high-resolution two-laser spectroscopy setup and its applications to solid state

(1)

high-resolution two-laser spectroscopy setup and its applications to solid state

devices

Supervisors Prof. dr. ir. C. H. van der Wal Prof dr. ir. B. J. van Wees dr. ir. J. P. de Jong T. S. Ghiasi, MSc Research group Physics of Nanodevices Student

F. A. van Zwol

Date Monday 23^rd October, 2017 Credits 60 ECTS

Master thesis

Zernike Institute for Advanced Materials

(2)

Abstract

Optical methods can be used to find properties of solid state materials like the band structure, impurities, excitonic properties and information about lattice vibrations of these materials.

In this thesis, we build a high resolution (< 1 MHz linewidth and < 10 MHz drift) two-laser spectroscopic setup which can be used to study these properties. We use the D⁰-D⁰X quantum spin system on shallow Si-doped gallium arsenide to confirm proper functioning of the setup.

We obtain single-and two-laser scans and show the two-photon resonance effect of coherent population trapping at a magnetic field of about 1 T. We also investigated the nuclear spin relaxation time and polarized the nuclear spin to increase this time, a main factor of the low spin coherence time of 2 ns for the quantum superposition state in this GaAs system. We then modified the setup for single-laser photocurrent spectroscopy and obtain photocurrent spectra on M oSe² field effect transistors of a thickness varying from monolayer to bulk. We identified the A⁰ exciton transition at 1.59 eV and a spin-forbidden dark exciton optical transition A^D0

at 1.62 eV from the photocurrent spectra of monolayer M oSe². The bandgap dependence on thickness of M oSe² is characterized, ranging from 1.54 eV for the monolayer to 1.46 eV for the thicker TMDC-layers.

(3)

Thanks to prof. dr. ir. C. H. van der Wal for the research opportunity, guidance and feedback.

Thanks to dr. ir. J. P. de Jong for the clear feedback and daily guidance.

Thanks to prof. dr. ir. B. J. van Wees for the nice research group and interesting conversations.

Thanks to T. S. Ghiasi, MSc. for your patience, guidance and help in writing this thesis.

Thanks to dr. ir. J. Quereda Bernabeu. For helping a lot and answering questions.

Thanks to ir. T. Bosma for the occasional feedback and helping with questions.

Thanks to the Quantum Devices/FND-research group for the great time during the project.

(4)

Chapter 1 Introduction

Optical methods can be used to find properties of solid state materials like the band structure, impurities, excitonic properties and information about lattice vibrations of these materials [1].

In this thesis, we build a versatile and easy to use two-laser spectroscopic setup and apply it to two different systems. We investigate optical transitions and two-photon effects on the D⁰X system in gallium arsenide and we explore optoelectronic properties of two-dimensional transition metal dichalcogenides.

Investigating the solid-state D⁰-D⁰X quantum system on gallium arsenide is used as the first test case of the setup, because a lot of experience and data of this system has already been gained by our group [2]. The lasers are used to address ensembles of electron-spins in our gallium arsenide sample, which has been shallow doped with silicon atoms. A neutral silicon atom, which can be seen as an artificial hydrogen atom called D⁰, is created in the sample.

The silicon atom becomes neutral because donor electrons are localized at the silicon atoms when the temperature of the sample drops below the ionization temperature. A superposition of the two spin ground states of this system can be created using two lasers, which makes the system applicable as a qubit. An electron hole pair, also localized at the silicon atom, called D⁰X supplies exited states for the transfer of the electrons [3].

The goal of past research on this system was to increase the spin coherence time as much as possible, because the quality of quantum memory systems is determined by the dephasing time of the quantum superposition state. A problem is that both gallium and arsenide have a non-zero nuclear spin, which decreases the spin coherence time due to noise introduced to the electron spin by the nuclear spin. We confirm proper working of the two-laser setup by driving optical transitions of the system and showing two-photon resonance effects like coherent population trapping (CPT), phenomenon where the system is trapped in a superposition of two spin states. The coherence time of this superposition state determines the quality of the quantum memory qubit and is largely influenced by the nuclear spin. We measure the nuclear spin diffusion of the sample and polarize the nuclear spin to influence the spin coher-

(7)

ence time. The setup is also used to answer open questions about the previous project next to confirming proper working of the setup.

In the second part, we use the setup to perform photocurrent spectroscopy on TMDC field- effect transistors (FETs). Photocurrent spectroscopy is an interesting technique for characterizing materials, because the response of the material is measured electrically instead of via photoluminescence. Phenomena that are invisible in the photoluminescence spectrum can thus be characterized by this technique. TMDCs have been chosen as a material because the bandgap of these materials is around 1 eV, which gives them interesting optical properties [4] [5] [6]. For our project, we fabricate FETs based on M oSe² because of its high spin orbit coupling, which can have applications in quantum computing due to the spin split bands [4].

Many electronic and optical properties of this group of materials are still unknown, and can be characterized using photocurrent spectroscopy [7]. We obtained photocurrent spectra of a FET with a monolayer M oSe2-crystal and characterized excitonic transitions. A very interesting result we found was a spin-forbidden dark transition, which should only be very weak but is strong in our case. We also characterized the bandgap size dependence on the number of M oSe2 crystal layers of the FETs.

A versatile new two-laser setup has thus been built which is capable of two-laser spectroscopy and characterizing TMDC FETs using photocurrent spectroscopy. Several ideas and improvements from previous setups have been considered as well as new ideas for building the two-laser setup.

In this project, we address the following questions:

1. What is the best approach for designing a complete computer-controlled instrumental two-laser setup?

2. Can we confirm proper functioning of this setup by doing measurements on the D⁰- D⁰X system in GaAs? Can we use it for addressing open questions in this project?

3. Can we apply this setup for studying spectroscopic properties of TMDC FET devices?

This thesis will explain how the two-laser spectroscopic setup has been built and show its applications on GaAs and M oSe² FET devices. Chapter 2 shows the technical details of the setup, while chapter 3 shows the first application of the setup on the D⁰X system in GaAs.

This chapter confirms that the setup works very well compared to the old setup. Chapter 4 shows the characterization of the exitonic transitions of M oSe2 FET devices. Next to the expected exciton and trion transitions, a forbidden transition is also found.

(8)

Chapter 2 Experimental setup

Outline

We present a two-laser spectroscopic setup suitable for showing coherent population trapping (CPT) on GaAs in this chapter. The chapter contains an overview of the components that we choose for this setup, as well as information about the various interfaces, settings and control software we used. It also contains an overview of how all these components and connections work together in the final design of the setup. Design Choices and specifications have been checked in practice against a set of requirements set by previous experiments on GaAs and suggestions for improvements based on these experiments. Additional requirements for performing experiments on transition metal dichalcogenides are discussed at the end of the chapter.

To summarize, we are looking for what the best approach for designing a complete computer- controlled instrumental two-laser setup is.

2.1 Setup requirements for two-laser spectroscopy on GaAs

We build the two-laser setup based on predefined requirements based on previous experiments on GaAs and literature, as well as requirements that arose while working on the setup.

Because we want to perform two-laser spectroscopy on gallium arsenide with the setup, some of the requirements below should be achieved for two lasers simultaneously. We made the following list of requirements the setup should satisfy:

1. The new two-laser setup should be able to do a variety of solid-state physics experiments.

Therefore, the laser should have a wavelength in the near infrared region (650-1050 nm), where a lot of optical centers of various materials are found.

2. The laser setup should be able to show effects like coherent population trapping (CPT),

(9)

which has a width of 100 MHz in optical frequency for a gallium arsenide mesoscopic spin ensemble [8]. The linewidth of the laser should be at least orders of magnitudes lower than this, so the upper bound for the linewidth of the laser is 1 MHz.

The long term drift of the laser wavelength should also be below 10 MHz, to ensure that a certain optical transition remains driven, while scanning or pumping with the other laser.

3. The most important requirement compared to the old setup is that the setup should also be mainly computer controlled. The first reason for this is that this makes it possible to write software to manage the underlying settings and make it easy for people without much experience with optical experiments to do measurements with the setup. The second reason is that it also speeds up experiments performed with the setup because data can be obtained unattended and consistently.

4. There is also a requirement for the power of the laser. The laser power should be up to a few milliwatts at the sample to prevent optical pumping and dark states. Another requirement on the power is that it should be stable when reaching the sample. The fluctuations should remain under 1 percent to achieve reproducible results, deduced from simulations in reference [8].

5. The optical transitions we want to address in the experiments are polarization dependent.

This is due to the fact that the coupling between the incoming light and the optical transition is sensitive to the angle of incidence of the light. This is described by the Rabi frequency:

Ωij =

d~ij· ~E0

~ (2.1)

Here, ~dij denotes the dipole moment of the material and ~E0the electric field vector of the incoming light. Adjustable polarizers should thus be used in the setup, so the light can be polarized. The polarization angle of the light should also not change unpredictably while propagating through the setup, so polarization maintaining single-mode fibers should be used where possible.

6. And lastly, measurements should not take extremely long, but the data should also be off sufficient quality. Based on the dynamics of the studied systems and the available equipment, a target speed for taking data points of 10 Hz is chosen as a good compromise between the number of data points taken and duration of measurements.

Now, we estimate a good scan rate to achieve this target rate of taking data points. A typical overview scan is 2 to 3 THz wide (4-6 nm). This is because we want to look at optical transition between excitons within the band gap of GaAs. The D⁰ ground state lies 5.8 meV below the conduction band of GaAs [9].

(10)

2.2. SETUP DESIGN OVERVIEW

Small details like CPT that we want to see are 100 MHz wide (can be calculated by the Lindblad equation [8]). The 2 THz overview scan should be taken in minutes, which means that a fast scan should be done at an optical scan rate of > 10 GHz/s, while the smaller scans to find CPT peaks should be done at an optical scan rate below 100 MHz/s.

The setup should support both speeds to allow for both types of scans.

Note that the 100 MHz width of the CPT peaks is only true for the D⁰ - D0^X system in gallium arsenide. The width of the CPT peak can be down to a few MHz in width in other systems [10], but we don’t aim for this in the two-laser setup.

2.2 Setup design overview

PC

Optodac (digital to analog converter)

ICE box 1 IP adress:

192.168.1.221

ICE box 2 IP adress:

192.168.1.222

NI-GPIB Interface

Shu!ers DAC channel 3 and 4

Noise eaters DAC channel 1 and 2

Sols"s laser 1 Sols"s laser 2

Keithley 2000

mul"meters Lockin ampliﬁers WS7 wave meter

Fiber transmi!er

Serial COM4 Network switch

USB

LAN

PCI

Fiber

LAN

GPIB

Coax

Fiber Switchbox

So$ware control connec"on Fiber

Reference beams

Figure 2.1: Overview of how various devices are connected to each other in the setup, with the type of connection indicated next to the connecting arrows. Certain settings that are hard-coded into the software are written below the name of the device.

2.2.1 Choice of control interfaces

A requirements for the setup is that it should be mainly computer controlled. Many measurement devices are already available in the lab from the previous setup. These devices can be connected to each other via multiple interfaces, while other devices have only one possible

(11)

interface. This section will provide some information and motivation about the interfaces that we used to connect the devices of the setup to the computer.

Table 2.1: Latency from the specifications of various protocols used in the setup. Note:Latency of TCP/IP depends strongly on the used equipment.

Protocol Latency (µs)

USB 125 [11]

PCI 0.3 [12]

Ethernet 2-1000 [13]

GPIB 30 [14]

The hardware interfaces that are used to connect the devices are indicated in the overview flowchart of figure 2.1.

Many devices in the setup only support GPIB (general purpose interface bus). Because this is not a standard type of interface on desktop computers, either an add-in PCI card or a USB interface had to be added. A PCI (Peripheral Component Interconnect) National Instruments GPIB/VISA card has been chosen over a USB-interface. The reason we choose PCI over USB was the extra latency of the USB-protocol compared to USB. After comparing the GPIB measurement readout times of both USB and PCI interfaces in practice (see results in table 2.2) with the specifications in table 2.1 however, it is clear that the difference in readout times is a lot bigger than the microseconds of difference we expected. Possible reasons for this difference are explained in appendix A.1.2. We expect all interfaces mentioned in table 2.1 to be fast enough to reach the measurement readout goal of 10 Hz.

Not all devices can be connected using GPIB. For example, the Solstis lasers are connected via ethernet, because it is the only option provided by the factory. Mixing of interfaces is thus unavoidable. That is why we made a software package to connect the various interfaces together (see figure 2.1).

Note: Because the parameters of the Solstis lasers are set before the measurement, the speed of communication to the laser while doing measurements has not been taken into account in this section.

(12)

2.2.2 Choice of software platform

Because we want to synchronize measurement devices with the lasers, we need a single software platform that can control all devices in the setup. We choose MATLAB for this, because all devices in the setup are compatible with MATLAB. The lasers have a MATLAB compatible API (application programming interface), while MATLAB can also communicate to measurement devices via a serial interface and GPIB. An advantage of MATLAB compared to the previously used Labview software is that the workings of MATLAB scripts are much more transparent.

This makes writing, understanding and debugging scripts more user friendly.

A big part of this project was making a set of well documented scripts that handle the basic functionality of all the lab devices. These scripts provide different layers of abstraction so a measurement program can be quickly assembled using either general functions or more specific lower level functions.

MATLAB and Labview were not the only candidates for a main software platform. A com- monly suggested open source alternative is Python [15]. Compared to python, MATLAB has commercial support available. The base functionality of MATLAB is also much greater compared to Python. A plethora of libraries have to be gathered to match the base functionality of MATLAB. All MATLAB functions also have good documentation available, while the availability and quality of documentation for Python libraries varies.

Advantages of Python over MATLAB are that it is much more resource friendly, open-source and more versatile. These advantages lead to a wider application ranger. Python code can be run off almost any device and would not only be limited to a desktop computer like MATLAB.

Despite all these advantages, MATLAB has been chosen over Python due to time constraints and familiarity within the research team [16].

(13)

2.2.3 Laser setup overview

Pump laser Solstis 1 Solstis 2

Keithleys Lockins Cryo-

Shutters stat

Switchbox Power attenuation unit

PC UTP TC/IP

USB GPIB

Voltage measurement

Wave meter

Normalization Noise eaters

A)

Transmission

Power attenuation unit

High Power dump

B)

~ 3W

~ 50mW Polarizing beam splitter

Fiber coupler

Polarization maintaining single mode fiber Laserbeam Waveplate (λ/2) Mirror

Lens

Fiber beam splitter Photodiode Con

-troller

Legend

Pump laser IR laser

Shutter

Switchbox Power attenuation unit

PC UTP TC/IP

USB

GPIB

Wave meter ICE box

Sample

Switchbox λ/4

C)

Voltage meter

Figure 2.2: Optical two laser spectroscopic setup. A) Shows an overview both optical paths. The pump laser drives the two passive Solstis infrared lasers. The high power laser part of the setup is physically separated using single mode fibers. Noise eaters provide power stability and get feedback from the light at the end of the optical path. Wave plates and polarizing beam splitters are used to linearly polarize the light and shutters are added to block the light when needed. The light is then coupled into a cryostat and is converted to a voltage by photo diodes outside of the cryostat. Voltage measurement devices and the ICE boxes are connected to the PC which uses MATLAB to combine the data together and control the setup. B) Shows the optical elements inside the power attenuation unit which reduces the laser power from 3 W to about 50 mW. C) Shows the changed setup that is used for photocurrent spectroscopy. Only one laser is used and no cryostat compared to the setup in A).

Figure 2.2 gives a picture that focuses more on the optical components of the setup, unlike figure 2.1 which focuses more on the control connections between components of the setup.

The most important device of the setup is the laser. Based on the requirements from section 2.1, the Solstis laser system from M Squared has been chosen for the setup. The wavelength range of this laser system is 697 – 1003 nm, right in the required near infrared region. The Solstis also supports computer controlled continuous scans spanning the entire

(14)

wavelength range at two different modes: medium and finescan. Medium is the fastest mode and supports speeds from 100 GHz/s to 1 GHz/s, while finescan supports 10 GHz/s to 1 MHz/s. The linewidth of the Solstis laser system is well below 1 MHz, also falls well within the requirement we set in section 2.1.

The supplied controllers (ICE boxes) of the laser automate the controls of all the internal optics of the laser. Every laser has one dedicated ICE box. The workings and settings of the ICE boxes are proprietary and can only be changed by the factory. They can however be controlled via an ethernet connection. The ICE boxes are connected to a dedicated ethernet switch and have static IP addresses. The settings of the Solstis can be changed via the supplied ICE box API-library or via the web interface on the ICE box. MATLAB can communicate with the lasers using the API, while the web interface is very convenient to monitor the laser parameters while an experiment is running.

The integration of this laser system into the optical table of the setup can be seen in figure 2.2.A and B, while the integration of the control connections can be seen in figure 2.1. The Solstis lasers are passive so an 18W Coherent Verdi 532 nm laser is used as a pump laser. The coupling between the pump laser and the Solstis lasers can be manually adjusted to optimize the power output of the Solstis lasers.

2.2.4 Wavelength stabilization

The lasers and the computer are connected to a High Finesse Angstrom WS7 wave meter (see figure 2.1). The wave meter receives light from the Solstis lasers using single-mode fibers to ensure a correct readout of the wavelength.

There is no direct data coupling between the wave meter and the Solstis lasers, but the data from the wave meter is sent via the PC to the network interface of the lasers.

Because the wave meter only has one fiber input coupling and there are two lasers in the setup of which the wavelength needs to be measured, we added an optical switchbox between the lasers and the wave meter. The optical switchbox is controlled via a software link between the ICE boxes and the wave meter.

The disadvantage of this optical switchbox is that in a two-laser scan, the wavelength of only one laser can be monitored by the wavelength meter simultaneously. A measurement of the long term drift of the laser is made in appendix A.1 to investigate if there was a maximum time the laser could be left unmonitored and unstabilized while still staying within the requirements from section 2.1. In general the drift stays lower the longer a certain laser is stabilized with the aid of the wave meter. A good compromise between laser drift of one laser and scanning time of the other has to be made when using two lasers.

For two-laser experiments, we made a matlab script that pauses a scan after a set time to stabilize the parked laser to minimize the drift of the stationary laser to the requirements set

(15)

in section 2.1.

2.2.5 Power and polarization stability

The setup needs a high polarization and power stability. The fluctuations in power should be below 1 percent and the fluctuations in polarization should stay below 10 percent [8], as stated in requirement 2. The first part of the setup (see figure 2.2.B) decreases the power to a level such that it is not dangerous to work with and does not instantly bleach materials the beam contacts. This part is called the ”Power attenuation unit.” The main part of the power attenuation is a polarizing beam splitter that dumps 98.3 percent of the laser power into a high power dump. The light coming out of this unit is coupled into a single mode fiber, so the dangerous high power is physically separated from the low power part of the setup. Because it is difficult to couple light into single-mode fibers, two mirrors have been added to the setup (Thorlabs E63) at various points to aid coupling of the laser light into single-mode fibers.

Coupling light into single-mode fibers is not only difficult, but couplings are also not 100

% efficient. The efficiency of this coupling can be improved by changing the polarization of the incoming light. For example, only 0.514 mW of an incoming light beam with an power of 1.9 mW could be coupled into a single-mode fiber. By optimizing the polarization with a λ/2 wave plate, this is increased to 1.6 µW

Polarization stability

Another requirement was that the laser beams should be polarizable. In the low power part in the setup, half wave plates and polarizers have been added in the optical path. Both of these optical elements have to be adjusted manually, so a required polarization can be achieved with the setup. The requirement of the setup was that the light was not only polarizable, but the stability of this polarization should also be high. This is achieved in general by using polarization maintaining single-mode fibers whenever possible and by polarizing the laser beams at multiple points in the setup.

To measure how stable the polarization was, we connected a polarizer and a photodiode to the end of the fiber going that would otherwise be going into the cryostat at the end of the setup (see figure 2.2). The power stabilizing noise eaters were disabled for this experiment.

Changes in polarization are thus measured as changes in power (via Malus law: I^out= I0cos θi

where I⁰ is the intensity of the incoming light and θⁱ the polarization angle of the incoming light). The output of each laser was then measured for half an hour. Figure A.2 shows how the power and thus the polarization fluctuates over one run for both lasers. The data of 12 of these runs were then combined in figure 2.3 to show the number of data points at each normalized power.

(16)

0.9 1 1.1

Normalized power 0

500 1000 1500 2000 2500 3000 3500 4000

Number of datapoints

Solstis 1

0.8 1 1.2

0 1000 2000 3000 4000 5000

6000 Solstis 2

Figure 2.3: The number of data points for each polarization, created from several figure A.2 (appendix) like measurements. The powers are measured at the fiber going into the dipstick with a polarizer between the fiber and the power meter, such that the normalized power indicates the polarization deviation from the average polarization.

Power management and stabilization

One way of managing the power of the laser beam is blocking off one or two laser beams entirely. We do this by shutters, represented by the dark blue rectangles in figure 2.2.A. They can be open and closed both manually and digitally. The shutters allow for quickly switching between single and two-laser scans, as well as doing background readings without switching the lasers entirely off.

The power requirements in section 2.1 also state that the fluctuations in power should be under 1 %. Leaving the laser at a fixed wavelength yielded a very stable value with fluctuations of less than 0.1 %. The power does however fluctuate a lot while scanning the wavelength of the laser, as figure A.2 in the appendix shows.

(17)

0 50 100 150 0.17

0.18 0.19 0.2 0.21

Normalization channel (V)

814 816 818 820 822 824

0.17 0.18 0.19 0.2 0.21

Time(s) Wavelength (nm)

Figure 2.4: Power stability versus wavelength and time during a laser scan from 814 nm to 824 nm. Notice that the power changes 20 % during the laser scan, even though noise eaters have been used to surpess this effect. Also notice that the power does not necessarily change with time but does change with wavelength.

Noise eaters are placed to prevent power fluctuations while performing laser scans as much as possible. The noise eaters remove noise in the power by a feedback loop and cut the power to a level below the fluctuations. The noise eaters use light that is split off from the end of the optical path as reference so fluctuations in power caused by the optical elements placed after the noise eaters also get taken into account.

The sensitivity of the noise eater can be controlled via an input voltage and a general sensitivity switch on the noise eater itself. A MATLAB script was made to automatically find a stable setting for the sensitivity based on the intensity response of the photo diode at the end of the setup.

While they are very good at preventing short term power fluctuations, long term fluctuations can still be seen in figure 2.4. A fixed percentage of the light going into the cryostat is split off using a beam splitter and is measured using a separate photodiode to compensate for this. This split off beam is called the normalization channel throughout this thesis.

2.2.6 Moving the sample with micro precision

The light from both lasers is combined and coupled into the cryostat (see figure 2.2.A) using fiber beamsplitters and fibers. More detail about the microscopic setup contained in the cryostat can be found in the PhD thesis of Sladkov [9]. The setup also contains a superconducting magnet. This magnet is controlled manually via a standalone controller due to safety concerns and will not be discussed in this thesis.

The sample inside the cryostat can be moved around by piezo’s in the X,Y and Z direction.

These piezo’s are controlled by the Attocube ANC-150 controller. This controller supports computer controlled movement via a serial interface and is integrated into the MATLAB control software package for the setup so it can be moved around automatically. This allows for automating sample movement while scanning the sample.

(18)

A difficulty in the movement of these piezo’s is that the step size of the piezo movement depends on environmental variables and is not always predictable due to hysteresis of the piezos. Nevertheless, a rough estimate of the step size can be made by dividing the known sample length by the number of steps it takes to cross it and the size of the laser spot can be estimated by the sharpness of peaks in the optical transmission spectrum. The estimated step size at liquid helium temperature is about 0.2 µm.

2.2.7 Voltage control of devices

Many devices that do not have a direct computer control interface can still be controlled by external voltages. The noise eaters as well as the shutters are controlled by these external voltages. The voltages in this setup are generated by a device called Optodac (see figure 2.1).

It consists of 8 DAC (digital to analog converter) cards and a shielded power supply. Voltages are controlled by the computer via an optical fiber control box to eliminate electrical noise created by the computer. A voltage between −5 V and 5 V can be generated with a stepsize of 0.00015 V. These voltages have very low electrical noise in the order of a tens of µV.

The main function of this device in this setup is accurately controlling devices that are triggered by changes in voltage.

2.2.8 Voltage readouts

We measure light intensity with two photo diodes in the setup, called the normalization (intensity) channel and the transmission (intensity) channel. The two voltages provided by the photo diodes are measured by two different sets of devices. The Keithley 2000 digital multimeters and lock-in amplifiers. These devices are connected via a GPIB-interface.

Several voltage readout devices use GPIB, as visible in figure 2.2. Two of them are the Keithley digital multimeters, which read the voltages from the photo diodes in the setup. The Keithley supports two readout modes.

One method uses the trigger model. A set of parameters is set in advance and a trigger command is the used to start the measurement. This mode allows for consistent times between measurements and fast (> 100 Hz) measurements. Up to 2048 values are then stored in the buffer of the Keithley digital multimeter. Reading and clearing the buffer after each measurement takes a few seconds. This is not ideal, because the setup needs to make measurements that are much longer than a few seconds. Using this method would leave gaps in the measurements.

A second method is doing one readout at a time. The measurement readout rate for this method is much slower compared to the trigger model. The readout time also varies about 1 ms from the average, yielding slightly varying times between readouts. The advantage is a relatively constant time between measurements on multi-hour measurements. The read speed

(19)

of this method is fast enough for our requirements.

To test how long a Keithley multimeter readout takes in practice, the average readout time is calculated from 100 measurements. The time a single readout took varied about 1 ms from the average.

Another suggestion to increase the readout speed according to the manual of the Keithley 2000 was to turn off the Keithley display. The impact on the measurement time in table 2.2 was small compared to the disadvantage of not being able to read the Keithley displays while measuring. Due to this, the decision has been made to leave the display of the Keithley on by default.

The type of GPIB interface also potentially affects the readout speed, as can be seen from the specifications in table 2.1. A USB GPIB interface is compared to a PCI model in table 2.2, as well as the effect of turning off the Keithley’s display.

Table 2.2: Average measured GPIB readout time per measurement for a Keithley 2000 digital multimeter

Method Avg. readout time ( ms)

Keithley 2000 connected via PCI with display off 13 Keithley 2000 connected via PCI with display on 16 Keithley 2000 connected via USB with display off 19 Keithley 2000 connected via USB with display on 21

Notice that the time per measurement is lower than the default 20 µs (one 50 Hz power cycle) of integration time of the Keithley. We can read the Keithley faster than the integration time, but it won’t update the measured value faster. The time between the readouts has no direct correlation to the integration time. We can also conclude that the readout rate for a single Keithley is well above the 10Hz required for the setup.

Distinguishing two laser beams with a lock-in amplifier

Because the setup needs to do two-laser spectroscopy, the transmission of both individual lasers have to be measured, while both laser beams are combined on the sample. The measurements however, are done with a single photodiode. Separating the signals of both lasers can be done by adding a chopper to one of the laser beams, which adds a frequency to one of the laser beams. The lock-in amplifier, which is connected to the photo diode can then be used to extract this signal from the total transmission.

This device is slower to read than the Keithley, with an average measurement time of 30 ms.

The total time to read all four voltage meters is about 85ms, yielding a readout frequency of 12 Hz and thus satisfying the readout speed requirement set in section 2.1.

(20)

2.3. SETUP REQUIREMENTS FOR TMDC’S

2.3 Setup requirements for TMDc’s

The requirements on the laser setup are different for investigating transition metal dichalcogenide (TMDc) based devices compared to two-laser spectroscopy on gallium arsenide. The modified laser setup for performing photocurrent spectroscopy is shown in figure 2.2.C. We changed the setup due to the following list of requirements to make measurements on TMDc’s possible:

1. Photocurrent spectroscopy only requires a single laser. We therefore disconnected the second laser from the measurement setup.

2. For laser spectroscopy gallium arsenide, the required spectral resolution should be able to resolve features under 100 MHz. The spectral resolution requirement on photocurrent spectroscopy is much less strict, it should be able to resolve features in the order of 0.001 eV or 241 655 MHz[17].

3. Laser scans on gallium arsenide are in the order of 20 nm wide in wavelength for an overview scan, while photocurrent spectra use the entire laser spectrum from 700 nm to 1000 nm (1.77 eV to 1.24 eV). This is because we want to study two different optical transitions A (at 1.22 eV and B (at 1.8 eV), which are about 0.5 eV apart (see figure 4.2)[4].

The laser cavity had to be desiccated to avoid absorption of laser light by particles in the air (mainly water) as much as possible[18].

4. The power stability over a scan should ideally be constant, but due to the inner workings of a laser the power fluctuates a lot during a scan (see figure 2.5) over the entire wavelength of the laser. The solution with the noise eaters proposed in section 2.2.5 is not possible, because it only accounts for small power fluctuations. should be recorded during scans. We added a photodiode to measure the power.

5. Heating by the laser beam affects the properties of the sample. We used a mechanical shutter during laser scans to allow the sample to cool down periodically.

6. Electrical contacts of the device have to be connected to a voltage meter. We added an electrical switchbox to the setup for this purpose.

7. Instead of only measuring voltages, a source-drain and gate voltage has to be generated to activate the devices. We added a Keithley 2450 to the setup to provide a source-drain voltage, while a Keithley 2410 has been added to provide a gate voltage.

(21)

1.2 1.3 1.4 1.5 1.6 1.7 1.8 Photon energy (eV)

0 0.005 0.01 0.015 0.02 0.025 0.03

Laser power (w)

Figure 2.5: Variation of the power during a laser scan versus wavelength.

8. The polarization should be tunable between circular and linear. Because we are looking at valleytronics where the optical transitions of the individual valleys are addressable by circular polarization, while both transitions are sensitive to linearly polarized light [4].

9. The scan rate should be consistent and stable. Due to thermal effects due to the laser light incident on the sample, a consistent time between measurements is required for reliable measurement results.

(22)

2.4. CONCLUSION

2.4 Conclusion

We looked for a good approach to designing a two-laser spectroscopic setup. The key requirement was that the setup should be computer controlled to make measurements easier and faster. The Solstis lasers with the ICE-box controllers already took care of automating the laser adjustments and scanning methods. The challenge was to make all measurement devices work together with the lasers. For this, MATLAB was chosen as a main software platform. Multiple GPIB interfaces have been considered and the fastest PCI-based GPIB card was chosen for the setup to help achieve the measurement readout speed of above 10 Hz, much faster than ever before.

Next to the requirements in automation, there were also requirements in the stability of the laser power. Noise eaters have been added to prevent power fluctuations in the laser while scanning the wavelength and a normalization channel was added to measure any power fluctuations that were not removed by the noise eaters. A short-term power stability of 1 % was achieved while scanning.

The setup is also a two-laser setup. The wavelength can only be stabilized with one wave meter. This is solved by alternating the use of the wave meter between lasers with a switchbox.

Computer controlled shutters make it possible to switch from a single to a two-laser setup very quickly, making the setup also very versatile together with the wide wavelength range of the laser of 697 nm to 1003 nm. A very promising and versatile two-laser spectroscopic setup has been designed and built, that satisfies all requirements set in the beginning. The real test will be in the next chapter, where the setup will be used to do actual measurements.

(23)

Single and two-laser spectroscopy on GaAs

Outline

This chapter contains the first test case for the two-laser setup. We test all parts of the setup to check if it is capable of the same experiments and measurements as the previous setup, which was used in the previous project. An aim of the previous project was to create a solid- state quantum memory system. Quantum memory systems consists of qubits, which have two states and a quantum superposition state of these two states. Photons are useful for trans- porting qubits, but an easily upscalable solid-state system to store and manipulate qubits locally is highly desired. A gallium arsenide sample has previously been used to research such a quantum memory system. The quality of quantum memory systems in general is determined by its dephasing time, which is the time it can store quantum information before environmental interactions randomize the phase of the coherent superposition. The main question of this previous research was to find out what determines the coherence time in the used gallium arsenide sample and to what extent can we control its environment to enhance this coherence time [3]. The same gallium arsenide sample that was used in this previous research with the previous setup is thus used to test the new setup. This is because many parameters and results of this sample are already known. Every experimental step increases the requirements on the laser setup. The first step that has been done was using the automated sample movement to do single-laser scans to find a suitable spot on the sample. Single-laser scans do not require any of the two-laser functionality of the setup to work and also don’t need a linewidth of below 1 MHz.

The two-laser scans after that increase the requirements for the setup. Now, two-lasers need to be stabilized, the signal of the two-lasers has to be separated. A magnetic field of over 3 T is required to split the ground states of the sample, yielding a stricter condition for the

(24)

3.1. SAMPLE

linewidth of the lasers.

Next, two-photon resonance effects like CPT (coherent population trapping) are investigated, which really require a narrow linewidth of under 1 MHz, the maximum power and wavelength stability our setup should be capable of.

Combining single-laser pumping and scans for CPT can yield information about the nuclear spin diffusion in our gallium arsenide sample, the last and hardest test case. The nuclear spin diffusion has been investigated, because random fluctuations of the nuclear spin are a main cause for dephasing the superposition state of our system. Some other questions beyond the scope of this thesis about the D⁰- D⁰X system have also been answered due to the easiness of taking data with this new setup.

3.1 Sample

The same sample as in the previous setup is studied. This is sample that consists of a gallium arsenide layer placed on a sapphire substrate. Gallium arsenide has been chosen as a material for the sample because can be grown with very high quality. The sample dimensions are 2x2 mm and the thickness is 10 µm. Silicon has been used as a dopant, with a concentration of 10³per cm⁻³. This is a very shallow doping and makes the sample barely conductive at room temperature.

The interesting part happens when the sample is cooled down. The free electrons are situated at the silicon donors when the temperature is well below the ionization temperature of silicon. The donor silicon atom is then neutral again due to localized electrons, creating an artificial hydrogen atom like D⁰ state. The ground state of D⁰ consists of a spin up and spin down state, each with equal energy. An additional free exciton (electron-hole pair), creates a D⁰X excited state where the ground state electrons can transfer to. Electrons in the ground state D⁰can be optically excited to D⁰X by allowed optical transitions.

3.2 Positioning the laser beam on the sample

The first step in the experiment is positioning the laser beam. This makes use of the automatic piezo movement system described in section 2.2.6. We do not position the laser on a random spot, but on a spot with certain ”good” properties. We use the laser to address a mesoscopic ensemble of sites on the sample, to improve the robustness of the quantum system and interaction with photons. We can determine the quality of a certain laser spot by obtaining a single-laser scan and analyzing the spectrum.

What we expect the spectrum of a good spot looks like can be predicted by considering that the energy between the ground and excited state of gallium arsenide depends on the strain of the sample. The strain within the laser spot therefore needs to be homogeneous, because

(25)

we want to address as many donor sites as possible in the laser spot simultaneously with a narrow laser bandwidth. When only a small amount of donor sites are addressed, weak features will disappear in the transmission spectrum of the single-laser scan.

A way of predicting the strain in the sample on a certain spot is by the height differences of the sample. Figure 3.1 is a height map of the sample. A good spot is expected in a place that is homogeneous in height, so the best spots are expected in the upper middle part of the sample. To verify this, we made a script that automatically moved the sample across the arrow indicated in figure 3.1 in small steps. A single-laser scan was obtained at every step.

Figure 3.1: A differential interference contrast microscope image of the sample. This image accentuates height differences on the sample. The long arrow shows the scanned direction, while the short bar indicates the distance between scans.

3.3 Single-laser spectroscopy

We use single-laser spectra to determine if the ensemble we adress within the laser spot is homogeneous. We choose a good spot on the sample by analyzing all single-laser spectra obtained on the sample, along the arrow in figure 3.1. There are multiple features in a single- laser spectrum which define a good spot, which can not be determined from an optical image,

(26)

3.3. SINGLE-LASER SPECTROSCOPY

like strain. These features are shown in figure 3.2, which shows a typical single-laser spectrum of a good and a bad spot on the sample. A good spot should have visible small peaks of the higher order free excitons (energy Ex(n) = E_gap− Ex/n², where Ex is the binding energy, n, the exciton order and Egap the bandgap energy [19]), indicated with an A in figure 3.2. The homogeneity of the strain within the spot is clearly visible in the width peak of the lowest order exciton (n=1). A non-homogeneous strain yields a wide peak, as indicated with B of figure 3.2. Compare that to the narrow peak in pink. The ”bad” laser spot where the blue spectrum in figure 3.2 spectrum has been obtained, is not glued to the substrate. A cavity is thus created between the gallium arsenide and the substrate. The light shining through the sample interferes with itself and creates visible Fabry–P ´erot oscillations in the spectrum, indicated with a D in figure 3.2.

The part of the spectrum we are most interested in is the peak of the optical transitions of D⁰X. The wavelength of this peak corresponds to the energy required for the transition from the ground state of D⁰to the excited state D⁰X. We can see in figure 3.2 that this peak, indicated with a C, is clearly visible in the pink spectrum of the good spot, while it is absent in the blue spectrum of the bad spot.

The laser setup is clearly capable of taking single-laser scans with a resolution high enough to distinguish all important features we look for in single-laser spectra. The time it took to find a good spot was greatly reduced because taking single-laser scans on the sample could be automated due to the computer controlled sample movement. In the end, a spot with a slightly more homogeneous strain was found.

(27)

818 819 820 821 822 823 824 Wavelength (nm)

-0.2 -0.1 0 0.1 0.2 0.3 0.4

Transmission (V)

A

B

C

Bad spot

Good spot

D

Figure 3.2: A typical single-laser spectrum of a typical good spot in pink and a typical bad spot in blue.

Several features are highlighted by arrows. A: Optical transition wavelength of higher order free excitons.

B: Optical transition of free exciton with the lowest energy (n=1) C: Shows the optical transition of the D⁰X excited state D: Fabry–P´erot oscillations due to a cavity between the sample and the substrate.

3.4 Two-laser spectroscopy

B

B* A* A

Δω Ground states

Excited states

D X

D

0

ω

₁

ω

₂

Figure 3.3: Energy level splitting due to an external magnetic field of a Lambda system. The ground states of D⁰are indicated with an arrow up for spin up and an arrow down for spin down. The frequency corresponding to the energy difference of both ground states is denoted by ∆ω. The frequencies of both optical transitions are indicated by ω1 and ω2. The frequency and polarizations of the transitions A, A*, B and B* can be found back in transition tables like 3.1. Figure adapted from [3].

(28)

3.4. TWO-LASER SPECTROSCOPY

The second test for the two-laser setup is to see if we can take two-laser scans. These scans impose stricter linewidth and stability requirements on the setup. Another requirement of two-laser scans compared to single-laser scans is that two ground states of the D⁰ - D⁰X system should be addressable by different optical transitions. We used a magnetic field to split the ground state into ω1for the spin up state and ω2 for the spin down state (see figure 3.3).

This yields the question of how much the ground state should be split for the ability to take two-laser scans. The energy difference between the ground states due to Zeeman splitting equals:

∆E = gµbBext (3.1)

.

Equation 3.1 can be used to estimate the external magnetic field needed to sufficiently split the ground state D⁰. The ground state levels are not addressed by one single wavelength, but they have resonance curve of wavelengths where they are driven. The FWHM of this curve is about 10 GHz for D⁰. The FWHM of the resonance curves for both transitions should not overlap for a sufficient splitting. This puts a stricter requirement on the line width of the laser, but is still far from the 1 MHz the setup should be capable of, as well as a minimum requirement of 1.74 T on the magnetic field.

We take two-laser scans by keeping one laser with ω¹ stationary at a transition, while scanning the other laser with frequency ω². The stationary laser drives a certain transition and after a while, all electrons have moved from the ground state to the excited state. In this case, no light will be absorbed anymore and thus the transmission spectrum will show no transmission. When we then use a scanning laser to drive a transition, different from the transition the stationary laser drives, electrons will move back to the ground state and light from the stationary laser will be absorbed again. The scanning laser thus changes the transmission of the light of the stationary laser.

Because a chopper needs to be added to the path of the stationary laser, there is a drop in power. The lock-in amplifier is used to extract the signal of the signal of this laser and while this device is very sensitive, it should still stay above background noise.

The optical transitions are sensitive to either H-polarization (oscillating in the plane of the magnetic field) or V-polarization or both polarizations. Figure 3.4 shows a typical two-laser spectrum of both polarizations. Sharp peaks in the transmission of the spectrum are indeed found, indicating several different energy levels. The two-laser spectra taken with a current of 20A in the superconducting magnet can be found in appendix A.2. From these scans, the transitions in figure A.5 have been found. The transitions at a magnetic current of 30 A can be found in the appendix.

We can clearly see in figure 3.4 the setup is capable of doing two-laser scans. The setup is also capable of performing two-laser scans at a magnetic field of 1.01 T, which is a much

(29)

weaker magnetic field than has been achieved ever before in our group. The next step is to find out if the setup can also show two-photon resonance phenomena in the next section.

Table 3.1: Optical transitions that have been found in two-laser scans with a magnetic current of 20 A.

The header with a grey background indicates the polarization and wavelength of the stationary laser. All transitions that have been found with the scanning laser are shown in this table. The D⁰- D⁰Xtransitions have been indicated with a letter corresponding to the labeled letter of a transition in figure A.5. The units of this table are GHz.

H 364948.53 H 364948.53 V 364961.80 V 364961.80 H 364993.18 (B*) V 365087.62 V 365076.10 H 365077.52

V 364994.47 V 365062.36 H 364950.07 (A) V 364961.67 (A*) V 365032.39

V 364983.05 (B)

819 819.5 820 820.5

Wavelength of scanning laser (nm) 0.09

0.095 0.1 0.105 0.11 0.115

Transmission of stationary laser (V)

H polarization scanned V polarization scanned

Figure 3.4: two-laser scan over both H and V. The stationary laser was V-polarized and kept at a wavelength of 819.526 nm. Several peaks indicating optical transitions are greater in the H-polarized scan compared to the V-polarized scan, indicating a that these transitions are more strongly driven by H-polarized light.

Note that the wavelengths are slightly shifted in the two-laser scans compared to the single-laser scans in figure 3.2 due to the external magnetic field being present in two-laser scans.

3.5 Coherent population trapping

The measurement that puts the strictest requirements on the two-laser setup is detecting coherent population trapping (CPT). This is a special case of electromagnetically induced trans- parency, an effect whereby a material is transparent to light resonant with one of the transi-

(30)

3.5. COHERENT POPULATION TRAPPING

tions in figure 3.3, by applying a strong electromagnetic field which is resonant with the other transition. When CPT occurs, the system is ”trapped” in a dark state, which is a coherent superposition of two spin states. The CPT shows up as a peak in a two-laser scans because the system will not absorb any light when it is trapped in the dark state.

A typical CPT peak in our system is only 100 MHz wide in the frequency spectrum [8], which puts a maximum of 1 MHz on the linewidth of the scanning laser. The drift of the stationary laser over time should also stay under 10 MHz. We expect CPT to occur when the frequency difference between ω¹ and ω² is equal to the frequency corresponding to the splitting of the ground states of D⁰from figure 3.3.

A typical CPT peak of the sample can be seen in figure 3.5. Note that his CPT peak is measured with a magnetic field of just 1.01 T, which is a much lower magnetic field compared to the previous setup. Indicating a very good wavelength stability and linewidth of this setup.

Laser scans in this chapter are frequently taken at a magnetic field of 1.97 T because it is a weaker magnetic field than has been used in the previous setup, but features are more clearly visible compared to the same scans at a magnetic field of 1.01 T.

Figure 3.5 shows that the setup clearly succeeds in showing CPT peaks in laser scans.

Open questions from the previous project can thus be addressed. A question in the previous project about this system was how CPT depends on the intensity of the laser. This is investigated and while the power required to observe CPT was different at various magnetic fields, the laser power required to observe CPT was lower at lower external magnetic fields.

821.206 821.21 821.214 821.218

Wavelength of scanning laser (nm) 2.55

2.6 2.65

Transmission (V)

Figure 3.5: Two-laser scan in the D⁰Xtransition of the GaAs sample. An increase in transmission can be seen around 821.2139 nm, indicating CPT. This CPT peak has been measured at a magnetic field of 1.87 T.

The stationary laser was at a wavelength of 821.1883 nm.

(31)

821.268 821.2685 821.269 821.2695 821.27 Wavelength of scanning laser (nm)

2.58 2.59 2.6 2.61 2.62 2.63 2.64

Transmission (V)

Figure 3.6: A close up of a CPT peak taken at a magnetic field of 1.01 T. This was the lowest magnetic field strength where CPT was observable. The stationary laser was at a wavelength of 821.282 00 nm, while the top of the peak can be seen at 821.268 95 nm.

3.6 Nuclear spin polarization

The coherence of localized spins in gallium arenide is detioriated by the non-zero nuclear spin of both gallium and arsenide. The orientation of these spins show random thermal fluctuations and interact via the hyperfine interaction. This is the dominant dephasing mechanism for electron-spins in our gallium arsenide sample and reduces the dephasing time of the bound-electron spin ensemble to a 3 ns. Open questions regarding this spin diffusion from the previous project were also investigated using this setup. A goal of the previous project was to lengthen the dephasing time of electron spin in the D⁰- D⁰X system. Information about the spin dephasing time can be extracted from the width of the CPT peak. The CPT peak has a spectral width because every site in the D⁰X does not have the same energy and is addressed by an optical transition with a slightly different wavelength. The CPT peak thus shows a distribution of the energies of all optically addressed sites within the laser spot. The ensemble spin dephasing T²∗ depends on the width this distribution.

The main contributor to the short dephasing time of our sample are the nuclear spins of gallium arsenide. Because the wave functions of the electrons are S-shaped, the highest prob- ability density of the electron is at the nucleus of the atom. This causes a strong interaction between the nuclear spin and the electron spin (hyperfine interaction). A D⁰ electron sees about 10⁵nuclear spins from the GaAs bulk of the sample. All these nuclear spins cause the electron spin to fluctuate a lot with their strong spin-orbit interaction, greatly lowering the

(32)

3.6. NUCLEAR SPIN POLARIZATION

spin coherence time T²∗.

The measured spin decoherence time in the investigated sample without applying tech- niques to increase this time is about 3 ns [3]. For a quantum computer, we need the coherence time to be as long as possible and at least in the microsecond region. One of the main goals of the previous project was to see how close the coherence time of the sample could be brought to the target of several micro seconds. The nuclear spin has a big influence on T²∗. That is why the saturation behaviour of the diffusion of nuclear spin has been investigated in this new setup.

Addressing questions regarding this is instrumentally interesting because the long term power and wavelength stability becomes more important in this part.

Nuclear spin diffusion

Ground states

Excited state

ħ

Nuclear spin bath of GaAs bulk D -D X lambda system

Hyperfine interaction

Wavelength of scanning laser (a.u.)

Transmission (a.u.)

A) B)

ΔE

1

2

3 4

5

0 0

Figure 3.7: A) Dynamic nuclear spin polarization. 1) An electron is excited by the laser. 2) The electron as a chance of falling back to the spin down state. 3) When the electron the falls back to the spin up state, ~ of angular momentum is then transferred to the surrounding nuclear spin bath, polarizing the nuclear spin out of equilibrium. 5) The nuclear spin influences the electron spin and changes the energy splitting ∆E between the spin up and spin down ground state. B) The change in energy difference between the spin up and spin down ground state causes a shift of the CPT peak.

A main mechanism of nuclear spin diffusion in the D⁰X-system is dynamic nuclear polarization (DNP). This process happens through hyperfine interaction between the nuclear and electron spin, as explained in figure 3.7. DNP causes a shift of the CPT peak, which can be seen in figure 3.8. This shift also increases with an increasing external magnetic field (see figure A.6 of the appendix). Data can then be obtained by setting the laser to a frequency ω² for a time (pump time) and then doing a CPT scan after a certain time to see how the CPT peak has been shifted.

Nuclear spin polarization is not an instantaneous process. The nuclear spin of atoms near the donor D⁰sites are polarized first and then the spin polarization diffuses outward, mainly via dipole-dipole coupling.

The spin diffusion rate of the nuclear spin within the laser spot is generally fast compared

(33)

to the diffusion rate of the nuclear spin into the bulk, outside of the laser spot, yielding two timescales. The nuclear spin diffusion into the bulk thus has a relatively small effect on the spin relaxation time within the laser spot. When the laser is then turned off, the nuclear spin will start to relax again. This causes the CPT peak to move back to its original position over time.

We repeated the DNP-experiment several times with an increasing pump time and plotted the results in figure 3.8. Exponential fits can then be used to find an average relaxation time of the nuclear spin polarization of 14.19 s from figure 3.8. The figure also shows that the spin relaxation time does not differ for various pump times, but there is a slight offset. We also investigated the power dependence of CPT, but no power dependence above a threshold value has been found.

The two-laser setup is thus capable of DNP and several open questions of the D⁰ - D⁰X system can be and have been addressed using this setup. The computer controls allowed a wide range of variables to be checked on overnight scans, yielding much more data to analyze.¹

0 5 10 15 20 25 30 35

Dark time (s) 0

20 40 60 80 100 120 140 160

CPT shift (MHz)

tau = 6.91s ± 0.91s tau =13.04s ± 3.6s tau = 9.68s ± 2.02s tau = 9.31s ± 2.04s tau = 10.74s ± 2.30s

Figure 3.8: CPT peak position after single-laser dynamic nuclear polarization (DNP), at a superconducting magnet strength of 1.97 T. The lines indicate exponential fits for multiple runs at the same settings. The relaxation times of the nuclear spin polarization τ are shown in the legend. The average relaxation time τ is 14.19±0.69s

1Note: DNP can also be done using two lasers. More about this can be found in [3], but this is beyond the scope of this thesis.

(34)

3.7. CONCLUSION

3.7 Conclusion

The question we wanted to answer in this chapter was: Can we confirm proper functioning of the setup designed in chapter 2 by doing measurements on the D⁰- D⁰X system in GaAs? The test started by finding a good spot on the sample to do experiments on. The automated sample movement was a great help in this first step. The setup was capable of doing single-and two- laser scans. Coherent population trapping (CPT) transmission peaks were visible at magnet field of 1.01 T, a much weaker magnet field than the previous setup needed to show CPT. The setup was also capable of measuring the relaxation time of the nuclear spin polarization of this system using the movement of the spectral position of the CPT peak.

Furthermore, the setup has been used to address questions that were still open from the previous project about the D⁰ - D⁰X system. The laser power dependence of CPT has been investigated. Powers required to observe CPT were very different at various magnetic fields and an exact dependence could not be identified. However, in general CPT appeared at much lower laser powers for weaker magnetic fields. Other open questions about the control of DNP buildup and relaxation could also be answered using this setup.

To summarize, proper function of the two-laser setup has been confirmed.

(35)

Photocurrent spectroscopy on transition metal dichalcogenide devices

4.1 Outline

We use the two-laser setup to characterize field-effect transistors based on transition metal dichalcogenides (TMDC). We compare the electrical and optical response of mono- and multilayer M oSe2-based FETs. We observe two excitonic transitions A0 and B0 by using photocurrent spectroscopy on the sample. We also observed an extra peak corresponding to a strongly spin-forbidden dark transition, which can be controlled by gate. We also obtain information on the dependence of the optical bandgap on M oSe2-cyrstal thickness from the photocurrent spectra of multilayer M oSe². ¹

4.2 Introduction

Atomically thin TMDCs are very attractive materials for two-dimensional (2D) optoelectronics, since they have a band gap in contrast to graphene [20]. The bandgap of around 1 eV is particularly interesting because it is in the visible spectrum. TMDCs have also shown high carrier mobilities for holes and electrons [21]. The bandgap of monolayer M oSe²is direct and about 1.54 eV and that of multilayer or bulk is indirect and about 1.34 eV [22], which falls right within the wavelength range of our laser setup. The structure of multilayer TMDCs consist of multiple layers of a hexagonal crystal structure, which are bound by weak Van der Waals forces and strong in-plane covalent bonding. The thickness of one M oSe²layer is 6 – 7 ˚A [23].

1This chapter is partly based on a published paper: DOI 10.1088/2053-1583/aa8aa0

high-resolution two-laser spectroscopy setup and its applications to solid state