Phase-resolved terahertz self-detection near-field microscopy

: At terahertz (THz) frequencies, scattering-type scanning near-field optical microscopy (s-SNOM) based on continuous wave sources mostly relies on cryogenic and bulky detectors, which represents a major constraint for its practical application. Here, we devise a THz s-SNOM system that provides both amplitude and phase contrast and achieves nanoscale (60-70nm) in-plane spatial resolution. It features a quantum cascade laser that simultaneously emits THz frequency light and senses the backscattered optical field through a voltage modulation induced inherently through the self-mixing technique. We demonstrate its performance by probing a phonon-polariton-resonant CsBr crystal and doped black phosphorus flakes.


Introduction
Terahertz (THz) frequency electromagnetic waves (30-300 µm wavelength) can resonantly interact with fundamental excitations of molecules and solids and thus offer an ideal tool for the optical characterization of emerging low-dimensional materials and biological-systems. Scattering-type scanning near-field optical microscopy (s-SNOM) displayed an exceptional potential for nanoscale imaging of material properties [1] as has been demonstrated also at 2.5 THz [2]. In s-SNOM, an incident light beam is focused on a sharp atomic force microscope (AFM) metallic tip strongly confining the radiation to the near-field region of its nanometric apex. Nanoscale resolved (10 -100 nm) optical images can be retrieved by analyzing the scattered radiation as a function of tip position, placing the tip in close proximity to the sample surface. Coherent imaging, i.e. detection of amplitude and phase contrast, is crucial to get information on the real and imaginary parts of the dielectric response and hence several interferometric approaches have been developed allowing amplitude and phase resolved s-SNOM imaging in the visible and infrared spectral ranges [3,4]. Solutions based on fs pulsed laser sources and electro-optic sampling detection in the THz range [5,6] or solutions based on microwave circuitry in the sub-THz range [7] have been reported. However, in spite of a strong demand to extend in the far-infrared the spectral coverage of coherent nano-imaging based on continuous-wave (CW) compact sources, conventional interferometric techniques [2,8,9] have suffered from the poor dynamic range of cryogenically cooled bolometric detectors needed to measure the typically small s-SNOM signals [2]. Progress in coherent THz nanoimaging would therefore greatly benefit from compact, room-temperature operating and fast detection systems for THz frequency operation.
Here we tackle the problem by conceiving a simple, potentially fast and compact s-SNOM system based on a THz quantum cascade laser (QCL) operated CW in the self-detection (SD) mode. A similar approach has been recently exploited to produce THz images with subwavelength spatial resolution [10,11]. In this work, we demonstrate amplitude-and phaseresolved background-free SD-s-SNOM imaging with 60-70 nm spatial resolution comparable to the scattering tip size, providing a key step forward to make THz nanoscopy a widely used tool. Our approach, based on a simple 2-parameter fitting of self-mixing interferograms, outperforms by far all previously reported attempts based on the use of self-mixing and either attaining incoherent near-field s-SNOM imaging, or using a rather complex 6-parameter fitting procedure for amplitude-like and phase-like images at diffraction-limited resolutions (> 100 µm) [12].

The general concept of SD-s-SNOM
In our experimental configuration, we use the radiation source, a THz quantum cascade laser (QCL) [13], (see Experimental Methods) not only for illuminating the tip but also for detecting the radiation field scattered from the tip, therefore avoiding the use of bulky THz cryogenic detectors. The near-field scattering is measured through the voltage change (ΔV) across the electrical contacts of the THz QCL, induced by the self-mixing effect [14,15]. This effect is based on the reinjection of a small fraction ( 4 2 10 10 − − ) of the emitted field that coherently interferes within the laser cavity.
The inherent stability of QCLs against optical feedback [16] has been recently exploited in a number of self-mixing interferometry (SMI) configurations, providing an interesting method to control the emission of THz QCLs by reconfigurable photo-generated anisotropic metamaterials [17], to trace the free carrier distribution in a semiconductor target [18], and to map the real and imaginary refractive index of polymeric materials [12], for example.
In our system ( Fig. 1(a)), the 2.7 THz QCL radiation is focused onto the apex of a Pt tip, which is sinusoidally dithered, normally and in close proximity to the sample surface at frequency Ω. The scattered field E s is collected by a parabolic mirror and focused back onto the QCL front facet along the same incident optical path, inducing the changes in the QCL voltage ( V Δ ) (see details on the experimental arrangement in Appendix 1). Hence, V Δ measures the optical response of the sample. To quantify this properly, we relied on the wellestablished Lang-Kobayashi (LK) model [19], and extended it to encompass a complex sample permittivity. In the very weak feedback limit, the voltage change at the QCL terminals can be written (see Appendix 2): where s and ϕ are the amplitude and phase, respectively, of the ratio between scattered and incident electrical fields of THz radiation, 0 2 / c ω π λ = is the unperturbed laser frequency, and L is the laser-to-tip distance which varies by the tip dithering and by the piezoelectric mirror (PZM) displacement. By using an optical attenuator (A), the feedback remains sufficiently low to keep the system in the validity range of Eq. (1). In this regime, the proposed experimental layout is similar to a single-arm homodyne interferometer [14], in which the laser output facet and the tip define the external cavity of our self-mixing interferometer.
Equation (1) contains all necessary physical quantities to predict observable amplitude and phase contrasts in SD-s-SNOM images. Although in conventional s-SNOM, the phase contrast is extracted by analyzing the detector signal variations induced by changing the optical path length of the reference beam of an interferometer, in SD-s-SNOM the amplitude and phase information is obtained from the analysis of the V Δ as a function of ΔL as explained below. Fig. 1. Self-detection scattering type near field optical microscope with nanometer resolution at terahertz frequencies. (a) Schematic diagram showing the experimental arrangements. The same optical components (two paraboloid mirrors, an attenuator (A), one fixed mirror and one piezo-actuated mirror (PZM)) focus the THz beam emitted by the QCL onto the apex of a Pt tip sinusoidally dithered at frequency Ω in close proximity to the sample surface and feeds the scattered radiation back into the QCL cavity to produce self-mixing. The distance L ≈60 cm between the QCL front facet and the tip can be varied by translating the PZM. (b) Self-mixing induced modulation from a Au surface of the voltage drop across the QCL obtained using the experimental arrangement (a) and measured by a lock-in amplifier to the n = 3 demodulation order, as a function of ΔL. The dots are the experimental data and the solid line is the bestfitted curve obtained using Eq. (2). The ΔV 3 signal show the expected λ/2 periodicity characteristic of self-mixing. (c) AFM topographic and (d) 3rd harmonic near-field THz imaging of the Au-on-silicon sample at a fixed mirror position. (e) Edge response profile extracted from (d) by averaging seven adjacent rows of pixels along the green horizontal line. The inset shows the corresponding spatial derivative, which has been fitted by a Gaussian function (red curve) with a full-width-at-half-maximum of 65 nm.

The implementation of SD-s-SNOM
In our arrangement, as is typical in s-SNOM for effectively distinguishing the near-field scattering from background scattering [1], V Δ is demodulated at low-order harmonics of the dither frequency using a lock-in amplifier, and the measured harmonic amplitudes The experimental n V vs L Δ Δ curves (e.g. ΔV 3 in Fig. 1(b)) demonstrates that our SD-s-SNOM system behaves similarly to an external interferometer and produces approximately sinusoidal interference fringes, as predicted by Eq. (2). Figures 1(c)-(d) provide direct proof that the SD-s-SNOM signal is sufficiently stable for homodyne THz imaging.

Deep sub-wavelength in-plane spatial resolution
Comparison between the AFM ( Fig. 1(c)) and the self-detected ( Fig. 1(d)) images of a gold film deposited on a SiO 2 -coated silicon substrate demonstrates the capability of our THz imaging system to achieve an in-plane resolution comparable with the employed tip apex sizes (50-70nm). Figure 1(e) shows the ΔV 3 signal collected along the green horizontal line in panel 1d traversing the edge of the gold-coated region. The inset of Fig. 1(e) shows the corresponding first-order spatial derivative. From the full-width-at-half-maximum of the Gaussian curve interpolating the derivative function (red curve in the inset of Fig. 1(e)), we can retrieve a remarkable spatial resolution of σ x = 68 nm along the horizontal (fast) axis, corresponding to ∼λ/1500.

Amplitude and phase contrast imaging capability
The phase-contrast at THz frequencies between gold and SiO 2 is negligibly small, as is usual for all combinations of non-resonant materials [9]. Hence, to provide a proof-of-principle of the amplitude and phase contrast imaging capability of the SD-s-SNOM, we selected a polar crystal (CsBr), which exhibits a strong phonon-polariton (Reststrahlen) resonance in the range 2.2 -3.3 THz [20]. We simulated the SD-s-SNOM near-field amplitude (s 3 ) ( Fig. 2(a)) and phase (φ 3 ) signals ( Fig. 2(b)) for both CsBr and Au as a function of wave number, using a single phonon oscillator model in the framework of the finite dipole model (see Appendix 2). Our findings show that, although Au exhibits the expected flat amplitude response and negligible phase throughout the THz frequency range investigated, CsBr shows a four-times stronger s 3 peak at 90 cm −1 , and a large phase signal φ 3 ≈150° at this wave number. Hence, we prepared a suitable sample comprising a CsBr crystal, coated with a thin (100 nm) Au film, and selected a THz QCL that operated single mode in continuous wave at 90.3 cm −1 (2.7 THz). These choices should ensure a strong near-field amplitude and phase contrast at CsBr/Au steps obtained by mechanically scratching the Au thin film to expose the underlying CsBr surface. Figures 2(c)-2(d) compare the AFM topography ( Fig. 2(c)) and the SD-s-SNOM image ( Fig. 2(d)) of a CsBr/Au step, simultaneously obtained while raster-scanning in the x-y plane underneath the tip and recording 3 V Δ at each pixel. To show the L dependence of both the 3 V Δ signal and the optical contrast between CsBr and Au, we changed the optical phase over a 2π range in π/3 steps ( 9.25 L m μ Δ = ), every Δy = 0.5 μm during the acquisition. We thus observe 13 striped regions in Fig. 2(d). In order to retrieve amplitude and phase from the collected optical images, each line in Fig. 2(c) and 2(d) were preliminarily horizontally shifted to straighten the CsBr/Au edge. Then the signal ΔV 3 was vertically averaged in each stripe and associated with the corresponding ΔL value. Finally, sinusoidal functions (Eq. (2) were fitted to the <ΔV 3 > vs ΔL curves to extract the amplitude s 3 and phase φ 3 as a function of the horizontal distance from the CsBr/Au edge ( Fig. 2(f) and Fig. 2(g)).
Then the height (Fig. 2(e)) and the signal ΔV 3 were vertically averaged on each stripe and associated with the corresponding L value. From the steep rise of the optical signal at the edge the Au film on the CsBr crystal, we can estimate the optical spatial resolution. To address the issue we have followed the same procedure detailed for Fig. 1(e). The full-width-at-halfmaximum of the Gaussian curves interpolating the derivative function of the amplitude and phase signals (red curves in the insets of Figs. 2(f) and 2(g)), give estimates of the spatial resolutions: 54 nm and 60 nm for the amplitude and phase signals, respectively. In Fig. 3, we demonstrate the strong sensitivity of SD-s-SNOM to changes in the amplitude and phase of the effective polarizability of the coupled tip-sample system. For this, we exploited the strong dependence of s 3 and φ 3 on the tip sample distance z (see Appendix 2) [21] and recorded ΔV 3 during approach curves obtained by progressively stepping the PZM position by ΔL = 9.25 μm. For each value of z, s 3 ( Fig. 3(a)) and φ 3 (Fig. 3(b)) were extracted by fitting sinusoidal functions (Eq. (2) to the ΔV 3 vs L curves. The prevalent near-field nature of 3rd harmonic signals is shown by the monotonic decrease of both the amplitude and phase signals for increasing z, with s 3 becoming negligible and φ 3 saturating for z > 60 nm. The approach curves of Fig. 3(a)-3(b) are nicely reproduced by the calculated near-field amplitudes (s 3 ) (Fig. 3(c)) and phases (φ 3 ) (Fig. 3(d)) of CsBr and Au, solving the full set of transcendental coupled LK equations, with no approximations, at each experimental L value (see Appendix 2). Similarly, our technique is expected to measure changes of the real and imaginary part of the effective polarizability as a function of the incident laser wavelength.

Application for 2D material imaging
Finally, we demonstrate the capability of SD-s-SNOM to image doped van der Waals layered materials. We selected black phosphorus (BP), an emerging layered semiconductor, which has been recently demonstrated to be a suitable material system in the THz for photodetectors [22,23], and near-field optical probes [25]. Here we study BP flakes with different thickness and carrier densities (see the estimate of free carrier density in black phosphorus flakes in Experimental Methods section) that were transferred onto a Si/SiO 2 substrate by mechanical exfoliation [23,26]. A thin (5 nm) SiO 2 protection layer was deposited via sputtering to encapsulate the material system and avoid degradation under ambient exposure, ensuring that flakes remained clean and stable for several months [22,23].  Fig. 4(e) and 4(f) were preliminarily horizontally shifted to straighten the BP/SiO 2 edge, in analogy to the analysis performed on the CsBr/Au sample (Figs. 2(c)-2(g)). Then, the height ( Fig. 4(g)) was vertically averaged on each stripe and associated with the corresponding L value. From the striped image of Fig. 4(f) we extracted the near field amplitude s 3 (Fig. 4(h)) and phase φ 3 (Fig. 4(i)) profiles at the BP/SiO 2 step edge. Apparently, no phase contrast between BP and SiO 2 is observed, due to the off-resonance excitation of both materials, as expected since in both cases optical phonon resonances fall in the mid-IR [27]. Incidentally, the detected nearfield scan results highly sensitive to the presence of oxidation-induced clusters on the surface of the investigated BP flake, which causes an abrupt change of the detected optical signal.

Conclusion
In conclusion, we have demonstrated a detector-less s-SNOM system operating at THz frequencies that provides both amplitude and phase contrast with deep sub-wavelength (60-70 nm) in-plane spatial resolutions. These achievements are expected to have a profound impact on the flourishing field of QCL-based THz imaging, in which near-field coherent detection has been missing, so far. The ability of SD-s-SNOM to resolve both the amplitude and phase of the THz field opens up the possibility of mapping the complex permittivity of a target using QCLs with a high spatial resolution, down to a few tens of nanometers. The proved sensitivity to thin-layered samples of our novel THz SD s-SNOM opens the way to further investigations of resonant 2D-materials and combined Van der Waals heterostructures, with potential impacts in plasmonics and optoelectronics.
Future experiments could make use of the tunable bandgap of BP [28] to control its modeactivation energy as well as exploiting resonant excitation to retrieve plasmonic features. The in-plane asymmetry of BP [29] may also allow for further tunability. Looking forward, complex heterostructures that combine BP with graphene, transition-metal dichalcogenides and hexagonal boron nitride have the potential to provide a robust technological platform for THz nanophotonics and ultrafast plasmonics.
Further development of SD-s-SNOM will also benefit from the availability of broadly tunable THz QCLs [30] and THz QCL combs [31,32]. Also, the inherent ultrafast response of QCLs to optical feedback perturbations, associated with their unipolar nature and the pslong lifetimes of inter-subband transitions, raises a compelling perspective for new timeresolved hyper spectral THz imaging systems with deep sub-wavelength spatial resolutions (< 100 nm or λ/1000).
The unique features of SD-s-SNOM, i.e. the inherent field sensitivity and compactness, can in principle lead to further elimination of intermediate optical elements and true access to near-field radiation, opening the way to many applications in fields ranging from biosensing to quantum optics [33]. Also, a full integration of sample and nanoscopy set-up inside the same cryostat may become possible, thereby allowing to exploit in situ the potential of coherent THz nanoscopy at cryogenic temperatures and under high magnetic fields, for understanding fundamental excitations, such as phonons, magnons, polaritons, and phase transitions in exotic semiconductor and organic nanostructures, superconductors, multiferroics, and metamaterials, just to mention a few examples.

Experimental arrangement
A bound-to-continuum single longitudinal mode THz QCL emitting, at a wavelength λ = 111 μm (2.7 THz, 90.3 cm −1 ) with a surface plasmon waveguide, was grown by molecular beam epitaxy employing a GaAs/Al 0.15 Ga 0.85 As heterostructure on a nominally undoped GaAs substrate. The 10-μm-thick active region was embedded between a doped ([Si] = 3.0 × 10 18 cm −3 ) 700-nm-thick GaAs bottom layer, and a doped ([Si] = 5.0 × 10 18 cm −3 ) 200-nm-thick GaAs top contact layer. The GaAs semi-insulating substrate was lapped down to a thickness of 150 µm. Laser bars 1.5 mm long and 150 μm wide were cleaved and mounted on a copper bar.
The QCL was mounted in a liquid helium continuous-flow cryostat fitted with a polymethylpentene window, and maintained at a fixed heat sink temperature of 15 K. To maximize the sensitivity to coherent optical feedback [14,15], the QCL was driven in continuous wave at a current I = 450 mA (J = 132 A/cm 2 ) which is just 5% higher than the threshold current density (J th = 125A/cm 2 ), using a highly stable current generator (Lightwave Electronics, mod. QCL 2000). The emitted THz beam was collimated using a 90° off-axis parabolic mirror with an effective focal length of 50 mm, and reflected by a mirror mounted on a piezo-controlled translator stage into the entrance optical port of a commercial near-field microscope (Mod. NeaSNOM, Neaspec, Martinsried, Germany). A 5 dB substrate-free, metal-mesh diffractive attenuator (Mod. 224, Lasnix, Berg, Germany) was placed in the beam. In the microscope, a second paraboloid mirror with an equivalent focal length of 25 mm focused the beam onto a Pt tip (Bruker, mod. RMN-25PT300) having a nominal apex radius of 50 nm, a shank length of 80 μm, and which was sinusoidally dithered at its resonant frequency Ω = 15380 Hz. The laser polarization lay in the plane containing the tip in order to induce an oscillating dipole in the tip efficiently. The average optical path length from the QCL front facet to the tip was L ≈ 60 cm. The scattered radiation was collected by the same focusing paraboloid mirror and coupled back into the laser cavity along the same incident optical path. The voltage modulation across the QCL terminals produced by the self-mixing effect was pre-amplified using a low-noise amplifier (DL Instruments, mod. 1201) and demodulated up to the highest harmonic order (n = 5) allowed by the electronic card of the NeaSNOM.
The density of free carriers n e in the black phosphorus samples was estimated by performing transport analysis on field effect transistors FET having flakes with identical thickness as active channels [20-22, 24, 25]. n e is connected with the FET threshold voltage Here, i E and s E are the incident and scattered fields, respectively, and p r is the far-field Coherent optical feedback in a semiconductor laser cavity is described by the Lang-Kobayashi (LK) model [14,15,35], in which two coupled differential equations describe the evolution of the excited state population N(t) and the optical field E(t). The crucial feedback contribution in this model is provided by an additive term, representing the retarded field reentering the laser cavity.
The LK equations read [35]: where N 0 is the carrier density at transparency, G n is the gain coefficient, τ = 2L/c, τ p = 30 ns and c τ = 35 ns are the photon life time and the cavity round trip time inside the QCL cavity, respectively, and e τ is the carriers decay time (1 10 ps − ), J is the current density, η is the internal quantum efficiency, d is the active layer thickness and q is the electron charge, γ is the feedback strength parameter, i.e. the fraction of the back-scattered field that efficiently couples with the lasing mode. A value 1.5 H α = for the linewidth enhancement factor was used, which is appropriate for a THz QCL close to threshold [36,37].
In our theoretical approach, we consider a feedback field provided only by the tipscattered field s E and neglect any contribution from the background field. The latter is a component well-known in s-SNOM that can be experimentally suppressed by recording signals at higher harmonics of Ω. Considering that the feedback path 2L depends on the oscillating tip position, we assume: , as appropriate under our experimental condition, the perturbation induced on the laser frequency can be neglected and one can assume that the frequency of the retro-injected laser coincides with 0 , ω radically simplifying the self-mixing formalism [14][15][16]. In fact, the sinusoidal line shapes unveiled in the experiments of Fig. 1 confirm that the system is in the very weak feedback self-mixing regime [19], with estimated values of the feedback coefficient C ≈ 0.03 -0.1. Hence, the LK equations are solved for the steady-state value of the carrier variation N Δ induced by the self-mixing (with respect to the free running laser value at threshold), assuming the tip and the PZM dynamics are slow with respect to field and carrier timescales: noting that V Δ is proportional to N Δ [16], we find the following expression: is the laser-target distance which may vary, if the PZM is moved, on time scales much longer than the tip period 2π/Ω.. Moreover, we approximate 0 L L ≈ in the following, since in our experimental configuration A z λ  . By introducing the spectral representation of the scattered amplitude in Eq. (4), we can rewrite the steady state equations accordingly, where the RHS term of Eq. (7) becomes: