We investigate the THz photo-response of nanometric-thick graphite flakes. The blueish curves in Figures 2a,b show the I-V characteristics of two graphite flakes of 14 to 4 nm in thickness, corresponding to 30 to 10 carbon-atom layers. The conductive AFM probe is kept at a fixed position on the flake. The behavior is slightly superlinear, which is a signature of a potential energy barrier for carriers in the tunneling process between platinum tip and the few-layer graphene. Thin layers of graphite and a metal typically form a low-resistivity ohmic contact Robinson et al. (2011); Xia et al. (2011); Lee et al. (2008), with a negligible potential barrier. In particular, if the work function of the graphene and the metal are different, a significant charge transfer occurs Giovannetti et al. (2008), facilitating the good electrical contact. In our case, the nanometric size of the contact given by the tip radius of the AFM probe makes the electric contact very sensitive to local defects, charge puddles, and contaminants on the sample surface. A small defect density at the nanojunction can easily induce the formation of a small potential barrier. This is especially true in AFM imaging in ambient conditions, where various types of nanometric organic and inorganic contaminants can be attached to the sample and the tip Rozhok et al. (2004). The tunneling barrier from the graphite flake to the tip dominates the I-V characteristic of the circuit. At higher applied voltages, the carriers can overcome the small potential barrier between metal-graphene nanojunction, leading to an increase in the conductivity and a superlinear behavior.

The red curves in Figures 2a,b show the voltage dependence of the THz photo-induced current, measured with a lockin amplifier, simultaneously to the cAFM measurements. We observe a THz-induced current signal at zero applied bias. This may be given both by the photo-thermoelectric effect and by THz rectification at the asymmetric AFM probe-graphite-gold junction Vicarelli et al. (2012); Bianco et al. (2015). In various previous works in the literature, it has been shown that the photo-thermoelectric effect in graphene is particularly strong Vicarelli et al. (2012). To get a quantitative estimate of the photo-thermoelectric current, we use the following approximate formula: I PTE = ( S G r − S P t ) ∇ T R , where S are the Seebeck coefficients of graphene and platinum, ∇ T is the electronic temperature gradient and R is the resistance of the device. Approximating the gradient with a total temperature difference Δ T , the amount of photo-induced current that we observe under zero applied bias is compatible with a local heating close to the tip apex of a few degrees, which is what is expected for this type of technique Schwartz et al. (2022), and which we already observed on graphene Venanzi et al. (2023). To further corroborate this hypothesis, we measure the THz-induced current as a function of the incident power. As shown in Figure 2c, the photo-induced current changes linearly with the THz power. This observation is compatible both with a photo-thermoelectric effect and the THz rectification. Moreover, we show in Figure 2d that the THz radiation induces a negative DC current.

By applying a bias voltage, we observe a nonlinear change of the photo-response that cannot be ascribed to the photo-thermoelectric effect. The photo-induced current shows a sharp positive slope at voltages close to 0.15 V and saturates at high applied voltages, both negative and positive. We rule out bolometric effect of graphite as the origin of the nonlinear dependence of the photo-induced current because the series resistance of graphite is very small given the nanometric thickness of the flake and the vertical transport through it. Rather, the resistance is dominated by the platinum-graphite nanojunction. In Figure 2 we report the second derivative of the I-V characteristics (see the Supplementary Material for further details). This quantity is proportional to the current due to THz-induced rectification Schlecht et al. (2019). As can be clearly seen from the figure, there is a good qualitative agreement between the d 2 I / d V 2 curve and the THz-induced current. This observation strongly corroborates that the THz-rectification is the main mechanism leading to the observed photo-induced current as function of voltage. The THz rectification at the junction may work similarly to a metal-insulator-metal (MIM) structure between the platinum tip, a narrow tunneling barrier, and the graphene. The asymmetric MIM structure leads to an asymmetric flow of current at the barrier, that results into a net DC current Citroni et al. (2022); Cocker et al. (2021). The net current depends nonlinearly on the applied voltage as expected. We finally note that using a basic parallel-plate capacitor approximation for our junction geometry, we can estimate a parasitic cutoff frequency which is well above our working frequency of 0.61 THz Casini et al. (2011), further supporting that the THz rectification is the dominant effect in the observed photo-induced current.

Finally, we mention the role of the geometric nanometric junction. In fact, as recently studied in graphene Zhu et al. (2013), our system can behave as a geometric diode where the nanometric platinum tip is the contact point, in contrast to the large gold substrate-graphene interface which acts as a counter-electrode. This represents a possible alternative explanation of the rectifying properties manifested by our sample.

We demonstrate the scanning capability of the setup by measuring AFM topography, cAFM, and THz-pcAFM maps (Figure 3) simultaneously. Figure 3a shows the AFM topography of the few-layer graphene flakes and Figure 3b shows a zoom-in in the 1 μ m × 1 μ m area where we measured the cAFM and THz-pcAFM maps. The edge of the few-layer graphene flake is barely visible in the AFM image. The flake thickness is about 2 nm as measured by AFM. The flake results to be 3-layer thick as confirmed by Raman spectroscopy (see Supplementary Material (SM)). We did not observe any evidence of damage induced by the AFM tip scanning in contact mode Goossens et al. (2012). Figures 3c,d are the cAFM and THz-pcAFM maps, respectively, measured under a small applied bias of about 2 mV to reduce the noise in the THz-pcAFM maps. In the cAFM map we observe a strong current when the AFM probe is on top of the gold substrate and a weaker current when the probe is on trilayer graphene, as expected. Conversely, in the THz-pcAFM we do observe a stable signal only when the probe is on the flake, while there is only noise when the probe is on gold. The photo-induced current signal under zero bias is likely given by the photo-thermoelectric effect. The high noise on gold comes from the high DC current that introduces noise into the lockin detection. From this trivial observation, we conclude that the technique is able to map nanometric inhomogeneities in a device that generates THz photo-induced current, e.g., a THz Schottky diode detector, and leads to complementary information with respect to conductive AFM.

We also note that during the AFM raster scan of the sample, sometimes the electrical contact between the conductive AFM tip and the sample is lost, as it can be observed on the top left corner of Figures 3c,d. The instability of the electrical contact is the main limitation of this technique, a problem discussed in the literature for cAFM Jiang et al. (2019). A critical feature for the stability of the electrical contact is the cleanliness of the sample. In our case, to avoid contaminants on the sample, we use ultra-flat gold freshly stripped from the template as substrate. The graphite flakes are obtained from a chemical vapor deposition-grown crystal. After peeling off a top layer of the graphite via mechanical exfoliation, the crystal is mechanically pressed on the gold without the use of tape or any other polymer. These leaves clean and flat graphite flakes of different thicknesses.

Interestingly, both the cAFM and the THz photo-induced current map show a strong increase of the current/photocurrent in a small region of the few-layer graphene flake, while the rest of the flake has a rather homogeneous current and THz photo-response. The area is about 150 nm wide, which is a dimension that is smaller than the diffraction limit of visible lasers. Techniques such as Raman spectroscopy that we used to probe the sample, typically average over these regions. We interpret this observation as a consequence of a charge puddle in the few-layer graphene flake Martin et al. (2008); Samaddar et al. (2016). Since we do not observe any clear corrugation in the AFM topography, the charge puddle is likely induced by charged impurities between the few-layer graphene and the gold substrate Zhang et al. (2009). It is fundamental to remark that the charge puddle results in an increase of both the DC current and photocurrent absolute signal. This means that the observed increase cannot be solely due to the photo-thermoelectric effect. In fact, a change in the Seebeck coefficient, which depends nonlinearly on the local charge density Castilla et al. (2019); Duan et al. (2016), would affect the photocurrent but not the DC current. As a consequence of this observation, we hypothesize that the main role of the charge puddle is to increase the number of available charges that can tunnel between the tip and the sample. Quantifying charge density from our measurements would be of crucial importance for the future refinement of the THz-pcAFM technique. However, for this kind of estimation it is essential to develop a precise quantitative model to describe the photocurrent. This quantitative analysis goes beyond the scope of this work, and would be indeed particularly tough without knowing some key parameters, such as the precise mechanisms governing the photo-induced current signal and the evaluation of the graphite flakes intrinsic doping level.

The observation of charge puddles in graphene has been obtained by means of scanning tunneling microscopy (STM) Zhang et al. (2009), single-electron transistor (SET) microscopy Martin et al. (2008) and other scanning probe microscopy techniques Zhao et al. (2021). In our experimental, the spatial resolution of our experimental setup is given mostly by the tip apex (20 nm of radius), enlarged by the current pathways that are not solely vertical. As a consequence, we are limited to observing charge puddles of about hundred of nanometers in size, different from what has been observed in STM experiments Zhang et al. (2009) where smaller charge puddles have been observed. Another important distinction lies in the sample cleanliness. In the STM and SET experiments performed in the aforementioned works, charge puddles are observed in an ultra-high vacuum chamber, after a proper thermal annealing of the sample. By doing so, the amount of adsorbates is significantly reduced with respect to our ambient temperature and pressure conditions. This may lead to significant changes in the origin of the puddles that in our case may be due to deep potential traps caused by adsorbates. Finally, while both SET and STM techniques involve a negligible flow of current between tip and sample, our THz-pcAFM setup relies on a good electrical contact. This implies a significant amount of charge transfer, and hence a change of the local Fermi energy of few-layer graphene depending on the relative work functions of platinum and graphite Giovannetti et al. (2008). In this sense, our approach is more perturbative and less accurate to quantify the charge density in the puddles.