation of plasmon electron dynamics with sub-cycle resolution instead of the observation of dynamics that have been averaged over the lifetime of the plasmon resonance.

33 CHAPTER 1. INTRODUCTION 18 Figure 1.13: Surface plasmon-based electron acceleration demonstrated by Irvine, et al. [60]. In this experiment a 27 fs laser oscillator pulse is used to both excite a SPR in a gold film in the Kretschmann geometry and ionize photoelectrons into the enhanced field. The electrons are then classically accelerated to high kinetic energies Overview of photoelectron spectroscopy from surface plasmon systems Recently, plasmon-enhanced photoelectron acceleration has been studied in propagating plasmons on thin metal films [58, 59, 60, 61, 62, 63, 64, 65] and extremely sharp metal tips [66, 67]. In these experiments, an ultrafast laser pulse is used to launch a propagating SPR wave. In the same laser pulse, photoelectrons released via multiphoton ionization are precisely spatially and temporally injected into the strong plasmon electric field and accelerated away from the surface. Because this type of experiment is very reminiscent of attosecond streaking, it has been explored in this dissertation as a pre-cursor to SPR enhanced attosecond streaking experiments. Figure 1.13 shows an experimental demonstration of photoelectrons being accelerated to high kinetic energies by a plasmon enhanced electric field. In this study 27 fs laser oscillator pulses are used to excite a propagating surface plasmon wave using the Kretschmann geometry. Photoelectrons are ionized via multiphoton ionization and are determined to be ponderomotively accelerated to extremely high kinetic energies after ionization. Extensive theoretical studies have shown this and similar experiments to be a classical acceleration effect [68, 59]. Variations of the photoelectron spectra as a function of CEP due to electron acceleration have also been predicted [69]. Because the phase of the enhanced plasmon field is related to the CEP of the driving laser field, electrons that are accelerated in the enhanced plasmon

34 CHAPTER 1. INTRODUCTION 19 field should be sensitive to the CEP for sufficiently short pulse durations. At short pulse durations, electrons ejected at different half-cycles of the plasmon electric field will experience significantly different enhanced field strengths and thus have different final kinetic energies. When the photoelectron spectrum is integrated over the duration of the driving laser pulse, this should result in distinct cutoffs in the photoelectron spectrum corresponding to individual half-cycles of electron acceleration. As the CEP of the driving laser pulse is varied, the distribution of photoelectron kinetic energies should vary as the distribution of half-cycle intensities changes. However, recent attempts to measure such CEP variation have so far proven to be unsuccessful [65]. The absence of the expected variation is explained by a small amount of surface roughness on the metal film that causes localized plasmon modes that oscillate out of phase with each other. Acceleration by the out-of-phase modes serves to wash out any CEP variation expected in the photoelectron spectra. This may have consequences for future experiments on SPR enhanced attosecond streaking and consideration of the exact system to study must be made carefully. 1.4 Summary The ideas presented in this chapter have described the history and theories of both attosecond pulse generation and surface plasmon resonances. Chapter 2 provides a detailed overview of the construction and operation of the apparatus that is used to conduct the experiments presented in this dissertation. In Chapter 3, I present the results of a study using a visible laser to eject electrons by multiphoton processes in the presence of plasmonenhanced electric fields. Finally, in Chapter 4, I demonstrate attosecond streaking from both a W(110) single crystal surface and an amorphous Cr thin film and discuss the additional challenges in performing condensed matter attosecond experiments over more common gas phase experiments.

35 20 Chapter 2 Experimental Apparatus 2.1 Overview As described in Chapter 1, the overall theme of this experiment is to laser-excite plasmonic resonances in metallic nanomaterials and to detect the field enhancement at the surface of the nanoparticles by electron ejection. Studies using excitation by both 800 nm infrared-visible laser pulses and extreme ultraviolet (XUV) isolated attosecond pulses, or the combination of the two, are desired. The first step towards accomplishing these goals was the construction of an experimental system to produce such light pulses and to detect electron ejection from surfaces. The apparatus generally consists of three main parts that will be described below: the generation of few-cycle 800 nm laser pulses, high-harmonic generation for the production of isolated attosecond pulses near 90 ev photon energy, and a time-of-flight (TOF) electron spectrometer and XUV grating spectrometer for detecting ejected electron kinetic energies and the HHG spectral distribution, respectively. In order to provide as useful of a guide as possible, the apparatus is described chronologically from the laser to the TOF detection apparatus. Figure 2.1 shows an overview of the entire system. 2.2 Laser System Oscillator The laser oscillator used is a Titanium:Sapphire Femtolasers Femtosource Scientific Pro oscillator [71]. The oscillator is pumped by a diode-pumped, frequency doubled, single longitudinal mode yttrium lithium fluoride (YLF) laser (Coherent Verdi V-5) that can provide up to 5 W of CW 532 nm light. The Ti:Sapphire oscillator crystal is typically pumped with 4.6 W and can produce 5 nj, 10 fs FWHM Gaussian pulses centered at 800 nm ( 100 nm FWHM bandwidth) at a repetition rate of 78 MHz. Multiple reflections from chirped multilayer dielectric mirrors (CM in Figure 2.2) are used to correct for dispersion in the oscillator

36 CHAPTER 2. EXPERIMENTAL APPARATUS 21 Chirped Mirror Compressor f=1m 0.7% Hollow-core fiber f=2m Oscillator Amplifier Slow loop f=0.5m High Harmonic Generation Time-of-flight spectrometer CCD XUV Spectrometer Verdi V-5 Femtolaser Compact Pro Evolution 15 Evolution-15 Figure 2.1: An overview of the experimental apparatus. To amplifier To f-2f interferometer BS2 CP WP CM PD FI BS1 OC CM L Ti:S RR GB Stretcher TOD mirrors PO Oscillator PL Figure 2.2: Schematic of the Femtolasers Femtosource Scientific Pro oscillator. PL - pump laser, L - lens, Ti:S - titanium sapphire crystal, CM - chirped mirror, WP - wedge pair, OC - output coupler, CP - compensating plate, BS1-50:50 beamsplitter, BS2-30:70 beamsplitter, FI - Faraday isolator, TOD mirrors - chirped mirrors for third order disperson (TOD) compensation, GB - 10cm long SF10 glass block, RR - retro-reflector, PO - pick-off mirror, PD - photodiode. Modified from Ref. [70]

37 CHAPTER 2. EXPERIMENTAL APPARATUS 22 To compressor PBFM PD IRFM2 PBS1 PO1 BW PBS2 RR1 C Ti:S RR2 VC P3 BW L4 PO2 IRFM1 P2 L3 L2 L1 From pump laser PC BC P1 P4 From stretcher Figure 2.3: Schematic of the Femtolasers Femtopower Compact Pro amplifier. L1,2 - telescope for pump beam, L3 - lens for focusing of pump beam, PBFM - pump beam focusing mirror, P1-4 - periscopes, IRFM1,2 - infrared focusing mirrors, RR1,2 - retro-reflectors, PBS1,2 - polarizing beam-splitters, PC - Pockel s cell, BC - Berek polarization compensator, PO1,2 - pick-off mirror, VC - vacuum chamber, BW - Brewster window, Ti:S - titanium sapphire crystal, C - Peltier cooling, PD - photodiode. Courtesy of Ref. [70] and to produce as short a pulse duration as possible. Before seeding the pulse to the amplifier it is temporally stretched to 3 ps to keep the peak power below the damage threshold of the Ti:Sapphire crystal and other optics in the amplifier. Multiple reflections from a pair of chirped mirrors (TOD Mirrors in Figure 2.2) are used to pre-compensate for third order dispersion induced by optical components in the amplifier. A schematic of the oscillator and stretcher is shown in Figure Amplifier The laser amplifier is a Femtolasers Femtopower Compact Pro 9-pass Ti:Sapphire amplifier. The crystal is pumped by a diode-pumped, frequency doubled YLF laser (Coherent Evolution 15) that can provide up to 15 W of 532 nm light at 1 khz repetition rate. The amplifier crystal is typically pumped in constant current mode at around 20 A while being cooled to 230 K and housed in vacuum at a pressure of < 50 mbar. The first four passes through the Ti:Sapphire crystal amplify all of the pulses in the oscillator pulse train. After the fourth pass a Pockel s cell is used to select the most energetic pulses from the oscillator pulse train and reduce the repetition rate to 1 khz. Five additional passes through the Ti:Sapphire crystal further amplify the 1 khz pulse train. By only placing the Pockel s cell after the first four passes, the buildup of amplified spontaneous emission is greatly reduced. Gain narrowing, which results when the central frequencies of the laser pulse are amplified more

38 CHAPTER 2. EXPERIMENTAL APPARATUS 23 Translation PO Output From amplifier Figure 2.4: Schematic of the Femtolasers Femtopower Compact Pro prism compressor. Courtesy of Ref. [70] than the spectral wings, decreases the pulse bandwidth to approximately nm FWHM, substantially less than the oscillator bandwidth. The amplifier output is then directed into a prism compressor for temporal re-compression. The amplified pulse bandwidth supports a transform-limited pulse duration of 25 fs, but imperfect compensation of fourth and higher order dispersion results in a final pulse duration of 30 fs. Typical amplified laser power measured before the compressor is 1 W Prism Compressor In order to temporally re-compress the amplified laser pulse, a prism based compressor is used instead of a more traditional grating based compressor. Prism compressors can often have higher throughput efficiency than grating compressors, and although it was initially thought that carrier-envelope phase stabilization would be problematic with grating compressors, this has been shown not to be the case [72]. Typical throughput for this compressor is approximately 80%, giving a final amplified output of 850 µj per pulse and 30 fs FWHM Gaussian pulses at a 1 khz repetition rate. A schematic is shown in Figure 2.3.

39 CHAPTER 2. EXPERIMENTAL APPARATUS 24 Time (mins ) Wavelength (nm) Figure 2.5: Spectral fringes measured in the slow loop. The first 8 minutes show spectra collected while varying the offset voltage that is sent to the phase locking electrons. The changing voltage varies the CEP and thus the positions of the spectral fringes. From 8 minutes onward, the offset voltage is only controlled by the slow loop error signal and long-term locking stability of the CEP is demonstrated. Inset: Fast loop beat signal measured on a spectrum analyzer.

40 CHAPTER 2. EXPERIMENTAL APPARATUS 25 PCF MO Glass Wedges P L DC P L BBO L P BPF L PD Figure 2.6: A schematic of the Mach-Zehnder interferometer used in the CEP fast loop. MO - microscope objective, PCF - photonic crystal fiber, L - lens, DC - dichroic mirror, P - polarizer, BBO - beta barium borate crystal, BPF - band pass filter, PD - photo diode.

41 CHAPTER 2. EXPERIMENTAL APPARATUS Carrier-Envelope Phase Stabilization f-2f Mach-Zehnder Interferometer A special feature of this laser system is the ability to actively lock the carrier-envelope phase (CEP). The CEP results from the carrier-envelope offset (CEO) frequency of the oscillator frequency comb and has been defined and discussed in detail in Section The CEP is locked using a method known as f-2f interferometry. In this technique it is required that the light spectrum span at least an optical octave, meaning that the highest frequencies contained in the pulse are at least twice the lowest frequencies in the pulse. If this condition is satisfied, the pulse can be split into two copies of itself and then one of the pulses can be sent through a frequency doubling crystal. Because the highest frequencies in the fundamental pulse are at least twice the lowest frequencies, the low frequency side of the doubled laser pulse will be spectrally overlapped with the high frequency side of the fundamental pulse. By spatially and temporally overlapping the pulses in this spectral region, a self-referencing beat signal can be measured. This beat signal is proportional to the CEO frequency and can be locked on using an electronic control loop [73]. In this system there are two stages of CEP locking. The first stage, called the fast loop, locks the CEP of selected pulses in the oscillator pulse train. A schematic of the fast loop is shown in Figure 2.6. Because the Femtosource oscillator does not span an optical octave on its own, the beam must be focused through a spectrally broadening material in order to perform f-2f interferometry. In the fast loop this source of white light generation is a photonic crystal fiber (PCF). The PCF is a microstructured fiber designed to have a strong nonlinear response to spectrally broaden low energy laser pulses. The fiber consists of a solid cladding with a microstructured center. A 50% beamsplitter is placed into the output of the oscillator (before the stretcher) to direct enough light to the f-2f for sufficient spectral broadening. Typical power measured after the PCF is mw. This light is then split in a Mach-Zehnder interferometer configuration. A dichroic mirror is used to send different frequency light to the two arms so that pulse energy is not wasted by doubling light that won t be in the spectral overlap region. The light in the green arm (bottom arm as shown in Figure 2.6) is focused into a 3 mm thick Type I beta barium borate (BBO) crystal for frequency doubling. The two arms are recombined and then passed through a narrow bandpass filter (typically 510 nm with a 10 nm bandwidth) to isolate the region of spectral overlap. A pair of glass wedges mounted on a translation stage can be used to tune the temporal overlap between the two arms for maximum beat signal. The signal is measured on a photodiode and fed into the CEP locking electronics. The inset of Figure 2.5 shows the typical beat signal measured on a spectrum analyzer. The beat signal is located at 30 MHz and a signal-to-noise ratio of 30 db is typically required for effective CEP locking. The fast loop is one of the most critical components of the experimental apparatus but also one of the most common problem areas. The main problem is drifting of the spectral overlap region between the two arms of the interferometer. Because a narrow bandpass filter

42 CHAPTER 2. EXPERIMENTAL APPARATUS 27 is used before the photodiode, drifting of the overlap region in wavelength on the order of less than 10 nm will cause the beat signal to drop substantially. Because this problem can happen on the timescale of experimental operations (several hours), it is not practical to change the bandpass filter in the middle of operation. Instead, two avenues are currently being approached to improve the system. The first is to try to eliminate spectral drifts by actively stabilizing the pointing of the oscillator beam into the PCF. By stabilizing the pointing of the input, the spectrum of the output should also be stabilized and hopefully reduce spectral drift. A longer term issue is drift within the oscillator itself, which can also lead to changes in the PCF output. Ultimately the oscillator will also need to be stabilized in one or several ways. The second avenue of improvement for the fast loop is to replace the narrow bandpass filter with a grating and a slit so that the spectral overlap can be visually determined and selected by an adjustable slit. This could potentially be automated to correct for spectral drifts in real time. The second stage of CEP locking, called the slow loop, is measured after the amplifier and compressor and is used to correct slow drifts in the CEP introduced during the amplification process. The f-2f configuration is the same as for the fast loop with one notable exception. Because the amplified pulse is of much higher energy than the oscillator pulse, the octave spanning white light can be generated by focusing into a 2 mm thick sapphire plate. The signal from the photodiode is then fed into a Labview controlled proportionalintegral-derivative (PID) control loop that produces an error signal. This error signal is sent into the CEP locking electronics as an offset voltage to pre-compensate for CEP drifts in the amplifier and compressor by adjusting the fast loop appropriately. Figure 2.5 demonstrates both control of the CEP (by varying the offset voltage, which results in a change of CEP and thus different positions of the spectral fringes) and long term locking stability of the spectral fringes from the slow loop f-2f. Typical root-mean-squared fluctuations are on the order of 200 mrad when averaged over 10 ms. When the spectral drift of the fast loop is stable, sustained phase locking can be achieved for greater than two hours CEP Locking Electronics The CEP locking electronics in this apparatus have been homebuilt through a collaboration with the group of Jun Ye at JILA at the University of Colorado Boulder. The system consists of three main parts: a phase detector, a fast PID control loop and a fast summer. The photodiode signal from the fast loop is sent into the phase detector where the beat signal is selected and compared to the oscillator repetition rate signal in order to determine the phase difference. This phase is then fed into the PID loop where a phase error signal is generated. In order to keep the phase constant this error signal is used to adjust the oscillator CEO frequency. However, before being sent back to the oscillator, the error signal is combined in a fast summer with the offset correction provided by the PID controller of the slow loop. This combined error signal is then sent to an acousto-optic modulator (AOM) that is placed in the pump beam of the oscillator. Small changes to the oscillator pump

43 CHAPTER 2. EXPERIMENTAL APPARATUS 28 power will change the peak power of the oscillator laser pulse and vary the refractive index of the Ti:Sapphire crystal due to the Kerr effect. This causes a change to the group and phase velocities of the laser pulse and thus changes the CEO frequency. By using the AOM to slightly deflect the pump beam and change the pumping power, the CEO frequency can be stabilized according to the error signal and provide a locked CEP. 2.4 Spectral Broadening in Hollow-Core Fiber As described previously, CEP does not typically become an important parameter until the laser pulse duration is in the few-cycle regime. The amplified laser pulses are already 25 fs, but for isolated attosecond pulse production much shorter laser pulses, on the order of 5 fs, are needed. Laser pulse durations are limited by the energy-time uncertainty principle, according to Heisenberg. According to this principle, the spectral bandwidth and the temporal duration of the laser pulse must satisfy the relationship E t, where E is the spectral bandwidth, t is the pulse duration and is the reduced Planck s constant. Because 2 of this, the spectral bandwidth of the pulse must be increased to obtain a shorter pulse duration. This is achieved by focusing the laser into a gas-filled 1 m long, hollow-core glass fiber with a 250 µm inner diameter [74]. The beam is focused using a 1 m focal length lens to achieve the optimum 64% ratio between focused beam diameter and the diameter of the core [75]. The fiber is housed in an outer tube that can be evacuated and backfilled with neon gas, simultaneously filling the hollow core with Ne. Typical gas pressures used range from 1.4 Bar-2.0 Bar. Spectral broadening occurs through the nonlinear process of self-phase modulation. This process relies on the nonlinear refractive index of the propagation medium, n(i). Because of the intensity dependence, a Gaussian laser pulse will induce a time-varying refractive index in the Ne gas that will in turn induce a shift of the instantaneous spectral phase. This phase shift will introduce new spectral components and give the pulse increased bandwidth, ultimately supporting a shorter temporal duration according to the uncertainty principle. Figure 2.8 shows the spectrum measured at the output of the hollow-core fiber and before temporal recompression. The output of the hollow-core fiber is very sensitive to the pointing stability of the input to the fiber. One of the most important components of this apparatus was the installation of an active beam pointing stabilization system between the amplifier and the input of the hollow-core fiber. Before installation of this system, the fiber input often had to be adjusted as often as every 30 min. After installation of the stabilization system, the fiber input almost never has to be touched, and the beam stabilization system typically only needs to be reset every few days unless significant alignment to the laser has occurred. The stabilization is provided by a commercial system from MRC Systems (Germany) and is diagrammed in Figure 2.7. It consists of two actively controlled mirrors mounted with piezoelectric actuators and two quadrant photodiodes to detect pointing fluctuations. By using two mirrors and two photodiodes, full control of the beam pointing position and angle is available. A fast

44 CHAPTER 2. EXPERIMENTAL APPARATUS 29 PZ1 L BS HCF PZ2 QPD2 QPD1 Figure 2.7: A schematic of the beam stabilization system for input into the hollow-core fiber (not drawn to scale). PZ1, PZ2 - Piezo actuated mirrors, L - focusing lens, QPD1, QPD2 - Quadrant photodiodes, BS - Beamsplitter, HCF - Hollow-core fiber.

45 CHAPTER 2. EXPERIMENTAL APPARATUS 30 Figure 2.8: Laser pulse spectrum measured after spectral broadening in a hollow-core fiber filled with 1.9 Bar of Ne gas but before temporal recompression. electronic loop provides feedback. 2.5 Chirped Mirror Compressor While propagation through the gas-filled hollow-core fiber introduces the desired spectral broadening, it also introduces a significant amount of dispersion resulting in a positively chirped pulse in the time domain. Positive chirp is a condition that occurs when lower frequency spectral components travel through a medium at higher velocities than higher frequency components. This results in a pulse that has the spectral components spread out along the time axis. In order to achieve the lowest pulse duration allowed by the uncertainty principle, the temporal phases of the various frequency components must be aligned. In this apparatus this is done through the use of a series of negatively chirped multilayer mirrors. These chirped mirrors are designed such the penetration depth into the surface of the mirror is a function of the wavelength of the light [76]. For a positively chirped laser pulse, negatively chirped mirrors allow the low frequency side of the pulse to penetrate further into the surface and thus have a longer round-trip reflection time. Higher frequencies penetrate less deep into the surface and have a faster reflection time. By calculating the amount of chirp on the input pulse, a chirped mirror compressor can be designed to balance the frequency components to a near Fourier transform-limited pulse duration. Figure 2.9 shows a second-order autocorrelation trace of the laser pulse after chirped mirror compression. A simulated fit to the data gives a FWHM pulse duration of 6.5 fs. The deviation from the

46 CHAPTER 2. EXPERIMENTAL APPARATUS 31 Autocorrelation Signal FWHM = 6.5 fs Measured Fit Time Delay (fs) Figure 2.9: A measured autocorrelation trace of the few-cycle laser beam after spectral broadening in the Ne filled hollow-core fiber and temporal recompression in the chirped mirror compressor. The black circles are experimentally measured points while the red line is the calculated fit to the measured data. The corresponding FWHM laser pulse duration is 6.5 fs. The deviation from the fit in the wings of the pulse results from higher order phase terms that manifest as pre- or post-pulses. These are not well characterized by autocorrelation but could be characterized using a pulse-reconstruction technique such as SPIDER or FROG. fit in the wings of the pulse results from higher order phase terms that manifest as pre- or post-pulses. These are not well characterized by autocorrelation but could be characterized using a pulse-reconstruction technique such as SPIDER or FROG [77, 78]. 2.6 Vacuum System Many portions of this apparatus are housed inside a high vacuum system with a base pressure maintained below Torr. A schematic of the vacuum chamber system is shown in Figure The chamber consists of four regions: high-harmonic generation chamber, differential pumping and filter chambers, experimental chamber and XUV spectrometer chamber. The high harmonic generation setup will be described in detail in Section 2.7, but briefly it consists of a nickel tube gas cell with holes drilled in two sides for the laser beam to pass through. Because of the high pressure required for efficient production of high harmonics,

47 CHAPTER 2. EXPERIMENTAL APPARATUS 32 Differential Pumping High Harmonic Generation CCD XUV Spectrometer Time-of-flight spectrometer Figure 2.10: Schematic of the vacuum chamber high gas loads result. Therefore the chamber is pumped with a 2000 L/s turbomolecular pump backed by an oil-free scroll pump. A metal baffle with a hole drilled for the laser to pass through separates this chamber from the differential pumping stage. Differential pumping is used to decrease the pressure inside the vacuum chamber from the high pressure inside the HHG chamber to the low pressure necessary in the experimental chamber. By separating the regions of high and low pressure with small apertures and by pumping the differential volume, the experimental chamber can be effectively isolated from the high gas load in the HHG chamber. This region is pumped by two 200 L/s turbomolecular pumps backed by oil-free scroll pumps. The differential pumping stage also contains linear vacuum feedthroughs on which a variety of filters for spectral selection and beam separation can be mounted. The third region is the experimental chamber. In this chamber the main photoelectron spectroscopy experiments are performed using the IR and XUV laser pulses and the timeof-flight electron spectrometer. The laser pulses are reflected and focused at near-normal incidence using a combination of a gold coated mirror and an XUV multilayer mirror (Section 2.9). The mirror is mounted on a multi-axis picomotor stage to allow for precise control over pointing of the laser beam while remaining under high vacuum conditions. The desired sample is mounted on a three-axis translation stage with encoded picomotors for reproducible positioning of the sample with respect to the focus of the laser beam. The reflected laser light from the sample is directed through a window to outside of the chamber and used to image the sample surface to allow for exact positioning of the laser focus with respect to the sample surface. Photoelectrons emitted from the sample are detected normal to the surface by a linear time-of-flight electron spectrometer described in detail in Section This chamber is pumped by a 200 L/S turbomolecular pump backed by an oil-free scroll pump. The fourth region is a homebuilt XUV spectrometer used for measuring the spectrum

48 CHAPTER 2. EXPERIMENTAL APPARATUS 33 Figure 2.11: Typical HHG cutoff spectrum generated by an 800 nm laser pulse in Ne gas. This spectrum was generated by laser pulses with an unlocked carrier-envelope phase. of light created in the high harmonic generation process. It consists of a grating/slit and a liquid nitrogen cooled backlight CCD camera. It is described in detail in Section High Harmonic Generation The primary tool used in generating isolated attosecond XUV pulses is a technique known as high harmonic generation (HHG). While the mechanisms have been described in detail in Section 1.2.1, the experimental details as implemented in this apparatus will be described here. A typical HHG spectrum measured from neon gas is shown in Figure A concave, silver coated spherical mirror with a focal length of 0.5 m is used to focus the few-cycle IR laser beam into a thin gas cell flowing with Ne gas. The focal spot size is 50 µm 1/e 2 radius and the peak intensity is up to W/cm 2. The gas cell consists of a nickel tube with 3.2 mm outer diameter and an inner diameter of 1.2 mm. Two holes approximately 0.5 mm in diameter have been drilled into opposite sides of the tube to allow the laser beam to pass through. In order to maintain the desired gas pressure inside of the tube, these holes are

49 CHAPTER 2. EXPERIMENTAL APPARATUS 34 covered with Teflon tape and the laser is allowed to burn through the tape. This provides for the smallest possible holes for the laser to pass through while still maintaining gas pressure inside of the cell. A typical tape job will last for 1-4 weeks depending on frequency of use. After this time, small changes in the laser pointing will broaden the hole to a larger diameter and decrease HHG efficiency and the cell must be re-taped. Throughout these experiments, neon gas is used as the HHG generation medium. This provides for a cutoff region around ev. The gas cell pressure is a critical parameter towards achieving the optimum phase matching conditions for HHG. The pressure used in this apparatus varies as the laser-drilled holes broaden, but it is typically around 150 Torr of Ne gas. Another critical parameter for phase matching conditions is the position of the gas cell with respect to the laser focus. In order to preferentially select the short electron recombination trajectories the entrance to the gas cell is placed approximately 3 mm after the laser focus. In optimal generation conditions, an average XUV photon flux of photons/second is estimated within the bandwidth of the molybdenum/silicon multilayer XUV mirror (described in Section 2.9). The photon flux estimate is an approximation based on known detector parameters such as the number of electrons necessary to generate a single count, the number of electrons generated per photon and the quantum efficiency of the detector. The transmission spectrum of the Zr filters (Section ) and the reflectivity of the XUV mirror must also be accounted for. For the generation of isolated attosecond pulses there is one additional parameter that must be considered beyond normal HHG - the carrier-envelope phase. As described in Section 1.2.2, a laser pulse with an unlocked CEP will generate a train of attosecond pulses that appear as a spectrum of discrete harmonics in the frequency domain. By locking the CEP and generating harmonics with a cosine pulse, the high energy region of the HHG spectrum can consist of a single isolated attosecond pulse, resulting in a continuum in the frequency domain. Figure 2.12 shows two HHG spectra generated with CEP locked laser pulses at different relative values of the CEP. Because it is not straightforward to measure the absolute CEP of the laser pulse, possible production of an isolated attosecond pulse is typically gauged by checking for spectral continuum at the high energy side of the HHG spectrum. The two spectra shown in Figure 2.12 are shifted by π phase relative to each other. In the spectrum collected at 0 relative CEP (red), discrete harmonics are observed. In the spectrum collected at π relative CEP (black), a continuum is obtained around 95 ev. This continuum can then be spectrally selected to obtain an isolated attosecond pulse. Experimentally, there are many parameters to be tweaked in order to obtain the optimal conditions for generating an isolated attosecond pulse with HHG. For fine controlling the pulse duration, a pair of glass wedges are inserted on a translation stage before the chirped mirror compressor. The amount of glass inserted into the beam path is adjusted to achieve the best continuum in the HHG spectrum. In addition to the pulse duration, the gas pressure inside the hollow-core fiber and the gas pressure inside the HHG gas cell are scanned for best HHG continuum on a routine basis.

50 CHAPTER 2. EXPERIMENTAL APPARATUS Normalized HHG Intensity Relative CEP Relative CEP Photon Energy (ev) Figure 2.12: Generation of a continuum in the HHG cutoff region by using a CEP locked 800 nm laser pulse. The two spectra are taken at zero and π relative CEP. The continuum at π relative phase is indicative of an isolated attosecond pulse in that energy region.

51 CHAPTER 2. EXPERIMENTAL APPARATUS 36 Figure 2.13: Calculated transmission of a 200 nm thick Zr foil filter. For use with the XUV spectrometer, two such filters are used in-line. 2.8 Metal filters Linear translation feedthroughs in the differential pumping region allow spectral filters to be placed in the optical path. In this apparatus, two types of filter are commonly used. For use when only the HHG spectrum is desired and no IR light should be present, two 200 nm thick Zr filters, 5 mm in diameter, are mounted in-line with each other. Figure 2.13 shows the calculated transmission spectrum for one of these filters. The other filter used is a single 200 nm thick Zr filter, 2.2 mm in diameter, mounted on a 1 inch aluminum coated pellicle, shown in Figure The Zr filter has the same transmission as shown in Figure 2.13, transmitting only the XUV light, while the aluminum coated pellicle has partial transmission of the IR light (ranging from 0.1% to 10% transmission, depending on the filter used and the desired amount of IR light). This filter is used to spatially separate the IR laser light and the XUV HHG light in order to subsequently introduce a time delay between them and perform a pump-probe experiment. 2.9 XUV Optics In order to select only the portion of the HHG spectrum that has an isolated attosecond pulse, special optics designed to reflect a narrow spectral range in the XUV must be used. In this case multilayered mirrors optimized for reflection near 93 ev have been designed and provided by the Center for X-Ray Optics (CXRO) at Lawrence Berkeley National Laboratory. Multilayer XUV mirrors work on the principle of interference [79]. The material type, number and spacing of layers are optimized such that each interface backscatters some light and only

52 CHAPTER 2. EXPERIMENTAL APPARATUS mm Figure 2.14: 200 nm thick Zr filter mounted on an aluminum coated pellicle. This filter serves to separate the copropagating XUV and IR light into an inner beam and an outer beam so that a time delay can be introduced between the two. Figure 2.15: Multilayer Mo/Si XUV mirror reflectivity shown compared to the HHG continuum. The mirror reflectivity is designed to spectrally select only the continuum region of HHG, leaving an isolated attosecond pulse after reflection.

53 CHAPTER 2. EXPERIMENTAL APPARATUS 38 Outer mirror mount Fiber aligner Piezo Micrometer/ translation stage Inner Mirror 5-Axis stage Figure 2.16: Schematic of the cored-mirror used in this apparatus. The central portion of the mirror has a multilayer XUV coating and can be moved in the beam axis independently from the outer mirror by a piezo-translation stage. The outer mirror (not shown here for clarity) is gold coated and can be moved in the x- and y-directions by picomotors to allow for precise spatial overlap between the inner and outer beams. the desired wavelengths will constructively interfere. In the case of HHG with neon gas, molybdenum/silicon multilayer mirrors are made to reflect around a central energy of 93 ev with a 4 ev FWHM bandwidth. Figure 2.15 shows the spectral reflectivity of the multilayer mirror used in this apparatus compared to the CEP-locked HHG spectrum. The maximum measured reflectivity is 70%. The reflectivity is clearly centered in the continuum region of the HHG spectrum. This allows for only the energy region with an isolated attosecond pulse to be reflected. In this apparatus the multilayer mirror serves not only as a spectral filter, but also to focus the XUV and IR light onto the sample and to introduce the pump-probe time delay between the two pulses. For this purpose a special cored-mirror geometry is used. A concave spherical mirror substrate with a 10 cm focal length and a surface figure of λ/10 (measured at 633 nm) or better has a 3 mm diameter core cut out of the center. The core is retained

54 CHAPTER 2. EXPERIMENTAL APPARATUS 39 to give two separate mirror substrates, the 3 mm diameter center piece and the outer mirror with the 3.5 mm hole in the center. The center portion is then coated with the multilayer Mo/Si XUV coating and the outer portion is given a high reflectivity gold coating. By attaching the center mirror to a piezeoelectric controlled translation stage and reinserting it into the outer mirror, the mirror will separately reflect the colinear XUV attosecond pulse from the center portion and the IR pulse from the outer portion. It also introduces a time delay between the two pulses by moving the inner mirror along the beam axis (z-direction) with respect to the outer. A schematic of the mirror design is shown in Figure The piezo stage in this apparatus has a minimum step size of 15 nm, corresponding to a minimum time step of 5.2 fs (twice the minimum step size because reflected light covers the distance twice). The outer mirror can be moved in the x- and y-directions by picomotors to allow for precise spatial overlap between the inner and outer beams. One major cause of experimental challenge regarding the cored mirror is achieving interferometric stability between the inner and outer mirror on the order of 15 nm. Even small vibrations in the apparatus, introduced by the turbomolecular pumps for example, can lead to a complete loss of timing information at the necessary temporal resolution. This stability can be checked by illuminating the inner and outer mirrors simultaneously with a CW red He/Ne laser and observing the interference pattern created. In order for attosecond experiments to be possible there must be stable fringe contrast. If the fringes are visibly shifting large amounts or blurring out entirely, steps must be take to reduce vibrations in the cored-mirror design. It is important to use sturdy mirror mounts and translation stages, but also to implement measures such as placing the chamber on vibration damping rubber foam or using vibration isolating bellows between the turbomoleuclar pumps and the vacuum chamber Time-of-Flight Electron Spectrometry TOF Spectrometer The primary method of experimental detection used throughout this dissertation is time-offlight (TOF) electron spectroscopy. For this purpose a home built TOF electron spectrometer was constructed. The principle of TOF spectroscopy is rather simple and relies only on basic classical physics. By using a defined source of particle emission, in this case photoelectrons ionized by the laser pulse, one can measure the length of time taken to travel to a detector a known distance from the source. By knowing the particle charge, mass and the distance to the detector, the kinetic energy of the particle can easily be determined. In this apparatus one of the most simple versions of a TOF electron spectrometer is used. A 0.59 m long µ-metal tube with an inner diameter of 45 cm is placed as close to the laser-surface interaction region as possible. µ-metal is a material with a very high magnetic permeability that allows it to effectively shield an enclosed region from the Earth s magnetic

55 CHAPTER 2. EXPERIMENTAL APPARATUS 40 field. It is important to use such a material for photoelectron spectroscopy because very light, charged particles such as electrons will have their trajectories significantly deflected by the Earth s magnetic field over the length of the TOF spectrometer. Such a deflection will lengthen the time-of-flight and cause a falsely low kinetic energy measurement. The tube is grounded to ensure field-free electron flight. At the end of the tube a micro-channel plate (MCP) detector is mounted. MCPs are glass plates with a large array of small channels permeating them. The interior of the channels are coated with a low work function material so that the impact of a particle will release a cascade of electrons. These electrons impinge on an anode at the rear of the microchannel plate and create a measurable electric current. In order to obtain easily measurable signal levels, multiple microchannel plates and electron accelerating voltages are typically used. In this apparatus a Chevron stack configuration is used. In this configuration two microchannel plates are mounted on top of each other with their channels angled such that the cascade of electrons from the first plate easily flows into the channels of the second plate. A grounded grid is placed in front of the first plate in order to provide a flat field while still allowing the electrons to pass through. The plates are then held at varying positive voltages to drive the electrons through to the anode. In this apparatus the front of the first plate is held at 160 V, the back of the first plate and the front of the second plate at 960 V, the back of the second plate at 1760 V, and the anode at 1920 V. A voltage dividing box is used to easily provide these potentials using only a single 4000 V input. A capacitive coupling box is used to obtain a voltage signal from the anode through a standard 50 Ω BNC connector without risking exposure to the high voltage being applied to the anode. A circuit diagram of the entire MCP apparatus shown in Figure Signal analysis The signal from the MCP is routed through an analog constant fraction discriminator to correct for timing errors introduced by the MCP pulse height distribution and then counted by a digital multi-channel scaler installed in a PC. The Fast Comtec multi-channel scaler has a bin size of 500 ps. The spectrometer resolution is determined by the length of the flight tube, the electron kinetic energy and the bin size of the multi-channel scaler. Figure 2.18 shows a plot of the TOF kinetic energy resolution versus electron kinetic energy over the range of electrons typically detected in this apparatus. Data is collected via a variety of commercial programs and self-made Labview routines and data analysis is typically done using self-made Matlab routines. Data collection times vary depending on the particular experiment, but for most samples each step is integrated for laser pulses ( 1 s) or less.

56 CHAPTER 2. EXPERIMENTAL APPARATUS 41 Figure 2.17: Circuit diagram of the MCP apparatus used in TOF detection. courtesy of Jordan TOF Products, Inc. Provided

57 CHAPTER 2. EXPERIMENTAL APPARATUS 42 Figure 2.18: Kinetic energy resolution of the TOF electron spectrometer as a function of electron kinetic energy. For valence electrons emitted directly by the HHG produced XUV pulse, around 90 ev, the energy resolution is 0.9 ev. The plot only accounts for the instrument resolution and does not include the bandwidth of the XUV pulse or any other contributions to the final experimental resolution. 400 Electron counts Time of Flight (ns) Figure 2.19: A sample TOF photoelectron spectrum collected from ionization of a gold nanopillar sample by the few-cycle IR laser and integrated over laser pulses.

58 CHAPTER 2. EXPERIMENTAL APPARATUS Electron Counts Time of Flight (ns) Figure 2.20: Raw data for a TOF photoelectron spectrum collected from ionization of a W(110) crystal by HHG produced 93 ev photons and integrated over laser pulses. The peak at 104 ns is caused by scattered photons and can be used for calibrating the spectrometer time zero. To the right of the photon peak a sharp peak from XUV emitted photoelectrons can be seen, followed by a large low energy electron background Sample TOF data Figure 2.19 shows a typical TOF electron spectrum collected after irradiation of a gold nanopillar sample with the few-cycle IR laser pulse. The spectrum is shown as raw timeof-flight data, integrated over laser pulses. The total number of counts is , giving an average of 5217 electrons detected per second. It is important to subtract off the time-zero before converting the time-of-flight to a kinetic energy scale. The time zero results from slight timing discrepancies between the 1 khz trigger signal from the laser and the time at which the pulse arrives in the interaction region and releases photoelectrons. This time is determined by both the optical path length from the laser to the interaction region and the delay of electrical signals in the various connections. It is most easily determined by observing the photon peak, which results from photons scattering off of the target surface and arriving at the detector near-instantaneously. It can be difficult to observe the photon peak with ionization by the few-cycle laser pulse alone, but typically it is very easy to observe when irradiating the surface with HHG radiation. Figure 2.20 shows a typical TOF electron spectrum collected by ionization of a W(110)

59 CHAPTER 2. EXPERIMENTAL APPARATUS 44 crystal using HHG generated XUV pulses (not an isolated attosecond pulse) centered at 93 ev. The spectrum is integrated over laser pulses with an average of 3311 electrons detected per second. The large low energy (long flight time) background is the result of electron kinetic energy loss via inelastic electron scattering inside the metal surface (this results from the fact that the XUV light will penetrate 3 nm into the surface [80] while the electron mean free path is only 0.4 nm at 90 ev kinetic energy [81]). This issue will be discussed further in Chapter 4. The small photon peak can be seen at 104 ns and is used for calibration of time zero. In order to convert the time-of-flight axis to a kinetic energy axis a transformation has to be applied to not only the independent axis but also to the dependent axis. This correction to the dependent axis is called the Jacobian and results from the non-uniform bin width of the scaler card in the kinetic energy domain. The procedure for converting the independent axis is straightforward: E ke = 1 2 mv2 = 1 2 m ( d t ) 2 (2.1) where d is the distance from the interaction region to the MCP detector and t is the electron flight time minus time zero. For the Jacobian correction to the dependent axis the following condition must be fulfilled: D(t) dt = D(E ke ) de ke D(E ke ) = dt D(t) de ke (2.2) where D(t) is the experimentally collected photoelectron counts as a function of time-offlight and D(E ke ) is the Jacobian corrected intensity corresponding to the kinetic energy bins, therefore: t = d m E 1 2 ke 2 m dt = d de ke 2 2 E ke D(E ke ) = d m E 2 ke D(t) (2.3) The intensity scale for the corrected kinetic energy spectrum can be brought back to physical units by multiplying the dependent axis by the correct ratio to make the integrated area under the spectrum equal to the integrated area under the time-of-flight spectrum. Figure 2.21 shows the data from Figure 2.20 converted to kinetic energy. The peak centered at approximately 86 ev results from valence band photoemission by the XUV pulse while the large low kinetic energy signal is the result of electrons scattered within the metal, as mentioned above. The peak is very broad due to both the broad bandwidth of the XUV mirror (4 ev FWHM) and the bandwidth of the tungsten conduction band, which spans almost 10 ev [82]. 3 2

60 CHAPTER 2. EXPERIMENTAL APPARATUS Electron Counts/Second Kinetic Energy (ev) Figure 2.21: The TOF data from Figure 2.20 after conversion to a kinetic energy scale and correction for the Jacobian. The peak centered at 86 ev is from XUV-induced valence band photoemission while the large low energy signal is the result of inelastically scattered electrons within the metal.

61 CHAPTER 2. EXPERIMENTAL APPARATUS 46 CCD Figure 2.22: Schematic of the XUV spectrometer XUV Spectrometer The final component of the experimental apparatus is a homebuilt XUV spectrometer. A schematic is shown in Figure The spectrometer consists of a slit and transmission grating that are used to disperse the light onto a backlit, liquid nitrogen cooled CCD camera (Princeton Instruments XO Pixis 400B). The slit width is 500 µm and the transmission grating consists of free standing 100 nm thick Si 3 N 4 bars with a line density of lines/mm. Direct transmission through the grating (zero order) can be used to measure the spatial profile of the HHG beam. Figure 2.23 shows the measured HHG beam profile generated with Ne gas. The 1/e 2 diameter is shown for x- and y-cross sections measured at the center of the beam. The beam is slightly asymmetric and slightly non-gaussian. The CCD chip is not wide enough to capture the fully dispersed spectrum if placed directly in line with the grating, so the camera is mounted on a flexible bellows that allows for the camera to be moved laterally to the position of the first order dispersion. This position can be easily calculated using the diffraction equation, λ/d = sin(θ), where λ is the XUV wavelength and d is the grating spacing. For 93 ev light the angle is 7.6. The distance between the grating and the CCD camera is 35 cm, resulting in a lateral CCD displacement of 4.7 cm. The spectrometer resolution is not usually considered for this apparatus since it is only used as a diagnostic, however resolution is better than necessary to observe individual harmonic orders spaced 3 ev apart at around 90 ev. The CCD camera reads out in two dimensions, but it is often convenient to look at the spectrum integrated along the vertical (non-diffracted) axis in order to see total photon flux as a function of photon energy. Figure 2.24 shows the raw two dimensional HHG spectrum measured at first order. The absorption from the Si L-edge in the grating can be seen as a rapid drop in transmission at 100 ev. This edge can be used along with the Al L-edge around 73 ev to calibrate the photon energy

62 CHAPTER 2. EXPERIMENTAL APPARATUS mm 2 1 mm mm mm Figure 2.23: Measured zero order transmission of the HHG radiation generated in Ne gas, averaged over 10 scans with 1 s integration each scan. 1/e 2 diameters of the x- and y-cross sections are measured as 6.86 mm and 5.4 mm, respectively. The white dashed lines show the positions at which the line-outs were measured.

63 CHAPTER 2. EXPERIMENTAL APPARATUS mm Photon Energy (ev) Figure 2.24: First order dispersion spectrum of HHG XUV radiation generated in Ne gas, averaged over 10 scans with 1 s integration each scan. scale of the spectrometer.

64 49 Chapter 3 Surface plasmon assisted electron acceleration in photoemission from gold nanopillars 3.1 Introduction Among the most intriguing phenomena in nanoscale materials today is the coherent electronic excitation in metals known as the surface plasmon resonance (SPR). The SPR is a collective oscillation of conduction band electrons that typically occurs at optical frequencies in noble metals [46]. For a short amount of time, these electrons oscillate in phase and create a strongly enhanced electric field at the surface of the metal/vacuum or metal/dielectric interface [2, 1]. SPRs have enormous potential for applications in medicine, communications, and electronics [6, 7], most of which take advantage of the strongly enhanced electric field created by the plasmon at the metal surface. Techniques such as surface-enhanced Raman spectroscopy exploit this near field enhancement to allow the sensitive spectroscopic detection of single molecules [8]. One area of significant interest is the plasmon response to excitation by high intensity, ultrafast laser pulses. Lasers that generate such pulses are becoming increasingly common and have opened the door to studying new regimes of light/matter interactions. The goal of the present work is to investigate the interactions of laser-ionized photoelectrons with localized surface plasmon electric fields excited in a lithographically prepared nanostructured array. The use of a nanostructured surface is advantageous because the SPRs are excited directly by ultrafast laser pulses without requiring special excitation geometries often used in studies of plasmon enhanced photoemission from flat gold surfaces [58, 59, 60, 61, 62, 63, 64] or extremely sharp metal tips [66, 67]. By measuring photoelectron kinetic energy spectra and electron yields as a function of laser excitation intensity, we observe photoelectron kinetic energies tens of ev higher than expected based on the laser excitation intensity. A classical

65 CHAPTER 3. SURFACE PLASMON ELECTRON ACCELERATION 50 ~7 fs ~30 fs CM HCF Ti:Sapphire, 1 khz, 800 μj, 30 fs TOF Figure 3.1: Schematic of the experimental apparatus. 30 fs FWHM, 800 µj laser pulses are spectrally broadened in a gas-filled hollow-core fiber (HCF) and temporally compressed to 7 fs FWHM with a series of multilayer chirped mirrors (CM). The laser is focused onto the sample surface and photoelectrons are detected using a linear time-of-flight spectrometer (TOF). electron acceleration calculation is used to model the data and to determine the average field enhancement from the nanostructures. Implications for possible studies of plasmon-enhanced attosecond photoelectron streaking are also briefly discussed. 3.2 Experimental Apparatus The measurements use a few-cycle femtosecond, visible-infrared laser pulse to excite a SPR in a lithographically prepared gold nanopillar sample and to simultaneously ionize photoelectrons from the sample surface. Photoelectron kinetic energies are measured as a function of excitation intensity using a linear time-of-flight (TOF) electron spectrometer. A schematic of the experimental apparatus is shown in Figure 3.1. The apparatus consists of a Femtolasers Femtopower Compact Pro multi-pass amplified Ti:Sapphire laser system that produces 30 fs full-width at half-maximum (FWHM), 800 µj laser pulses at a repetition rate of 1 khz. A 1 m long, 250 µm inner diameter hollow core glass fiber filled with 1.9 Bar of Ne gas is used for spectral broadening through self-phase modulation followed by temporal compression with a series of negatively chirped mirrors to a pulse duration of 7 fs FWHM. The laser spectrum extends from 540 nm to 930 nm

66 CHAPTER 3. SURFACE PLASMON ELECTRON ACCELERATION 51 a) b) 250 nm Intensity Wavelength (nm) Figure 3.2: (a) Scanning electron microscope (SEM) image of the gold nanopillar array. (b) Dark-field scattering measurement of a single nanopillar from an identically prepared sample with a larger pitch to allow for measurement of a single particle. (1% level of intensity). The laser pulse is focused at grazing incidence, 75 to the sample surface normal, using a near-normal incidence spherical mirror with a high reflectivity gold coating and a 10 cm focal length to a spot size of approximately 60 µm (1/e 2 diameter). The grazing angle stretches the spot size along the direction of propagation to 300 µm. The sample is housed in vacuum at a pressure of < Torr. No steps were taken to clean the sample surface. Emitted photoelectrons are detected normal to the sample surface by the field-free TOF electron spectrometer using a micro-channel plate (MCP) detector. The acceptance angle of the spectrometer is 2 and the total flight length is 0.59 m. Signal pulses from the MCP are processed by an analog constant-fraction discriminator to correct for the MCP pulse height distribution and then counted with a digital multi-channel scaler with a 500 ps resolution. The energy resolution of the TOF spectrometer varies with electron kinetic energy, ranging from 11 mev for 5 ev electrons to 1 ev for 100 ev electrons. At the count rates present in this experiment, the probability of missing an electron count during the detection electronics pulse-pair resolution dead-time ranges from 0.2% at the lowest count rates to 10% at the highest count rates. p-polarized light is used throughout this experiment to excite SPRs normal to the sample surface and parallel to the TOF axis Nanopillar Sample The sample investigated consists of free-standing gold nanopillars attached to a 10 nm thick binding layer of chromium. A surface consisting of 12 nm of gold on top of 10 nm of chromium was coated with 300 nm of photoresist and then patterned using electron-beam lithography.

67 CHAPTER 3. SURFACE PLASMON ELECTRON ACCELERATION 52 The exposed photoresist was chemically removed and gold was then electroplated onto the surface. Finally, the unexposed photoresist was chemically removed and ion sputtering was used to remove the tops of the pillars and the Au plating base layer, leaving free-standing gold pillars on top of a conductive thin film of chromium. Individual nanopillars are cylindrical with a diameter of 100 nm and a height of 285 nm and are arranged in a cubic lattice with 250 nm pitch. The tops of the pillars become partially rounded during the plasma etching step. Figure 3.2a is a scanning electron microscope (SEM) image taken at 29.0 from the surface normal. The nanopillar shape and aspect ratio were chosen such that the SPR along the long axis of the nanopillar (parallel to the TOF axis) is resonant within the laser bandwidth. Dark-field scattering measurements of individual nanopillars from a sample with larger pitch (4 µm) but otherwise prepared identically show a broadband plasmon resonance (Figure 3.2b), centered near 700 nm, that is well overlapped with the laser bandwidth. Control experiments were performed on a commercially available 50 nm thick gold film coated onto a Si 111 wafer (Ted Pella #16012-G). The flatness of the Si wafer allows for a gold surface with only 2.5 nm root-mean-squared surface roughness, which is measured by atomic force microscopy. Because of a mismatch in momentum, laser photons cannot couple to a surface plasmon wave in a flat gold surface unless special excitation geometries such as the Kretschmann configuration are used[1, 55]. Under the experimental geometry presented here, no coupling should occur and plasmon enhanced effects should not be observed from this sample. Independent measurements of the damage threshold of the gold surface were made by raising the laser intensity to the point where photoelectron spectra collected at lower intensities were no longer reproducible. Measurements presented here are collected below the damage threshold. 3.3 Results and Discussion Photoelectron Spectra In order to determine the interaction of ionized photoelectrons with the surface plasmon field, photoelectron kinetic energy spectra are recorded as a function of the excitation laser intensity. The few-cycle laser pulse is used to excite the plasmon resonance and simultaneously inject photoelectrons by multiphoton photoemission into the enhanced plasmon electric field. The work function of polycrystalline gold ranges from 4.7 ev to 5.2 ev [83]. The broadband laser pulse has < intensity in the spectral range below 527 nm (half of 4.7 ev), therefore photoemission should require at least three laser photons to eject an electron into the continuum, even at the high energy side of the laser bandwidth. The excitation intensity is varied using a variable neutral density (ND) filter, the dispersion of which is pre-compensated by chirped mirrors. The laser pulse energy is measured at each intensity step and is used along with the pulse duration and the measured focal spot size to determine the intensity. Scanning the variable ND filter does not produce a detectable change in pulse

68 CHAPTER 3. SURFACE PLASMON ELECTRON ACCELERATION 53 a) b) Figure 3.3: (a) Photoelectron kinetic energy spectra taken from a flat gold surface as a function of excitation intensity. (b) Photoelectron spectra taken from the gold nanopillars at the same intensities as (a). Strong acceleration of photoelectrons to high kinetic energies is indicative of photoelectron emission in the presence of plasmon-enhanced electric fields. Because of the inability of photons to directly excite a SPR in flat gold, a minimal increase in kinetic energy is present in (a). duration, which is monitored by second-order interferometric autocorrelation. By decreasing or increasing the excitation intensity, the plasmon electron oscillation in the nanopillars can be driven more weakly or more strongly, respectively. We expect this change in field strength to result in varying degrees of acceleration experienced by photoelectrons injected into the plasmon field. Figure 3.3a shows a series of photoelectron spectra taken from the reference flat gold sample as a function of excitation intensity. Each spectrum is integrated over laser pulses. As previously noted, a plasmon oscillation cannot be directly excited on the flat gold surface by the laser because of the momentum mismatch between the laser photons and the surface plasmon resonance, therefore there should be no field enhancement. Figure 3.3b shows a similar series of photoelectron spectra taken from the gold nanopillar sample. As the laser intensity is increased, the maximum kinetic energy measured increases substantially and an increasingly strong secondary peak is formed between 10 ev and 40 ev. The missed electron counts due to the detector pulse-pair resolution do not significantly alter the shape of the spectral distribution when corrected for. The dramatically increased electron kinetic energy with increasing excitation intensity in the nanopillars compared to the minimal increase at the same intensities in the flat gold spectra strongly suggests an enhanced plasmon-fieldbased acceleration mechanism. To further investigate the details of this mechanism we consider both the ionization and the acceleration processes in the following sections.

69 CHAPTER 3. SURFACE PLASMON ELECTRON ACCELERATION 54 a) b) 3 α I α I 2 Figure 3.4: Log-log plot showing the total number of detected photoelectrons as a function of excitation intensity, I, for (a) the flat gold surface and (b) the gold nanopillar sample. While the flat gold surface demonstrates the expected third order multiphoton dependence, only a second-order dependence is observed in emission from the nanopillars Total Electron Emission Scaling with Laser Intensity To determine what effect the nanopillar SPR has on the photoelectron ionization process, a measurement is made of the total number of electrons detected as a function of the excitation intensity. Figure 3.4 shows the integrated photoelectron yield versus laser excitation intensity, I, for both samples on a log-log scale. The error bars are determined by the probability of missing electron counts during the detection electronics pulse-pair resolution dead-time. The data are the same as is shown in Figure 3.3 but with additional data points that are not displayed in Figure 3.3 for clarity. In both the nanopillar case and the flat gold case, a linear slope fits the observed trend, suggesting multiphoton ionization where the slope of such a fit results in a n th order intensity dependence, where n is the number of photons required to exceed the work function of the metal [62]. Figure 3.4a shows the measured photoionization intensity dependence for the flat gold surface. In the flat gold case the expected I 3 dependence is observed, indicating a three photon multiphoton ionization process and no plasmon enhancement. For the gold nanopillar sample, Figure 3.4b, an I 2 dependence is observed, corresponding to one fewer photon needed to exceed the work function than expected for multiphoton ionization with the laser pulse. Although the very weak intensity of the laser pulse below 527 nm could play some role, this is unlikely. An I n 1 dependence has been previously observed in multiphoton ionization from a gold surface in the presence of coupled surface and interface plasmon waves [62] and in localized plasmon hot spots on a Cu surface [84]. The effect is attributed to the promotion of valence band electrons to an electronically excited state via the SPR, giving the electrons an amount of energy, ω plasmon,

70 CHAPTER 3. SURFACE PLASMON ELECTRON ACCELERATION 55 a) b) Figure 3.5: (a) Spectra modeled from classical electron trajectory calculations (black lines) compared to the experimental data (symbols). Each trace is offset by one order of magnitude from the previous trace for clarity. In the model, multiphoton emission is followed by classical acceleration in an enhanced field. An average field enhancement of 32 brings the model in close agreement with the experimental data. The intensities shown in the legend are the enhanced intensity values, (I E 2 ), used for the calculation. The experimental data is the same as shown in Figure 3.3a. (b) The experimental data (symbols) compared to a range of modeled spectra calculated for average field enhancement factors from (shaded areas). Each trace and shaded area is offset by two orders of magnitude from the previous trace for clarity. above the Fermi level and permitting one fewer photon for multiphoton ionization. An overall two-photon power dependence can occur, depending on the relative magnitudes of the cross sections for promotion to the excited state and ionization. When compared to the I 3 dependence observed for the flat gold surface, this corroborates the concept that a plasmon enhancement occurs in the nanopillars and is responsible for the observed accelerations in Figure 3.3a Classical acceleration model The acceleration of the electrons in the plasmon field is modeled using a one dimensional classical electron trajectory calculation. In this calculation 1000 electrons are released into the enhanced electric field of an assumed 7 fs FWHM Gaussian laser pulse at every time step, which are spaced by 4.8 as, resulting in a total of 7.5 million electrons. The frequency of the enhanced field is assumed to be the same as the frequency of the laser pulse. A full calculation of multiphoton ionization from a metal surface is beyond the scope of this work; instead, initial electron kinetic energies are chosen randomly from a log-normal fit to the

71 CHAPTER 3. SURFACE PLASMON ELECTRON ACCELERATION 56 photoelectron spectra from the flat gold surface shown in Figure 3.3b. The amplitude of the log-normal fit is the only parameter varied for initial spectra at the different intensity values. After the initial release of electrons into the enhanced electric field, the position and velocity of each electron is calculated at each time step by integrating the classical equations of motion for a charged particle in an electric field. The final electron velocity as a function of ionization time, v f (t i ), is described by: v f (t i ) = v(t i ) + t i qe(t) m e dt (3.1) where v(t i ) is the initial electron velocity, q is the elementary charge, m e is the electron mass, and E(t) = E 0 σ 2π e (t a)2 /(2σ2) e i(ωt+φ) (3.2) where E 0 is the peak value of the enhanced electric field, σ = FWHM/(2 2 ln 2) with the FWHM laser pulse duration, a is the center of the Gaussian pulse, ω is the angular frequency of the electric field and φ is the carrier-envelope-phase (CEP) of the exciting laser pulse. The final electron velocity distribution as a function of electron ejection time is weighted by a temporal emission probability I(t) 2, corresponding to a second-order multiphoton ionization process. Electrons with negative final velocity are excluded from the resulting spectrum because they would not reach the detector. Because of the Gaussian spatial mode of the laser focus, not all ejected electrons will experience the peak intensity value. To account for this, the spectra presented here are constructed by integrating over individual spectra calculated at a range of intensity values over the spatial extent of the laser focus. The contributions from each spectrum are weighted according to the area illuminated by that intensity. In addition, these spectra are averaged for 5 values of the CEP over a range of 2π. The model is constructed as if the emission were from a flat surface (without nanostructures) with a uniform enhancement over the spatial profile of the laser pulse. In reality, the nanostructured surface will have an inhomogeneous field enhancement and ejected electrons will experience different fields. Additionally, the degree of coupling between localized plasmon modes is unknown at this time. The enhancement factor considered below represents an average of all of these possible inhomogeneities. The single free parameter in the calculation is therefore the average field enhancement factor due to the nanopillars, which is adjusted to give agreement between the observed and modeled photoelectron spectra. The same enhancement factor is applied to all of the spectra. The lifetime of the plasmon oscillation, phase-lag between the plasmon field and the exciting laser field, and surface recollision effects are not included in the calculation. In order to directly compare the modeled spectra to all of the experimental data, the absolute electron yields of all the modeled spectra were scaled by a single value. This value is the ratio between the integrated photoelectron yield of the spectrum measured at the highest excitation intensity and the integrated photoelectron yield of the spectrum calculated at the

72 CHAPTER 3. SURFACE PLASMON ELECTRON ACCELERATION 57 highest enhanced intensity. This scaling places the modeled traces on the same absolute scale as the measured data while preserving the relative scaling of the modeled spectra produced by the calculation. Figure 3.5a shows the experimental nanopillar data (symbols) compared to the modeled spectra (black lines), where a field enhancement of 32 times the experimentally used field strength is chosen. The uncertainty in the enhancement factor is estimated to be ±7 and is determined by qualitatively comparing spectra calculated at various enhancement values to the experimental data. Figure 3.5b shows the experimental data (symbols) compared to a range of spectra modeled with enhancement factors ranging from 25 to 39 (shaded areas). The calculated spectra qualitatively reproduce the main features of the experimental data. Post-ionization acceleration of photoelectrons in the enhanced electric field results in the shifting of electrons from the initial kinetic energy distribution to higher kinetic energies and the formation of a secondary maximum between 10 ev and 40 ev. The fact that the secondary maximum is stronger in the modeled spectra and offset by several ev from the experimental data may result from non-uniform acceleration of photoelectrons due to the inhomogeneity of the enhanced field across the nanostructured surface. In addition to the electron kinetic energies, the relative electron yields of the spectra modeled using second-order multiphoton emission at the enhanced excitation intensities match well to the experimentally observed yields. When combined with the evidence for plasmon-assisted multiphoton ionization described previously, the results of this model support a two step process of plasmon-enhanced multiphoton ionization followed by classical electron acceleration in a plasmon-enhanced field. In addition, surface rescattering effects are expected to contribute higher energy electrons to the spectra [85, 86], and their exclusion from this calculation may result in the failure of the model at the highest observed kinetic energies. Moreover, plasmon dephasing rates, which may be dependent on the amplitude of the launching field, have not been accounted for and could account for deviations of the data from the model as a function of intensity at higher electron kinetic energies. Since only an average enhancement factor is used, and because the highest kinetic energies derive from the highest field regions of the nanopillars, it is also not surprising that deviations at high kinetic energies are observed. Previous calculations of thin-film propagating plasmon-enhanced electron acceleration that include a more detailed description of the surface plasmon field produce similar results as this simple model [59, 68]. 3.4 Conclusions In conclusion, we observe photoelectron kinetic energies in photoemission from lithographically prepared gold nanopillars that are consistent with electron acceleration in electric fields with average strengths between 25 and 39 times higher than the experimentally used laser field strengths. Reference measurements from a flat gold surface do not produce such high electron kinetic energies at the same excitation intensities. The presence of a plasmon-

73 CHAPTER 3. SURFACE PLASMON ELECTRON ACCELERATION 58 induced field enhancement is further supported by analysis of the excitation intensity dependence of the total electron emission yield. Multiphoton emission is observed for both the flat gold surface and the nanopillar sample, with the expected three-photon ionization process for the flat gold, yet a two-photon ionization process from the nanopillars indicative of plasmon-enhanced multiphoton ionization. Classical electron trajectory calculations support the concept that the electrons are first ionized via multiphoton ionization and then subsequently accelerated in the enhanced electric field of the nanopillars to high kinetic energies. These results provide the basis for the possibility of SPR-enhanced attosecond streaking from localized plasmon resonances in nanostructured surfaces. Such a concept has been explored theoretically for both nanostructures [87] and roughened metal surfaces [88]. In such an experiment, the attosecond streak camera scheme [41] would be modified to utilize the plasmon-enhanced electric field as a probe instead of the intense few-cycle laser pulse. Electrons are emitted by an isolated attosecond pulse in the presence of a plasmon field that has been excited by the femtosecond pump pulse. As the electrons are emitted they will then be streaked by the sum of the plasmon field and the laser field. By taking advantage of the SPR field enhancement, it may be possible to reduce the contribution of the laser field itself sufficiently that information on the SPR lifetime and dynamics of the oscillating SPR electrons can be obtained directly from the streak trace. Given the observation of a significant field enhancement and multi-ev streaking of photoelectrons presented here, we expect that attosecond streaking studies of plasmon dynamics in metal nanostructures is possible.

74 59 Chapter 4 Application of attosecond streaking to condensed matter targets 4.1 Overview The extension of attosecond streaking from the gas phase to condensed matter systems is desirable due to the fact that ultrafast, correlated electron dynamics occur in many solids, from superconductors to plasmonic materials. Many of these systems are not completely understood, and a more thorough understanding of the electron dynamics that lead to their unique properties is important. As has been described previously in this dissertation, one system of particular interest is the surface plasmon resonance (SPR). Unfortunately, condensed matter presents many challenges that are not present in the gas phase, and the direct application of existing attosecond techniques is not always possible. This chapter will describe the efforts to adapt attosecond technology to solids and the challenges and results associated with that effort. 4.2 Previous experiments in the literature Thus far, only one experiment using isolated attosecond pulses to probe electron dynamics in a condensed matter material has been published [40]. In this experiment, Cavalieri and coworkers used isolated attosecond pulses and the streak-field technique with 800 nm, IR laser pulses to measure a time delay of 110 ± 70 as between the photoemission of electrons from the delocalized valence band of a W(110) single crystal and electrons from the 4f state of the same. Figure 4.1 shows the streaking data from their experiment. Panel (a) shows the photoelectron spectra collected at two different time delays between the XUV attosecond pulses and the IR laser pulses. The blue line is the raw TOF data collected far from zero delay of the XUV and IR pulses (indicated by position (1) in panel (b)), where zero delay is define as temporal overlap of the maxima of the pulse envelopes. In this case negative time

75 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 60 Figure 4.1: (a) Photoelectron spectra collected at different time delays between the XUV attosecond pulses and the IR laser pulses. The positions of the delays are shown as dashed lines in (b). Peaks from the tungsten valence band (83 ev) and 4f state (56 ev) can be seen in a spectrum taken far from zero time delay (blue line), and the Fermi level is denoted by E f. This same spectrum is also shown after subtraction of the large multiphoton background emission and numerical smoothing (red line). These peaks broaden out from streaking by the IR laser field in the spectrum measured at zero time-delay (black line). (b) The full streaking spectrogram after subtraction of the multiphoton background emission.

76 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 61 delays correspond to arrival of the IR pulses before the XUV pulses. Emission peaks from the valence band (83 ev) and the 4f state (56 ev) can be clearly seen. In addition to these peaks, a large multiphoton background from the IR laser pulses can be seen starting at around 50 ev and rising rapidly to below 30 ev (this background is discussed further in Section 4.3.1). The red line is the same data after subtraction of the multiphoton signal and numerical smoothing. The black line is data taken near zero delay time, corresponding to position (2) in panel (b), after multiphoton background subtraction and numerical smoothing. When compared to the peaks in the spectrum measured far from zero delay (position (1)), the peaks corresponding to the tungsten valence band and 4f state are clearly broadened (with the resulting decrease in amplitude) by the presence of the IR laser field. The streaking to higher kinetic energies is particularly clear from the increase in signal level above the Fermi level (E f ). Figure 4.1 shows the full streaking spectrogram constructed from individual spectra integrated for 60 s at each delay time and after subtraction of the multiphoton background. Figure 4.2 shows the two streak traces from Figure 4.1b after cubic-spline interpolation. The conduction band is shown in the upper panel and the 4f state is shown in the lower panel. The region from 65 ev to 83 ev had been removed so that the streak traces can be compared more directly. A small temporal shift between the two streak traces is highlighted by the dashed white lines. This delay is quantified by performing center-of-mass analysis on the streak traces, shown in Figure 4.2b. These center-of-mass plots are used to determine the delay of 110±70 as. The delay is suggested to originate from two factors. The first is that the final state bands of the valence electrons are calculated, using a static band structure, to have stronger dispersion than those of the 4f electrons. This results in a lower effective mass and thus a larger group velocity for the valence band electrons, ultimately leading to a faster escape time from the surface. The second factor is the fact that the 4f electrons have a longer inelastic mean free path (IMFP) than the valence electrons and thus originate 1 Å deeper in the surface. The longer average distance of travel caused by the increased IMFP and the slower velocity of the 4f electrons are said to result in the temporal delay observed in the streak traces. Treating photoemission in the presence of a streaking laser field from a complex condensed matter band structure is extremely complicated, however, and three alternative theoretical treatments have already emerged. Kazansky and Echenique have expanded from the original calculation to a time-dependent calculation and found that the static band structure approximation does not hold [89]. They conclude that the group velocity, which results from interference of electron wavepackets scattered from atoms in the lattice, does not have time to form in the attosecond timescale (because not enough scattering events occur) and thus cannot be the reason for the observed delay. Instead, they find that the time delay results mainly from the difference in the initial electronic states, localized for the 4f electrons and delocalized for the valence band electrons. Zhang and Thumm use a similar model to that of Kazansky [90], but they allow the few-cycle IR streak field to penetrate into the surface, which Kazansky does not. In the presence of the streak field, they find that the delay results from interference between 4f electrons emitted from different atomic layers of the solid.

77 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 62 Figure 4.2: Streaking data from a W(110) crystal from Ref. [40]. (a) Streak traces following cubic-spline interpolation of photoemission from the valence band (upper panel) and the 4f state (lower panel) from a W(110) single crystal. A very small time delay between the two streak traces is highlighted by the dashed white lines. (b) Center-of-mass plots for the valence and 4f streak traces in (a). The resulting delay is 110±70 as, where the error results from the calculation of the center-of-mass.

78 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 63 This interference does not occur for the valence band because it is considered completely delocalized in their model. Finally, Lemell and coworkers used a classical transport theory to model the results [91]. They find that the delay results from a combination of the larger emission depth of the 4f electrons and also slowing of some of the valence electrons by inelastic scattering, resulting in final valence electron kinetic energies commensurate with the kinetic energy of the 4f electrons but at a delayed time. No consensus explanation has yet been formed. 4.3 Streaking results from a W(110) single crystal In order to verify the production of isolated attosecond pulses by the apparatus constructed in this dissertation, streaking experiments from a W(110) single crystal are demonstrated and discussed below Photoelectron background emission One of the primary challenges of XUV photoelectron spectroscopy from condensed matter materials as compared to gas phase studies is the presence of a large photoelectron background. There are two major sources of photoelectron background present in these experiments: inelastically scattered electrons resulting from XUV ionization deeper in the material than the electron inelastic mean free path and multiphoton ionization by the fewcycle, 800 nm laser pulses. The first source of background results from the fact that the 90 ev photons of the attosecond pulse will penetrate 3 nm into the surface (the distance is measured at the point where the intensity is 1/e of the initial intensity) of the condensed matter target (measured normal to the surface) [80]. Electrons will be generated all along the path of the XUV light, yet the electron escape depth (equivalent to the electron IMFP for detection normal to the surface) is only 0.4 nm for 90 ev kinetic energy electrons [81]. This value can be determined from the so-called universal curve for electron IMFP, shown in Figure 4.3, which applies to photoemission from many different solids. Figure 4.4 shows a schematic demonstrating the process. This results in a very large background of inelastically scattered electrons and a relatively small amount of direct photoemission. To calculate the percentage of direct photoemission as a function of the total emitted electrons, one must integrate over both the ionization probability and the inelastic scattering probability. This results in the expression: λ I = I 0 (4.1) λ + d p where I 0 is the number of photons that penetrate the surface (this assumes that every photon will produce one electron), λ is the inelastic mean free path and d p is the attenuation length

79 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 64 Figure 4.3: The universal curve for electron inelastic mean free path (IMFP) taken from Ref. [81]. For electrons with kinetic energy around 90 ev, the IMFP is only 0.4 nm. Because this length is shorter than the penetration depth of XUV radiation into the sample, a background of inelastically scattered electrons results. XUV Emitted Electrons 1 1/e 15 deg e- 1/e 1 z ~0.4 nm Probability of electron escape without scattering ~3 nm e- z Figure 4.4: A schematic illustrating the source of the large inelastically scattered photoelectron background resulting from XUV photoemission. The 93 ev light penetrates 3 nm normal to the surface (z-axis), releasing electrons all along the path of the light, while the IMFP is only 0.4 nm for 90 ev electron kinetic energy. Electrons released deeper than this have very little probability of escaping the surface without inelastically scattering.

80 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 65 a) 0.1 b) 0.1 Electron Counts/Second Electrons emitted without scattering Electron Counts/Second Detector saturation Kinetic Energy (ev) Kinetic Energy (ev) Figure 4.5: (a) Photoelectron kinetic energy spectrum from a W(110) single crystal surface ionized by HHG generated XUV pulses centered at 93 ev. The peak centered at 86 ev is from electrons that escape from the surface without scattering (indicated approximately by grey shaded area), while the large low energy background results from electrons that are inelastically scattered within the metal. (b) Comparison of XUV only photoelectron emission from W(110) (black line) to photoemission from the XUV plus the few-cycle, 800 nm laser pulses (red line). Instability in the HHG flux has led to a slight decrease in overall signal between the two measurements. The two pulses are positioned at zero time overlap and the intensity of the 800 nm laser pulses is typical for a streaking experiment. The broad peak centered at 34 ev is the result of multiphoton emission by the 800 nm laser pulses and is saturating the detector below 34 ev (indicated by the arrow). In addition, increased amplitude above 95 ev shows streaking of electrons from the valence band peak to higher kinetic energies. of the light, measured normal to the surface. For the parameters given above, this results in only 11% of total electron emission escaping the surface without inelastically scattering. Figure 4.5a shows an example of ionization from a W(110) crystal surface using only the HHG generated XUV light centered at 93 ev. The spectrum is integrated over laser pulses with an average of 3311 electrons detected per second. The peak at approximately 86 ev results from direct valence band photoemission by the XUV pulse while the large low kinetic energy signal is the result of the inelastically scattered electrons. The integrated number of photoelectron counts under a Lorentzian peak fit to the XUV peak is 6% of the total number of detected electrons. The second major source of photoelectron background that is relevant for attosecond streaking experiments from condensed matter is the multiphoton ionization induced by the few-cycle, 800 nm laser pulses. While multiphoton background does exist in streaking from gas-phase targets, it is usually not a problem because of the high ionization potentials of gas phase atoms and molecules. In comparison, most solids have relatively low work functions

81 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 66 that can permit a large amount of multiphoton ionization at typical streak-field intensities. For example, the ionization potential of Ne atoms is 21.6 ev, while the work function of a W(110) crystal surface is only 5.5 ev [92]. This means that fourteen 1.55 ev, 800 nm photons are needed to ionize a Ne atom while only four of the same photons are necessary to ionize from the tungsten surface. Figure 4.5b demonstrates the multiphoton background produced by the 800 nm, few-cycle laser pulses in a typical streaking measurement. The black line is the same data as shown in Figure 4.5a, while the red line is measured with the 800 nm laser pulse overlapping with the XUV pulse at zero time delay. Instability in the HHG flux has led to a slight decrease in overall signal between the two measurements. The peak from direct XUV emission at 86 ev has been broadened out to both lower and higher kinetic energies (the CEP is not locked in this measurement, so electrons are streaked to both higher and lower kinetic energies within the integration period). The peak centered at 34 ev is the result of multiphoton ionization from the few-cycle, 800 nm laser pulses. Multiphoton emission is high enough that the detector is saturated below 34 ev, indicated by the arrow in the figure. Saturation only occurs when the 800 nm laser pulse is present (red line), not in the XUV only spectrum (black line) Demonstration of attosecond streaking Figure 4.6 shows the first demonstration of attosecond streaking from the apparatus developed in this dissertation (data collected by Joseph Robinson, with whom the system was jointly developed). The spectrogram is constructed from a series of photoelectron spectra collected from a W(110) single crystal at varying time delays between the attosecond XUV pulses and the IR laser pulses. Negative time delays represent time at which the attosecond XUV pulses arrive before the IR laser pulses. Each time step (200 as) is integrated over laser pulses. The sample is mounted at the Brewster s angle of 15 grazing incidence angle in order to minimize the intensity of the reflected 800 nm laser beam [93]. The streakfield intensity is estimated to be W/cm 2 based on measurements of the pulse energy, pulse duration and focal spot size. The data is normalized to the integrated electron yield between 75 ev and 110 ev at each time step to account for fluctuations in the HHG yield. No smoothing or other processing has been applied. In this spectrogram, only the valence band is clearly seen because the emission from the 4f state is suppressed by surface contamination. The apparatus currently does not support ultra-high vacuum (UHV) conditions (typically described as < Torr), therefore cleaning and preparation of the sample surface would have very little effect as the surface would be quickly contaminated again. In the supplemental material of Ref. [40], it is stated that, at a pressure of < 10 9 mbar, the 4f photoelectron peak is only present for 3 hours after heating in front of an oxygen doser to 1400 K followed by repeated flash heating to above 2200 K. After this time the surface contamination becomes enough that 4f electrons cannot effectively escape the surface without being inelastically scattered. New equipment

82 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 67 Figure 4.6: First demonstration of attosecond streaking from the apparatus developed in this dissertation. The spectrogram is constructed from a series of photoelectron spectra collected from a W(110) single crystal at varying time delays between the attosecond XUV and the IR laser pulses. Negative time delays represent the time at which the attosecond XUV pulses arrive before the IR laser pulses. Each time step (200 as) is integrated over laser pulses.

83 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 68 Figure 4.7: Spectral centroid analysis of the streaking trace presented in Figure 4.6. The centroid is calculated from data between 70 ev and 110 ev and clearly demonstrates the sub-optical-cycle resolution of the streaking spectrogram. currently being constructed in the Kaindl laboratory will allow future experiments to achieve the proper surface conditions. Spectral centroid analysis is used to more clearly visualize the modulation of the kinetic energy distribution. The centroid of the kinetic energy spectrum measured between 70 ev and 110 ev is shown in Figure 4.7 and is calculated according to: 110eV 70eV Centroid = E kc(e k ) 110eV 70eV C(E (4.2) k) where E k is the kinetic energy and C(E k ) is the corresponding number of electron counts at each kinetic energy value. The centroid of the kinetic energy distribution is clearly modulated as a function of the delay between the attosecond XUV pulse and the few-cycle IR pulse. The expected period for an 800 nm laser field is 2.66 fs, however the average peak-to-peak spacing in Figure 4.7 is 3.15 fs, corresponding to a wavelength of 945 nm. While this wavelength is contained in the laser pulse bandwidth, the discrepancy is surprising. One possible source of error is spatial chirp. Longer wavelength light will diverge at a faster rate than shorter wavelength light. Due to the split mirror geometry, the outer portion of the beam is used as the streak field and the inner portion is removed by the Zr filter. It is possible that a spatial chirp results in a streak-field wavelength that is longer than expected. Another possible source of error is drift in the piezoelectric-driven translation stage that controls the

84 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 69 temporal delay between the XUV attosecond pulse and the few-cycle laser pulse. The effect of such a drift is unknown and requires further investigation. One technique that has been useful in attosecond gas-phase experiments is the pulse reconstruction algorithm known as frequency-resolved optical gating for complete reconstruction of attosecond bursts (FROG-CRAB) [94]. The technique can be used to reconstruct pulse duration and phase information for both the isolated attosecond pulses and the fewcycle streak field laser pulses used in a streaking experiment. Unfortunately, the central momentum approximation used in the FROG-CRAB technique fails if the electrons ejected by the attosecond pulses are not ionized from a well-defined state. In the case of W(110), the valence band is nearly 10 ev wide [91, 82]. Beyond this, the strong background signal from both the XUV-induced inelastic scattering and the IR-laser-induced multiphoton emission cause deterioration in the performance of the FROG-CRAB algorithm. Because of this, it will be desirable in future experiments to have the capability of performing gas phase measurements in the same apparatus as the condensed matter measurements. With this capability, the isolated attosecond pulse could be characterized in the gas phase and then used for a condensed matter experiment. 4.4 Streaking results from amorphous Cr thin film In addition to streaking from a tungsten single crystal surface, streaking from an amorphous chromium thin film has also been observed using the apparatus constructed in this dissertation. As mentioned above, the only previous isolated attosecond pulse measurement from a condensed matter surface used a single crystal surface with Brewster s angle incidence of the 800 nm laser pulses (minimizing the reflected beam). It was thus unknown how streaking from a strongly reflecting thin film may behave. Cavalieri has suggested that the phaseshifted reflection of the 800 nm laser pulses may affect the streaking results [95]. Here we have shown that streaking from an amorphous thin film with a strongly reflected beam is possible. Figure 4.8 shows streak traces taken with isolated attosecond pulses from (a) W(110) and (b) a 10 nm thick film of amorphous Cr. The spectra have both been normalized for fluctuations in HHG yield and numerically smoothed. The two samples are mounted sideby-side in the vacuum chamber and the measurements were taken on the same day. The samples are both mounted at a 15 grazing incidence angle, which is the Brewster s angle for tungsten but not for chromium. Each trace is constructed from a series of scans integrated for laser pulses and with time steps of 200 as. Because of the noise present in this experiment, spectral centroid analysis does not provide a clear picture of the streaking trace. To more clearly visualize the photoelectron streaking in the two spectrograms, Figure 4.9 shows the summation of the photoelectron yield of each streak trace between 93 ev and 110 ev at each point along the time axis. Sub-optical-cycle resolution of the streaking laser field is clearly visible in both spectrograms. As opposed to Figure 4.7, the expected

85 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 70 a) b) Kinetic Energy (ev) Kinetic Energy (ev) Time Delay (fs) Time Delay (fs) Figure 4.8: Comparison of attosecond streaking from (a) a W(110) single crystal and (b) a 10 nm thick amorphous chromium thin film. The 800 nm streak field intensity was not the same in both measurements, which accounts for the different amounts of streaking in the kinetic energy domain.

86 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 71 a) b) W(110) Cr Thin FIlm Figure 4.9: Summation of the photoelectron yield between 93 ev and 100 ev for (a) a W(110) single crystal surface and (b) a 10 nm thick amorphous chromium thin film. Sub-optical-cycle resolution is clearly visible in both spectrograms. periodicity of 2.6 fs (corresponding to an 800 nm streak-field wavelength) is observed in both streak traces in Figure Surface plasmon enhanced attosecond spectroscopy The contents of this section have been published in Chemical Physics Letters, 463 (2008) 11-24, Ref. [87]. In this section, we propose and discuss a novel spectroscopic technique that directly accesses light-induced potentials in nanoparticles and molecules with sub-cycle precision using laser light as a pump and the attosecond pulse as a probe. A laser pulse induces a microscopic charge displacement in a quantum system and the time-delayed attosecond pulse ionizes an electron that on its way out samples the charge-migration-induced local electric potential of the initial bound-state orbital of the electron. One phenomenon of particular interest, chosen here as an example to illustrate the method, is the surface plasmon resonance (SPR) in metal nanoparticles. The SPR is a collective oscillation of the conduction band electrons that exists at optical frequencies for the noble metals [46]. The period of these oscillations is thus only a few femtoseconds, and the lifetime of the coherent motion is on the order of 10 fs [50, 48, 47, 51]. By using the attosecond pulse as the probe, we will be able to directly observe the charge oscillation of the nanoparticle by measuring its surface potential in real time. Instead of using the electric field of the laser pulse to provide temporal resolution by streaking, the laser-driven sub-cycle-oscillating dipole field of the surface plasmon in the vicinity of the particle is strong enough to lead to acceleration and deceleration of emitted electrons. The SPR is excited using the optical laser pulse and the attosecond pulse intro-

87 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 72 duced at a variable time delay to ionize an electron near the nanoparticle surface. The kinetic energy of this electron measured far away from the particle is the sum of the excess energy (attosecond photon energy minus ionization potential) and the negative of the surface potential (due to the negative electron charge) that results from the transiently-induced electric dipole of the particle. The soft x-ray attosecond pulse must have a sufficiently high photon energy to ensure the initial kinetic energy of the electron is great enough that it will escape the plasmon electric dipole field in less than one half cycle, preventing cycle-averaging of the electron acceleration in the surrounding dipole field. By measuring the photoelectron energy distribution as a function of time delay between the attosecond pulse and the laser pulse, the transient surface potentials, and thus the microscopic dipole response of the particle, will be mapped out with sub-femtosecond precision. In the case of a spherical nanoparticle, the SPR can be well-modeled by use of the dipole approximation for a sphere embedded in a dielectric medium [96]. In this approximation, the surface potential is: V = E r cos θ ɛ ɛ m ɛ + 2ɛ m (4.3) where E is the electric field strength of the driving pulse, r is the radius of the particle, θ is the angle of emission with respect to the laser polarization, ɛ is the complex dielectric constant of the sphere, and ɛ m is the dielectric constant of the embedding medium. In the calculation shown below, vacuum is assumed as the embedding medium, giving ɛ m = 1. Experimentally obtained frequency-dependent dielectric data for gold are used [97], resulting in a plasmon decay time of 9.3±0.9 fs. The particle size is chosen as r = 60 nm. The SPR is driven on resonance with a 6 fs-fwhm Gaussian pulse centered at 530-nm wavelength with an intensity of W/cm 2 intensity, orders of magnitude less than typical streak field intensities required for atomic species [36, 41, 12], because of the large polarizability of the nanoparticle that enhances the electric field near its surface. The surface potential is calculated as a function of time and a spherical emission probability distribution is considered for the electrons. The resulting photoelectron spectrum is convoluted with a 3.6 ev FWHM, 500 as transform-limited probe pulse to give a time-dependent photoelectron spectrum (model simulations shown in Fig. 4.5). If the time-dependent field of the laser pulse is precisely known (which can, for instance, be accomplished by a conventional streak-field measurement), the full spectroscopic set of amplitude and phase information is available for both the dipole response and the laser field. After Fourier-transforming this data to the spectral domain, a full reconstruction of the frequency-dependent dipole response (resonance curve) is then possible by dividing the complex-valued spectrum of the dipole potential by the complex-valued spectrum of the laser field. One chemically attractive application of this spectroscopic technique is to probe the dynamics of the SPR when different molecules are adsorbed on the surface of the particle. In techniques such as surface-enhanced Raman spectroscopy, the amount of signal enhancement achieved is directly proportional to the coherence time of the plasmon oscillation [98, 15, 51].

88 CHAPTER 4. CONDENSED MATTER ATTOSECOND STREAKING 73 a) TOF vo e- F surface E kin E 0 (ev) b) Time (fs) 2x Electric Field Strength (V/m) Figure 4.10: Simulation of a proposed measurement of field-induced attosecond time-resolved dipole potentials. a) Schematic of the dipolar charge distribution induced in a metal nanoparticle by the exciting laser field. Electrons are freed by the attosecond pulse with an initial velocity v 0 and sample the induced electric field, experiencing a force F surface, before they are detected by a time-of-flight spectrometer with a small collection angle of 15. b) Simulated time-dependent photoelectron kinetic energy spectrum as a function of time-delay between the attosecond and the 530-nm laser pulse, where E 0 = v 2 0/2 is the kinetic energy in the absence of the plasmon excitation. A 6 fs laser pulse (white line) at W/cm 2 intensity excites the plasmon which is then probed by a time-delayed 500 as pulse. A temporal broadening of the dipole potential response function mapped out by the intensity maxima of the photoelectron spectral distributions compared to the driving pulse shows the finite lifetime of the plasmon resonance (sustained dipole oscillations at late times after the driving laser pulse is over). In addition to resolving the decay time of the plasmon resonance, individual plasmon oscillations are observed with sub-cycle resolution, permitting the possibility to unravel nonlinear dynamics.