a Schematic of threshold power of a photo-detector. The minimum detectable powers of two photo-transistors are determined by their detectivity (P1 & P2). Another detector with optimized limit of detection present better threshold power (P3), even though its responsivity (R) is lower. b Origin of threshold power in a photo-transistor (defined by Source (S), Drain (D) and Gate electrodes). c channel charges below (regime 1) and above (regime 2) threshold. d The structure of a photo-transistor under light (??) illumination, E_c represent the conduction band energy. Band-bending defines the surface potential (ψ_S), ?????? represents quantum capacitance, q is the elementary charge; e surface potential and charge density as functions of incident light power for the photo-transistor channel, the arrow indcates the photocurrent flow.

a Schematic of device structure, showing BP channel, graphene contacts (with source and drain electrodes V_s and V_d), bottom gate (V_bg), and two partial top gates (V_tg1 and V_tg2) as well as the dielectric layer of hexagonal Boron Nitride (hBN). b Scanning electron microscopy (SEM) photograph (with false color) of a representative device in (a), the colored dashed line indicate different layers of the device: yellow for bottom hBN and white for top hBN, while the solide white line indicates the bottom gate, the scale bar is 10 ?m. c, d illustrate the schematic band-diagram profiles for the principal operation of different photo-transistors. Specifically, they depict the dark (dashed blue line) and light-illuminated (red line) cases in a photo-tunneling transistor and a conventional field-effect transistor (FET), respectively. With a band-to-band tunneling (thermionic) charge injection mechanism, the device has a steep (smooth) Sub-threshold Swing (SS) and response (does not respond) to weak light, I_d represents the source-drain current, E_f denotes the fermi level and E_v indicates the valence band energy of the BP channel. e Room-temperature transfer curves of the device in (b) at different modes. Without top gate (V_tg), the device is an ambipolar FET (black curve). With V_tg1 = 6 V (V_tg2 = −6 V), the device is configured as p-type (n-type) tunneling-transistor V_ds = 0.1 V. f, g are enlarged transfer curves for tunneling transistor mode. The dashed triangles show subthreshold slopes of SS = 60 mV/dec for comparison.

Concretely, the top gate defines a highly doped region in the channel underneath. The distinct doping difference between the gated region and the rest intrinsic part results in a large energy barrier inside the channel. Applying a bottom gate voltage mainly shifts the Fermi level of the intrinsic part. At sufficiently large gate bias, band-to-band tunneling occurs when a proper tunneling window is built up (i.e., the valence band of the intrinsic region aligns with the conduction band of the highly doped gating part for p-type transistor, or vice versa for n-type case). Replacing thermionic emission by the cold carrier injection mechanism of band-to-band tunneling, the device is able to present a low SS that may break the Boltzman tyranny²⁵. We measured the transfer curves of the same transistor configured (by top gate) in different modes. (We have fabricated five devices as summarized in Supplementary Table I and consistently observed similar results). According to the barrier direction, the device can be configured to both p-type and n-type, which is suitable for complementary logic application. The reconfigurable working mechanism is summarized in Supplementary Note 3. The results are compared in Fig. 2e (V_ds = 0.1 V, same for the following measurements unless specified). Without top gate, the device is a conventional ambipolar FET. The lowest SS (for one decade increase of source-drain current) is ~300 mV/dec. In contrast, with positive Vtg₁ (negative Vtg₂), the device is a p-type (n-type) tunneling FET. The lowest room-temperature SS is 48.5 (given in Fig. 2f) and 43 mV/dec (given in Fig. 2g), respectively. It should be noted that due to different contact barrier, electron current is much smaller than the hole current (it also explains the asymmetric transfer curve in Fig. 2d. see Supplementary Note 4 for details). And this asymmetry does influence our results, as we mainly focus on the p-type case which does not suffer from contact problem.

To further confirm the band-to-band tunneling nature of the sharp SS, we performed temperature-dependent transport measurements for the device. Fig. 3a, b summarize the transfer curves at different temperature with and without top gating (positive Vtg₁). We also extracted the lowest SS for each curve and plot the results in Fig. 3c. Obviously, without top gate-defined junction, the SS presents a typical thermionic emission feature, that is, linearly grows with increasing temperature. In stark contrast, SS is nearly unchanged with increasing temperature when the channel is top-gated. This temperature-independent switching clearly evidences band-to-band tunneling^26,27,28 occurs when a tunnel window is opened by back gate modulation. But owing to some non-ideal effects, it should be noted that the device is likely operated by a tunneling dominated mixed mechanism. For example, under high V_bg bias, the p-n junction barrier may be too low to block all thermal injection. It results in a certain portion of thermal current (rather than a pure tunneling current), especially at high current density case. However, this characteristic does not affect the weak light harnessing proposed in this manuscript.

Fig. 3: Band-to-band tunneling properties of the photo-tunneling transistor.

a, b Temperature-dependent transfer curves for the device configured in tunneling transistor (a) and conventional FET modes (b), V_ds = 0.1 V. c Measured lowest SS extracted as functions of temperature. While for conventional FET (red), SS linearly increases with temperature and obeys a thermionic injection rule; the SS of tunneling transistor (black) is temperature independent, exhibiting a band-to-band tunneling nature, the dashed lines are described as indicating the overall trend for visual guidance. d SS as functions of source-drain current extracted from the transfer curves in (a) and (b) measured at T = 250 K.

Figure 3d plots the SS as a function of source-drain current extracted from the transfer curves measured at 250 K. Two major disadvantages of tunneling FET can be clearly identified. Firstly, the SS rapidly degrades with increasing current. The total voltage required to fully switch on the device and thereby the power consumption is still relatively high. Secondly, the ON-state current for a tunneling FET is lower than that of a conventional FET. Fortunately, these disadvantages are no longer critical for sensitive photo-detection, because the device is mainly operated around the lowest SS region for weak-light detection purposes.

Ultra-weak infrared detection of the photo-tunneling transistor

We next move to characterize the turn-on threshold for the photo-tunneling transistor. We extracted the minimum detectable power for mid-wave infrared light (2.5–4.8 μm wavelength) from the transfer curves under dark and different MIR illumination powers. Figure 4a plots the photo-currents as functions of incident MIR light power for the photo-tunneling transistor and conventional FET modes (conventional phototransistor), extracted from Fig. 4b and Supplementary Fig. 6 (V_bg = −0.3 V and −2.5 V for photo-tunneling transistor and conventional FET, respectively. All nearby the lowest SS points where the minimum power is achieved). For comparison, typical detectable threshold powers for recent IR photodetectors^{15,23,29,30,31,32,33,34,35,36,37} are provided in Supplementary Table 2. The results show that the photo-tunneling transistor significantly improves the turn-on threshold power even with lower responsivity/detectivity.

Fig. 4: Photo-response of the photo-tunneling transistor.

a Measured photo-current as a function of incident light power for conventional FET (red, V_bg = −2.5 V) and photo-tunneling transistor (black, V_bg = −0.3 V). Inset: enlarged curve at the low power region. b Representative photo-current as a function of incident power and back-gate voltage. The corresponding SS is also shown on the top axis. The white dash-dot indicates the turn-on threshold. c, d Photo-current mapping for photo-tunneling transistor (c) and conventional FET (d) under 35 pW light illumination. T = 80 K, V_ds = 0.1 V. The photo-tunneling transistor sensitively responds to the light. For comparison, no photo-current is generated in a conventional FET. The dashed lines define different structures of the device. Specifically, yellow dashed lines indicate the top gates (V_tg1 and V_tg2), blue dashed lines indicate the source and drain regions, and the white dashed line indicates the black phosphorus (BP) channel.

To clearly see the difference, we mapped the photo-currents of the device in different modes by a scanning photocurrent microscopy. Figure 4c, d show the representative mapping results excited by 35 pW MIR irradiation. Even with such a weak incident light, a photovoltaic-like response driven by externally applied electric field is apparently observed in the photo-tunneling transistor³⁸. For comparison, no photo-current is generated in the conventional FET. Note that this improvement of threshold cannot be attributed to detectivity enhancement. While the noise figures of the two modes are similar (Supplementary Fig. 7a), the overall responsivity of the photo-tunneling transistor across the entire power range (except around the threshold region) is almost universally lower than that of the FET (Fig. 5a). As a result, although the conventional FET mode is of higher detectivity (Supplementary Fig. 7b), its minimum detectable light power is inferior. This unexpected phenomenon unambiguously highlights the important role played by carrier injection in a photodetector.

Fig. 5: Correlation between SS and turn-on threshold light power.

a Responsivity extracted from the photo-response measurements for conventional FET (red) and photo-tunneling transistor (black). The photo-tunneling transistor has better threshold power but lower responsivity. b Measured threshold power as a function of temperature for conventional FET (red) and photo-tunneling transistor (black). The curves can be categorized into two different regions, showing the critical role played by charge injection in determining the threshold power. c Transfer curves of the device with different top gate voltages. The SS is continuously tuned by the Vtg₁. Grey dashed lines show subthreshold slopes of SS = 60 mV/dec for comparison. V_ds = 0.1 V. d Measured threshold power as a function of SS corresponding to the configuration in (c). The dashed lines in (a) and (d) are described as indicating the overall trend for visual guidance.

We finally examine the direct correlation between the turn-on threshold and SS of a photo-transistor. There are two ways to adjust the SS along the carrier injection of photo-transistors. At first, temperature (T) simultaneously influences the thermionic injection rate and noise of photo-transistors. Figure 5b illustrates the temperature-dependent threshold power for photo-tunneling transistor and conventional FET. The curves can be categorized into two intervals. At sufficiently low T (~80–100 K) the threshold power is nearly temperature-independent. And at higher T, the threshold powers exponentially increase. We assign the two intervals as SS-restricted and noise-/noise-SS jointly restricted regions, respectively. In both cases, the photo-tunneling transistor exhibits much better performance (about 30 and 5 times better at 80 K and 300 K).

In addition, the carrier injection rate reflected by SS can be continuously tuned through top gate voltage. Specifically, the band-to-band tunneling current can be described by the Wentzel–Kramer–Brillouin approximation^39,40. It is suggested that the tunneling probability relates to the tunneling window, material bandgap, carrier effective mass and screening tunneling length. On one hand, the top gate modifies the tunneling energy barrier and thereby the tunneling window. On the other hand, it also controls the screening tunneling length by changing the spatial charge region across the homojunction. Therefore, it allows to continuously tune the SS through the partial top gate. Figure 5c illustrates the transfer curves under different positive Vtg₁ (A similar results were also observed for n-type device under different negative Vtg₂, see Supplementary Information note 2 for details). The SS gradually decreases with increasing gating. Figure 5d plots the corresponding turn-on threshold powers. The threshold power exponentially decreases with the SS (T = 80K, where the SS is dominant over noise), verifying carrier injection is the determining factor for the lowest detectable power of a photo-transistor. Interestingly, we found a single photon limit can be theoretically achieved with SS = 5.3 mV/dec by extrapolating the measured curve.