Phase-Resolved Detection of Ultrabroadband THz Pulses inside a Scanning Tunneling Microscope Junction

Coupling phase-stable single-cycle terahertz (THz) pulses to scanning tunneling microscope (STM) junctions enables spatiotemporal imaging with femtosecond temporal and Ångstrom spatial resolution. The time resolution achieved in such THz-gated STM is ultimately limited by the subcycle temporal variation of the tip-enhanced THz field acting as an ultrafast voltage pulse, and hence by the ability to feed high-frequency, broadband THz pulses into the junction. Here, we report on the coupling of ultrabroadband (1–30 THz) single-cycle THz pulses from a spintronic THz emitter (STE) into a metallic STM junction. We demonstrate broadband phase-resolved detection of the THz voltage transient directly in the STM junction via THz-field-induced modulation of ultrafast photocurrents. Comparison to the unperturbed far-field THz waveform reveals the antenna response of the STM tip. Despite tip-induced low-pass filtering, frequencies up to 15 THz can be detected in the tip-enhanced near-field, resulting in THz transients with a half-cycle period of 115 fs. We further demonstrate simple polarity control of the THz bias via the STE magnetization and show that up to 2 V THz bias at 1 MHz repetition rate can be achieved in the current setup. Finally, we find a nearly constant THz voltage and waveform over a wide range of tip–sample distances, which by comparison to numerical simulations confirms the quasi-static nature of the THz pulses. Our results demonstrate the suitability of spintronic THz emitters for ultrafast THz-STM with unprecedented bandwidth of the THz bias and provide insight into the femtosecond response of defined nanoscale junctions.

Sampling the THz voltage across the STM junction via THz-induced modulation of the photocurrent requires that the photocurrent reacts quasi-instantaneous on the time scale of the applied THz field. However, several effects such as photoelectron propagation in the tip-sample gap or long-lived hot carriers in the photoexcited tip or sample may disturb the measurement and have to be ruled out. As discussed by Yoshida et al., 1 THz-Streaking back into the tip can be circumvented by operating at high DC bias and low THz fields a , whereas lifetime effects from hot carriers can be neglected when operating in the regime of multiphoton photoemission above the potential barrier. Another effect, not considered by Yoshida et al., 1 is space charge, that is, the Coulomb repulsion experienced by electrons in a photoelectron cloud containing multiple electrons. Due to the nanometer-sized volume at the tip apex, even a low number of electrons can lead to high photoelectron densities, leading to strong Coulomb interaction between the photoelectrons. As a consequence, 'leading' photoelectrons at the front of the space charge cloud are accelerated away from the tip faster than by the DC acceleration alone, whereas 'backside' photoelectrons close to the tip surface are accelerated back towards the tip by the space charge cloud in front of them. Those back-accelerated electrons can then be steered back into the tip by THz fields much lower than the DC field due to the described effect of space charge acceleration. Moreover, these photoelectrons are not rapidly accelerated away from the tip as expected from the large inhomogeneous DC field at the tip, but can remain a significant amount of time close to the tip surface. Hence, they not only experience the instantaneous THz field at the time of photoemission, but can be steered back into the tip by a THz field at a much later delay within the THz pulse, i.e., the overall process is not instantaneous and can act as an effective low pass filter distorting the measured THz waveform.
In Figure S1 we plot THz waveforms measured in the STM junction at 1 µm gap distance and 8 V DC bias for different photocurrents. The applied THz voltage is 0.12 V at the maximum peak. As can be seen from Figure S1a), the measured waveform clearly depends on the number of electrons excited per laser pulse, with a most pronounced deformation at the beginning of a It should be noted that, even at low THz and high DC fields, the THz field always affects the photoelectron trajectories in the tip-sample gap, depending on the emission time and flight time of the photoelectrons with respect to the arrival and duration of the THz pulse. As we detect the total current and average over all energies and arrival times of the photoelectrons, our measurement is not sensitive to small modulations of the electron trajectories, but is only sensitive to a reduction of the total electron yield by THz-streaking back into the tip. the pulse before the main half cycle. This is accompanied by a reduction of the THz bandwidth as plotted in Figure S1b), revealing the effective low pass filtering due to space charge.
Moreover, Figure S1c) shows that in this regime the shape of the THz waveform also depends on the THz polarity. These deformations and deviations with polarity are most pronounced at the beginning of the THz pulse. This is not surprising as those photoelectrons excited shortly before the arrival of the main pulse experience the strongest THz field, whereas the photoelectrons excited at later times only experience the ringing oscillations after the main THz cycle. For even higher electron numbers, severe waveform distortions are observed as plotted in Figure S1d), with distortions and polarity dependence occurring over the full THz waveform.
Understanding the exact waveform deformations in the presence of space charge requires a detailed analysis of the dynamics of the photoelectron cloud in the tip-sample gap and is beyond the scope of this work. At this point, we conclude that elimination of space charge contributions is crucial for reliable THz waveform sampling by the employed technique as described in the S5 main manuscript. In our experiments we proof the absence of space charge effects to our measurements by systematically reducing the number of photoelectrons to a value at which no waveform distortions and no spectral filtering is observed by further lowering the photocurrent.
These observations point out the importance for a careful consideration of non-instantaneous effects potentially disturbing the sampled THz voltage transient.

Dependence of THz waveforms on DC bias and THz voltage
The propagation of photoelectrons in the tip-sample gap (and potentially back to the tip) depends on the combined forces of the DC field, the THz field, and space charge interactions.
Moreover, waveform distortions from photoassisted tunneling of electrons close to the Fermi level with longer lifetimes may be expected to contribute at high DC bias and short gap distances. As the lifetime of hot electrons depends on the energy window sampled by the THz pulse, distorting effects from hot carriers should also depend on the DC bias and THz voltage amplitude. Hence, undistorted instantaneous THz near-field sampling requires that the measured THz waveform does not depend on the DC bias and incident THz field strength within the linear range of the IPh-V slope (see Figure 1c)). Figure S2b) shows THz waveforms measured for three different DC biases with the corresponding IPh-V curve plotted in Figure   S2a). All waveforms exhibit the same shape and voltage amplitude as expected from the linear slope of the I-V curve. Figure S2c) shows THz waveforms measured for two incident THz amplitudes at 10 V bias, and again we find that the sampled THz waveform does not depend on the applied THz voltage. In addition to the constant THz waveform observed for opposite THz polarities, as plotted in Figure 1d) in the main manuscript, these results confirm the interpretation and validity of our sampling approach.

Antenna model
The model used to fit the tip antenna response in Figure 2 treats the tip as an RLC electronic circuit with resistance R, capacitance C and inductance L connected in series. In this very simple model, the incident THz electric field applies a voltage to the antenna that leads to a current induced inside the antenna of

Measurement of incident THz electric fields
To analyze the THz electric field incident to the STM junction, we pick the THz beam before Hence, if ℎ det ( ) is known, the THz electric field can be obtained by deconvolution of the EOS signal with the detector response, which depends on the properties and thickness of the electrooptic medium and the sampling pulses. After calculation of ℎ det ( ), the deconvolution is performed numerically as described in more detail in references 4,5 . Figure S3a) shows the deconvoluted waveforms of the THz electric field incident on the STM tip for the four STE positions plotted in Figure 3b). As mentioned in the manuscript, the measured EOS amplitude at frequencies >5 THz is comparably low, leading to increased noise at higher frequencies in the broadband deconvolution process. This can be mitigated by using thinner detection crystals.
However, echos in the first picoseconds caused by reflections inside the detection crystal can complicate the analysis of the transfer function in this case.

Dependence of THz spectra and amplitude on STE position
As discussed in the main manuscript, the THz spectra can vary slightly with STE position. Figures S4a) and S4b) show the THz amplitude spectra of the tip-enhanced THz waveforms plotted in Figures 3b) and 4b), respectively. We observe a slight bandwidth reduction upon moving the STE away from the NIR focus. We assign such variation of the THz spectrum to the imperfect imaging of the THz focus onto the tip at STE positions outside the NIR focal plane. The variation of the peak THz voltage amplitude with STE position ∆ STE is shown in Figure S4c). Maximizing simultaneously the THz amplitude and bandwidth requires a modified STE geometry optimized for our laser parameters, which is currently under investigation.

Estimation of tip-sample distance at the set point
The actual tip-sample distance under NIR illumination is estimated as follows: First, we estimate the change in gap distance due to removal of the photoinduced current contribution by measuring the prompt (~30 ms, which is the time required to reach again the set point current) reduction in gap distance upon blocking the NIR beam, which gives ∆ ph ≅ 1.6 nm. Second, we estimate the absolute distance of the 'dark' junction by measuring the z-dependence after thermal equilibration and extrapolation of the I-z-curve to the quantum conductance G0. This with respect to the tip axis as and is polarized linearly in the --plane, as indicated by the red arrows in Figure S5b). The simulation volume is cut in half along the y-direction (out of plane) according to the symmetry given by the THz beam direction to reduce the computational cost.
The materials properties of the tip and sample are determined from the complex dielectric functions of tungsten and silver, whose real and imaginary parts are calculated from the Lorentz Drude model with the Lorentz-Drude parameters taken from Rakic et al. 6 .
Figures S5a)-d) show spatial distributions of the tip-enhanced THz electric field in the tipsample junction for two THz frequencies. As can be seen from the color scales, the THz field enhancement increases significantly with lower THz frequencies and shorter gap distances. At S10 far distances b) and d), the strongly inhomogeneous THz near-field is confined to the tip apex with a spatial extent given by the tip radius, and decays rapidly with larger distances away from the tip surface. At nanometer distances much smaller than the tip radius, the field becomes spatially more localized inside the tip-sample gap, and becomes homogeneous like in a platecapacitor at the junction center ( , ) = (0,0). We obtain the THz-field-induced potential difference from the computed fields by line integration of the z-component of the THz nearfield along the center of the junction at ( , ) = (0,0). We confirmed that line integration along other pathways yields the same THz-induced potential difference.