All-electronic Nanosecond-resolved Scanning Tunneling Microscopy: Facilitating the Investigation of Single Dopant Charge Dynamics

Podsumowanie

We demonstrate an all-electronic method to observe nanosecond-resolved charge dynamics of dopant atoms in silicon with a scanning tunneling microscope.

Streszczenie

The miniaturization of semiconductor devices to scales where small numbers of dopants can control device properties requires the development of new techniques capable of characterizing their dynamics. Investigating single dopants requires sub-nanometer spatial resolution, which motivates the use of scanning tunneling microscopy (STM). However, conventional STM is limited to millisecond temporal resolution. Several methods have been developed to overcome this shortcoming, including all-electronic time-resolved STM, which is used in this study to examine dopant dynamics in silicon with nanosecond resolution. The methods presented here are widely accessible and allow for local measurement of a wide variety of dynamics at the atomic scale. A novel time-resolved scanning tunneling spectroscopy technique is presented and used to efficiently search for dynamics.

Wprowadzenie

Scanning tunneling microscopy (STM) has become the premier tool in nanoscience for its ability to resolve atomic-scale topography and electronic structure. One limitation of conventional STM, however, is that its temporal resolution is restricted to the millisecond timescale because of the limited bandwidth of the current preamplifier¹. It has long been a goal to extend STM's temporal resolution to the scales on which atomic processes commonly occur. The pioneering work in time-resolved scanning tunneling microscopy (TR-STM) by Freeman et. al.¹ utilized photoconductive switches and microstrip transmission lines patterned on the sample to transmit picosecond voltage pulses to the tunnel junction. This junction-mixing technique has been used to achieve simultaneous resolutions of 1 nm and 20 ps², but it has never been widely adopted due to the requirement of using specialized sample structures. Fortunately, the fundamental insight gained from these works can be generalized to many time-resolved techniques; even though the bandwidth of the STM's circuitry is limited to several kilohertz, the non-linear I(V) response in STM allows faster dynamics to be probed by measuring the average tunnel current obtained over many pump-probe cycles. In the intervening years, many approaches have been explored, the most popular of which are briefly reviewed below.

Shaken-pulse-pair-excited (SPPX) STM takes advantage of the advancements in ultrafast pulsed laser technologies to achieve sub-picosecond resolution by directly illuminating the tunnel junction and exciting carriers in the sample³. Incident laser light creates free carriers that transiently enhance conduction, and modulation of the delay between the pump and probe (t_d) allows dI/dt_d to be measured with a lock-in amplifier. Because the delay between the pump and probe is modulated rather than the laser's intensity, as in many other optical approaches, SPPX-STM avoids photo illumination-induced thermal expansion of the tip³. More recent extensions of this approach have extended the timescales over which SPPX-STM can be used to investigate dynamics by utilizing pulse-picking techniques to increase the range of pump-probe delay times⁴. Importantly, this recent development also provides the ability to measure I(t_d) curves directly rather than via numerical integration. Recent applications of SPPX-STM have included the study of carrier recombination in single-(Mn, Fe)/GaAs(110) structures⁵ and donor dynamics in GaAs⁶. Applications of SPPX-STM face some restrictions. The signal SPPX-STM measures depends on free carriers excited by the optical pulses and is best suited to semiconductors. Additionally, although the tunneling current is localized to the tip, because a large area is excited by the optical pulses, the signal is a convolution of the local properties and material transport. Finally, the bias at the junction is fixed at the measurement timescale so that the dynamics under study must be photoinduced.

A more recent optical technique, terahertz STM (THz-STM), couples free-space THz pulses focused on the junction to the STM tip. Unlike in SPPX-STM, the coupled pulses behave as fast voltage pulses allowing for the investigation of electronically driven excitations with sub-picosecond resolution⁷. Interestingly, the rectified current generated from the THz pulses results in extreme peak current densities not accessible by conventional STM^8,9. The technique has been used recently to study hot electrons in Si(111)-(7x7)⁹ and image the vibration of a single pentacene molecule¹⁰. THz-pulses naturally couple to the tip, however, the necessity to integrate a THz source to an STM experiment is likely to be challenging to many experimenters. This motivates the development of other widely applicable and easily implementable techniques.

In 2010, Loth et al.¹¹ developed an all-electronic technique where nanosecond voltage pulses applied on top of a DC offset electronically pump and probe the system¹¹. The introduction of this technique offered a critical demonstration of unambiguous and practical applications of time-resolved STM to measure previously unobserved physics. Although it is not as fast as junction mixing STM, which preceded it, applying microwave pulses to the STM tip permits arbitrary samples to be investigated. This technique does not require any complicated optical methodologies or optical access to the STM junction. This makes it the easiest technique to adapt to low temperature STMs. The first demonstration of these techniques was applied to the study of spin-dynamics where a spin-polarized STM was used to measure the relaxation dynamics of spin-states excited by the pump pulses¹¹. Until recently, its application remained limited to magnetic adatom systems^12,13,14 but has since been extended to the study of carrier capture rate from a discrete mid-gap state¹⁵ and charge dynamics of single arsenic dopants in silicon^15,16. The latter study is the focus of this work.

Studies on the properties of single dopants in semiconductors have recently attracted significant attention because complementary metal oxide semiconductor (CMOS) devices are now entering the regime where single dopants can affect device properties¹⁷. In addition, several studies have demonstrated that single dopants can serve as the fundamental component of future devices, for example as qubits for quantum computation¹⁸ and quantum memory¹⁹, and as single atom transistors^20,15. Future devices may also incorporate other atomic-scale defects, such as the silicon dangling bond (DB) which can be patterned with atomic precision with STM lithography²¹. To this end, DBs have been proposed as charge qubits²², quantum dots for quantum cellular automata architectures^23,24, and atomic wires^25,26 and have been patterned to create quantum Hamiltonian logic gates²⁷ and artificial molecules^28,29. Moving forward, devices may incorporate both single dopants and DBs. This is an attractive strategy because DBs are surface defects that can easily be characterized with STM and used as a handle to characterize single dopant devices. As an example of this strategy, DBs are used in this work as charge sensors to infer the charging dynamics of near-surface dopants. These dynamics are captured with the use of an all-electronic approach to TR-STM that is adapted from the techniques developed by Loth et al.¹¹

Measurements are performed on selected DBs on a hydrogen terminated Si(100)-(2x1) surface. A dopant depletion region extending approximately 60 nm below the surface, created via thermal treatment of the crystal³⁰, decouples the DB and the few remaining near-surface dopants from the bulk bands. STM studies of DBs have found that their conductance is dependent on global sample parameters, such as the concentration of dopants and the temperature, but individual DBs also show strong variations depending on their local environment¹⁶. During an STM measurement over a single DB, the current flow is governed by the rate at which electrons can tunnel from the bulk to the DB (Γ_bulk) and from the DB to the tip (Γ_tip) (Figure 1). However, because the conduction of the DB is sensitive to its local environment, the charge state of nearby dopants influences Γ_bulk (Figure 1B), which can be inferred by monitoring the DB's conductance. As a result, the conductance of a DB can be used to sense the charge states of nearby dopants, and can be used to determine the rates at which the dopants are supplied electrons from the bulk (Γ_LH) and lose them to the STM tip (Γ_HL). To resolve these dynamics, TR-STS is performed around the threshold voltages (V_thr) at which the tip induces ionization of near-surface dopants. The role of the pump and probe pulses is the same in the three time-resolved experimental techniques presented here. The pump transiently brings the bias level from below to above V_thr, which induces dopant ionization. This increases the conductance of the DB, which is sampled by the probe pulse which follows at a lower bias.

The techniques described in this paper will benefit those wishing to characterize dynamics occurring on the millisecond to nanosecond timescale with STM. While these techniques are not limited to studying charge dynamics, it's crucial that the dynamics are manifested via transient changes in the conductance of states that can be probed by STM (i.e., states on or near the surface). If the conductance of the transient states does not differ significantly from the equilibrium state, such that the difference between the transient and equilibrium currents multiplied by the probe pulse duty cycle is smaller than the systems noise floor (typically 1 pA), the signal will be lost in the noise and will not be detectable by this technique. Because the experimental modifications of commercially available STM systems required to perform the techniques described in this paper are modest, it's anticipated these techniques will be widely accessible to the community.

Protokół

1. Initial Setup of Microscope and Experiments

1. Begin with an ultrahigh vacuum cryogenic-capable STM and associated control software. Cool the STM to cryogenic temperatures.
   NOTE: Throughout this report, ultrahigh vacuum refers to systems that achieve <10 x 10^-10 Torr. The STM should be cooled to cryogenic temperatures; this is especially important when investigating the charge dynamics of dopants, which are thermally activated at modest temperatures. Other chambers may be at room temperature.
2. Ensure that the STM tip is equipped with high-frequency wiring (~500 MHz).
   NOTE: By using pulse shaping methods, a great increase in the time response of a STM with standard cryogenic temperature coaxial wiring (~20 MHz) has been reported by Grosse et al.³¹
3. Connect an arbitrary function generator with at least two channels to the tip (Figure 2), which will be used to prepare the cycles of voltage pulse pairs used for pump-probe experiments.
4. Configure the arbitrary function generator so that the pump and probe voltage pulses are generated independently and summed before being fed into the tip.
5. Apply the DC bias voltage used for imaging and conventional spectroscopy to the sample (V_DC).
6. Connect two radio frequency switches to the arbitrary function generator's output channels.
7. Configure the switches so that the tip will be grounded during STM imaging and conventional spectroscopy, and the effective bias is V_DC + V_tip during pump-probe experiments (Figure 4A).
8. Collect the tunneling current for all measurements through a preamplifier connected to the sample.

2. Preparation of the H-Si(100)-(2x1) Reconstruction

1. Cleave a sample from a 3-4 mΩ·cm n-type arsenic doped Si(100) wafer by scratching the back of the wafer with a silicon carbide scriber and gently snapping the sample off the wafer with glass microscope slides.
2. Affix the sample to an STM sample holder and introduce it to an ultrahigh vacuum chamber adjacent to the STM chamber.
3. Degas the sample by resistively heating it to 600 °C (a pyrometer can be used to monitor the sample's temperature) and holding it at that temperature for at least 6 h in ultrahigh vacuum.
   NOTE: The pressure will initially rise as the sample and sample holder degas, but should stabilize near the base pressure (<10^-10 Torr) after several hours.
4. Allow the sample to cool to room temperature before continuing.
5. Degas a tungsten filament in the same chamber as the sample by resistively heating the filament to 1800 °C and waiting for the system to recover to base pressure. Turn off the filament before continuing.
   NOTE: The sample can remain in the chamber during this step because it is passivated by the native oxide layer at its surface, and any contamination of the sample surface caused from this step will be removed subsequently. The filament's temperature will have to be calibrated to a specific current/voltage applied to the filament using a pyrometer.
6. Remove the oxide from the sample surface by flashing the sample to 900 °C and holding it at that temperature for 10 s before cooling it to room temperature. The pressure will increase several orders of magnitude from the base pressure during the flashing procedure. After each of the flashes found throughout this procedure, wait for the sample to cool to room temperature and the system to recover to base pressure before continuing.
   NOTE: Flashing is defined within this report as heating and cooling the sample with high ramp rates, on the order of 100 °C/s.
7. Progressively flash the sample to higher temperatures while attempting to reach a final flash of 1250 °C. Abort any flash where the pressure rises above 9x10^-10 Torr to prevent the sample surface from getting contaminated. Record the voltage/current used to achieve the 1250 °C flash (the light given off by the heated filament in step 2.6 will prevent a pyrometer from giving an accurate reading of the sample's temperature, and thus this setpoint should be used). On the final flash, determine the voltage/current required to heat the sample to 330 °C as the crystal is cooled, then let the sample cool to room temperature, and let the system recover to base pressure before continuing.
8. Leak H₂ gas into the chamber at a pressure of 1x10^-6 Torr and heat the tungsten filament to 1800 °C.
   NOTE: This has the effect of cracking the H₂ to atmoic hydrogen³².
9. Hold the sample in these conditions for 2 min before flashing the sample to 1250 °C_, holding it at that temperature for 5 s, and cooling it to 330 °C.
10. After 1 min of exposure at 330 °C, simultaneously close the H₂ leak valve, turn off the tungsten filament, and let the sample cool to room temperature.
    NOTE: These high flash temperatures affect the distribution of dopants in the sample. Heating to 1250 °C has been found to induce a ~60 nm dopant depletion region near the sample surface³⁰.
11. Verify the sample's quality by taking STM images of the surface.
    NOTE: Good samples will have large (>30 nm x 30 nm) terraces with a defect rate of <1% (dangling bonds, adsorbed molecules, adatoms, etc.) and will demonstrate the classic Si(100)-(2x1) reconstruction³², which features dimer rows running antiparallel to one another across step edges (Figure 3B).

3. Assessing the Quality of the Pump-probe Pulses at the Tunnel Junction

1. Approach the STM tip to the sample surface by engaging the current feedback controller with a current setpoint of 50 pA and a sample bias of -1.8 V.
   NOTE: Under these conditions, the tip is estimated to be <1 nm from the sample surface. The STM tip used in this work was produced by chemically etching polycrystalline tungsten. It was sharpened further using a nitrogen etching procedure, which is well described in Rezeq et al.³³.
2. Look for an area on the sample surface free from large surface defects by taking large area scans (e.g., 50 nm x 50 nm).
3. Position the STM tip over an H-Si on the surface, which appear as the dimer rows in STM images (Figure 3B).
4. Turn off the current feedback controller
5. Set V_DC to -1.0 V, V_pump to -0.5 V, V_probe to -0.5 V, the width of the pump and probe pulses to 200 ns, and the rise/fall time of the pulses to 2.5 ns (Figure 4A).
6. Send a series of trains of pump and probe pulses where the relative delay of the pump and probe is swept from -900 ns to 900 ns.
7. Plot the tunneling current as a function of the delay between the pump and probe. It will likely demonstrate strong ringing (oscillations in the tunneling current as a function of the relative delay between the pump and probe pulses, Figure 4B).
   NOTE: Python and Origin software were used to plot, analyze, and evaluate the data collected for this manuscript.
8. Repeat steps 3.1–3.5, but increase the rise/fall times of the pulses. The ringing will decrease as the rise/fall times are increased.
   NOTE: It's desired to eliminate ringing to provide the most accurate spectroscopic results, however, the time-resolution of these techniques is limited to the width of the pulses used. 25 ns rise times were used for this work.

4. Time-Resolved Scanning Tunneling Spectroscopy (TR-STS)

1. Position the STM tip over a silicon DB, which appear as bright protrusions at negative tip-sample biases (Figure 3B).
2. Turn off the STM current feedback controller.
3. Send a train composed of only the probe pulse with a repetition rate of 25 kHz. Over a series of pulse trains, sweep the bias of the probe pulse over a range of 500 mV from the DC bias of -1.8 V.
   NOTE: This simple experiment is analogous to conventional STS where the conductance is sampled over a range of biases.
   1. Configure the duration of the pulse trains (each with a different bias) such that the resulting spectra have a signal to noise ratio >10.
4. Send a train composed of pump pulses set at a fixed bias (such that V_DC + V_pump > V_thr) with a repetition rate of 25 kHz. In these experiments, set the V_DC, V_pump, and V_thr to -1.8 V, 500 mV and -2.0 V, respectively.
   NOTE: Pump pulses can have arbitrarily long durations (1 µs is typically sufficient).
5. Send a train composed of the pump pulses with the probe pulses followed by a delay of 10 ns. In these experiments, set the amplitude of the pump pulse as 500 mV and of the probe pulse sweep from 50 to 500 mV.
   NOTE: In this experiment, the probe pulse is sampling the state prepared by the pump pulse, rather than the equilibrium state sampled in conventional STS.
   1. Subtract the signal obtained when only the pump pulse was applied when displaying/evaluating the signal collected from this step.
6. Compare the probe only and the pump + probe signals by plotting them in the same graph. Any hysteresis in the two signals is an indication of dynamics that can be probed with time-resolved STM techniques.
   NOTE: By keeping the range of the probe pulse fixed and coarsely scanning the DC offset (in 0.25 V steps, for example), one can efficiently map the entire energy range of the sample to identify dynamics accessible to the technique. Pulse durations can be modified depending on the experiment. The width of the pump pulse needs to be longer than the rate at which the dopant is ionized, such that it consistently ionizes the dopant. In general, probe durations should be of the same order as the dynamical process under study, such that the maximum signal can be measured without sampling an average of the two conductance states. When searching for energies at which dynamics exist, it is recommended that durations of the probe are minimized such that only one state of the system is measured to enhance hysteresis. As the relaxation time constants are found, the duration of the probe pulse can be increased to improve the signal to noise ratio.

5. Time-resolved STM Measurements of Relaxation Dynamics

1. Position the STM tip over a silicon DB and turn off the STM current feedback controller.
2. Send a train composed of pump pulses set at a fixed bias (such that V_DC + V_pump > V_thr) with a repetition rate of 25 kHz. In these experiments, set the V_DC, V_pump, and V_thr to -1.8 V, 400 mV and -2.0 V, respectively.
   NOTE: Pump pulses can have arbitrarily long durations (1 µs is typically sufficient).
3. Send a train of pump and probe pulses. Ensure that the probe pulses have an amplitude smaller than the pumps and comparable to the range at which hysteresis occurs (V_probe < V_pump, V_probe + V_DC≈V_hystersis).
4. Sweep the delay between the pump and probe pulse up to several tens of µs.
5. Subtract the signal obtained when only the pump pulse was applied. In these experiments, set the V_DC, V_pump, and V_probe to -1.8 V, 400 mV and 210 mV, respectively. Set the relative delay sweep from -5 μs to 35 μs.
   NOTE: If the signal obtained from the previous step is well fit (R² > 0.80) by a single exponential decay function, then the lifetime of the transient state prepared by the pump pulse can be extracted from the fit.

6. Time-resolved STM Measurements of Excitation Dynamics

1. Send a train composed of pump pulses set at a fixed bias (such that greater than V_DC + V_pump > V_thr) with a repetition rate of 25 kHz. In these experiments, set the V_DC and V_thr to -1.8 V and -2.0 V, respectively. Set V_pump between 220 and 450 mV.
2. Sweep the duration of the pump pulse from several nanoseconds to several hundred nanoseconds.
3. Send a train of pump and probe pulses. The probe pulses should have an amplitude smaller than the pumps and comparable to the range at which hysteresis occurs (V_probe < V_pump, V_probe + V_DC≈V_hystersis). In these experiments, set V_probe to 210 mV.
4. Subtract the signal obtained when only the probe pulse was applied.
   NOTE: If the signal obtained is exponential, it indicates that the pump pulse is preparing the transient state (ionized dopant) at a rate that can be extracted from the fit (R²>0.80). The protocol described above is specific to the experiments and equipment described herein. There are many potential avenues for readers to customize their own experimental setup for studies of other systems. For example, the general techniques are not limited to cryogenically cooled STMs; any tip material can be used, and they do not require nitrogen etching. Furthermore, a suitably programmed arbitrary function generator could be used to generate double-pulse waveforms, which would negate the need to sum two independent channels. Lastly, lower bandwidth cabling could be used³¹.

Wyniki

The results presented in this section of the text have been previously published^15,16. Figure 3 illustrates the behavior of an example selected DB with conventional STM. A conventional I(V) measurement (Figure 3A) clearly depicts a sharp change in the conductance of the DB at V_thr = -2.0 V. This behavior is also observed in STM images taken at -2.1 V (...

Dyskusje

The variant of TR-STS in which the pump pulse is not applied is comparable to conventional STS, except that the system is being sampled at a high frequency rather than continuously. If the durations of the probe pulses are appropriate (>Γ_LH), the TR-STS signal acquired without the pump pulse can be multiplied by a constant proportional to the experiment's duty cycle to coincide exactly with a conventional STS measurement. This is only possible because the measurements are made without the...

Materiały

Name	Company	Catalog Number	Comments
Low Temperature Scanning Tunneling Microscope	Scientaomicron		Custom-made with 500MHz bandwidth wiring
Arbitarary Function Genorator	Tektronix	AFG3252C
RF Power Splitter/ Combiner	Mini-Circuits	ZFRSC-42-S +
RF Switch	Mini-Circuits	X80-DR230-S +
Non-Contact Infrared Pyrometers	Micron Infrared	MI 140