The device structure of our MAPbCl₃/GaN-based light-modulated BJT is schematically shown in Figure 1a with a bottom-illumination scheme. The precursor solution of perovskite is directly spin-deposited on a standard epitaxial-grown GaN film on sapphire substrate, with a pair of parallel T-shaped gold electrodes (W/L = 3000/50 μm = 60 μm) thermally deposited on top for photocurrent collection. For photosensing mechanism analysis, the control single-layer photoconductive devices based on an MAPbCl₃ film on quartz substrate and GaN film on sapphire substrate are, respectively, fabricated in a similar way to that of gold electrodes, as shown by the structures in Figure S1.

We fabricate our MAPbCl₃/GaN PD and the control single-layer at the optimal concentration (2M) of the precursor solution of MAPbCl₃. The concentration of the precursor solution, as well as the thickness of the MAPbCl₃ film deposited on the GaN surface, is initially optimized to minimize the dark current level and maximize the light current level. Figure S3a,b, respectively, compares the dark current and photocurrent levels of MAPbCl₃/GaN and control PDs with different concentrations of MAPbCl₃ precursor. The thickness of the MAPbCl₃ layer increases with the precursor concentration, to suppress the dark current of the MAPbCl₃/GaN PD. The photocurrent of MAPbCl₃/GaN PD reaches the maximum for precursor concentrations above 1 M. To maximize the light over dark on/off ratio, the optimal concentration to deposit the MAPbCl₃ film is chosen as 2M. The optimal thickness of the MAPbCl₃ film is measured to be around 516 nm by cross-sectional SEM, as shown in Figure S4. Figure 1d shows the comparison of the typical current–voltage (I–V) characteristics of the MAPbCl₃/GaN PD and the corresponding control devices under dark and 365 nm light illumination. Unlike the MAPbCl₃ PD with a low dark/light current level and the GaN PD with a high dark/light current level, the dark current of MAPbCl₃/GaN PD is about 4 orders of magnitude suppressed compared with the one of GaN PD, while the light current is increased by more than 300 times compared to the MAPbCl₃ PD. The significantly enhanced light over dark on/off ratio of MAPbCl₃/GaN PD cannot be only attributed to the optimized MAPbCl₃ film quality on the GaN surface, as the charge transfer properties of the MAPbCl₃/GaN PD are also very different from those of the control devices. Figure 1e shows the linear-scale I–V curves of our MAPbCl₃/GaN PD under different illumination levels. Under both negative and positive bias, the photocurrents of the heterogeneous device sharply increase below the 0.5 V range and then saturate above the 0.5 V range. Moreover, the saturated photocurrent of MAPbCl₃/GaN PD increases linearly with the light intensity, which is highly similar to the characteristic output curves of a BJT with common emitter configuration. (32) As shown in Figure 1f, the bias is applied on the lateral electrodes of the heterogeneous device (as emitter and collector, respectively) for photocurrent collection, with the optical pump on the GaN film as the base terminal to inject the photogenerated carriers. The linear-scale I–V curves of our MAPbCl₃/GaN PD and the control devices under dark conditions are also compared in Figure S5. It is clearly shown that the MAPbCl₃ film forms an Ohmic contact and the GaN film forms a Schottky contact with the gold electrodes. The saturated trend can still be observed in the dark current of MAPbCl₃/GaN PD, which confirms that the electric field is applied on the heterostacked layers.

With the comparisons of the dark and light I–V characteristics between the heterogeneous and single-layer devices, we can summarize that the MAPbCl₃/GaN heterointerface plays an important role in the photocurrent generation, as schematically illustrated by the photocarrier dynamics model in Figure 1f. At 365 nm, the GaN film contributes the most photogenerated carriers, with a large absorption coefficient and bottom-illumination scheme. The electron–hole pairs generated in the GaN film diffuse to the MAPbCl₃/GaN interface, and are then separated by the heterojunction and extracted by MAPbCl₃ and the electrodes. That is, the photocurrent flows through the perovskite and GaN layers dominate the photoresponse of our MAPbCl₃/GaN PD. We further measured the I–V curves of one MAPbCl₃/GaN device with a discontinuous (split in the middle of the electrodes) perovskite film under 365 nm illumination, as shown in Figure S6. Even the perovskite film covering the GaN film is discontinuous; a similar output characteristic curve can still be obtained as the continuous one.

As the photoresponse is dominated by the heterostacked layers, we can analyze the photocarrier dynamics of our device with a MAPbCl₃/GaN/MAPbCl₃ p-n-p junction model, as shown by the equilibrium energy diagram in Figure 2a. The energy diagrams of MAPbCl₃ and GaN films in isolated status are shown in Figure S7d, which are determined by ultraviolet photoelectron spectroscopy (UPS) as shown in Figure S7a,b, with the optical band gaps derived from the Tauc plots in Figure S7c. The evaluated conduction band, valence band, and Fermi level of MAPbCl₃ and GaN are consistent with the previous studies. (19,33,34) It is observed that MAPbCl₃/GaN forms a type II heterojunction with a well-aligned conduction band, which can be further verified by the steady-state photoluminescence (PL) spectra analysis. Figure 2b compares the PL spectra of GaN and MAPbCl₃/GaN films optically pumped on the GaN layer. The PL emission of the GaN film is significantly quenched with a MAPbCl₃ coating layer, which indicates that the photogenerated holes can be effectively transferred from GaN to MAPbCl₃ with the energy shift in the valence band. Figure 2c compares the PL spectra of MAPbCl₃ and MAPbCl₃/GaN films with those optically pumped on the MAPbCl₃ layer. Due to the well-aligned conduction band between MAPbCl₃ and GaN, the photogenerated electrons and holes are mostly kept in the MAPbCl₃ layer, resulting in nearly unchanged PL spectra.

Then, we consider this light-modulated BJT under bias and illumination to summarize the operation model for photosensing, as schematically shown by the energy diagrams in Figure 2d, with the inherited characteristics of the photoconductive gain of photoconductors and the internal field of photodiodes. As the bias is applied on the electrodes on the MAPbCl₃ film, the superposition of the external electric field over the p-n-p junction results in the barrier of one junction increasing and the other one reducing. On the one hand, the enhanced junction facilitates the photogenerated charge separation at the heterointerface, leaving the holes in the MAPbCl₃ layer (to be directly collected) and the electrons in the GaN layer. On the other hand, the reduced junction helps the electrons accumulated in the GaN layer to diffuse into the MAPbCl₃ layer with a flat band alignment to be effectively collected. Therefore, no matter whether the photocarriers are generated in GaN or MAPbCl₃ films, the photoresponse of our light-modulated BJT is dominated by the enhanced junction with efficient charge separation and transfer, showing a saturated photocurrent like a photodiode operated at reverse bias. In addition, with the large distance between the lateral electrodes, the large trap states in the GaN layer can temporally capture the photogenerated electrons during charge transport, (35,36) which results in the multiplication of photocurrent with photoconductive gain.

To directly illustrate the necessity of band alignment in our BJT for photosensing, we replace the MAPbCl₃ layer with MAPbI₃ in our device and compare the I–V characteristics of the MAPbCl₃/GaN and MAPbI₃/GaN PDs under the same white light illumination, as shown in Figure S8a. The energy diagrams of MAPbI₃ and GaN are shown in Figure S8b based on the published literature. (37) Without the well-aligned conduction band, both the light and dark I–V curves of MAPbI₃/GaN PD are photoconductive-type without the saturated current. In particular, the photocurrent of MAPbI₃/GaN PD is significantly lower than the one of MAPbCl₃/GaN PD, although it is well known that MAPbI₃ has a good light absorption up to 780 nm. (38) Besides, a similar device of MAPbBr₃/GaN PD with type II band structure is reported, (39) with no saturated current observed. Therefore, the band alignment is the essential factor affecting the charge dynamics.

The photodetection properties of our MAPbCl₃/GaN-based light-modulated BJT are carefully characterized, with comparison to the control devices. Figure 3a shows the measured spectral responsivities of our MAPbCl₃/GaN PD and the single-layer control devices. The MAPbCl₃/GaN PD exhibits a photosensing range between 360 and 400 nm, with two peaks at 370 nm (0.43 A/W) and 390 nm (0.45 A/W), corresponding to the excitonic absorption peaks of GaN and MAPbCl₃, respectively. The MAPbCl₃ PD shows a low responsivity over the whole absorption range, indicating the large recombination loss in the single-layer MAPbCl₃. The responsivity of GaN PD is high, which matches well with the former reported similar photoconductors with rich trap states, contributing a high photoconductive gain. (35,40,41) Figure 3c shows the measured noise current spectral densities of MAPbCl₃/GaN PD and the control single-layer devices. The noise spectrum of GaN PD shows a typical flicker noise (1/f) with dominative trend, with the noise level 5 orders of magnitude overall higher than the ones measured in the other two devices. This could be attributed to the high carrier density and large trap state distribution in the GaN film, as 1/f noise is related to carrier density fluctuation. (42−44) The measured noise spectrum of MAPbCl₃/GaN PD is similar to the one of MAPbCl₃, showing a low noise level approaching the shot-noise limit (evaluated with the dark current of MAPbCl₃/GaN PD) and the measured instrument limit. The specific detectivity is the normalized figure-of-merit used to compare the sensitivities in different photodetectors. With the measured spectral responsivity, current noise density (at 7 Hz modulation frequency), and active area, the spectral specific detectivities of different devices are calculated and compared in Figure 3b. At 370 nm, the specific detectivity of our MAPbCl₃/GaN PD is evaluated to be 4.11 × 10¹² Jones (cm Hz^1/2 W^–1), which is more than 200 times higher than the one of the control MAPbCl₃ PD (1.89 × 10¹⁰ Jones), and more than 5 × 10⁴ times higher than the one of GaN PD (8.20 × 10⁷ Jones). At 390 nm, the MAPbCl₃/GaN PD reaches a peak detectivity of 4.30 × 10¹² Jones.

The response speed is an essential parameter in evaluating the operation bandwidth of a PD. Figure 3d shows the transient photocurrent response of different devices with square-wave-modulated light illumination. The response speed of PDs could be evaluated with the rise/fall time, with rise time being the time difference from 10 to 90% of the peak amplitude on the leading edge of the pulse, and fall time being the time difference from 90 to 10% of the trailing edge of the pulse. As for the former reported photoconductive devices, (40,41) the response speed of GaN PD is slow with the rise/fall time evaluated to be 0.17/25.46 s (full transient curve is shown in Figure S10). For MAPbCl₃ PD, a slow trapping and detrapping process can be observed in the rising and falling curve, (45,46) with the rise/fall time evaluated to be 0.19/2.07 ms. In contrast, the MAPbCl₃/GaN PD has a sharp photocurrent response with the rise/fall time evaluated to be 70.50/71.83 μs; the detailed rising and falling curves are shown in Figure 3e. This could be related to the higher crystallinity of MAPbCl₃ film on the GaN surface and the enhanced internal field that facilitates the charge separation and extraction.

We notice that both the MAPbCl₃/GaN and GaN PDs discussed above exhibit a narrow-band spectral response range, with suppressed photoresponse below the 350 nm wavelength range. This could be attributed to the charge collection narrowing (CCN) effect (49) of the 4.6 μm-thickness GaN film adopted in the devices with a bottom-illumination scheme. Therefore, we fabricated the MAPbCl₃/GaN PD with a 80 nm-thickness GaN film on sapphire substrate and calibrated the photodetection properties. Figure 4a shows the comparison of the I–V characteristics of the light-modulated MAPbCl₃/GaN (80 nm) PD, and the control MAPbCl₃ and GaN (80 nm) PDs under dark and illumination. The dark current of the MAPbCl₃/GaN (80 nm) PD is 4 orders of magnitude lower than the one of GaN PD, while the photocurrent is 3 orders of magnitude higher than the one of MAPbCl₃ PD. This indicates that the heterogeneous device with a thinner GaN film can still effectively reduce the dark current level and improve the photocurrent level, thus significantly enhancing the on/off ratio. However, the dark current of MAPbCl₃/GaN (80 nm) PD is about 10 times higher than that of the MAPbCl₃/GaN (4.6 μm) device at 0.5 V bias. This could be attributed to the poor quality of the thin GaN film grown on sapphire with unreleased lattice stress, resulting in a higher density of defect states. (50) Figure S12a shows that the dark current of GaN (80 nm) PD is about 50 times higher than the one of GaN (4.6 μm) PD at 0.5 V bias. This corresponds to a higher crystallinity of the GaN (4.6 μm) film, which is confirmed by the XRD pattern comparison to the GaN (80 nm) film in Figure S12b,c. The saturated photocurrent output can also be observed in the light intensity-dependent linear-scale I–V curves of the MAPbCl₃/GaN (80 nm) PD in Figure 4b, which indicates that the photosensing model of a light-modulated BJT is unchanged with the thickness of the GaN film adopted.

With the inherited properties of the photodiode and photoconductor, our MAPbCl₃/GaN-based light-modulated BJT provides a structure panel to balance the trade-off in different photosensing properties. As shown in Figure 5a, we compare the responsivity versus 3 dB bandwidth of our devices and the previously reported similar UV PDs, with the indication of a gain-bandwidth-product (GBP). The 3 dB bandwidth is evaluated with the formula f_3?dB = 0.35/t_f, where t_f is the fall time. (51) Our devices show the best GBP value, with a responsivity comparable to the reported photoconductors and a faster response speed outperforming most reported photodiodes. Not only in GBP, our device shows excellent comprehensive properties outperforming the published similar visible-blind UV PDs and the counterpart commercial products (comparisons of different figures of merit are summarized in Table S1). Figure 5b shows the comparison of the comprehensive photodetection properties of our device to those of the commercial GaN visible-blind UV PD and silicon UV-enhanced broadband PD. Our device is superior to the commercial GaN UV PD in all of the parameters, which approaches the 5H requirement.

The crystallinity of perovskite and GaN films was analyzed by an X-ray diffractometer (XRD-7000, Shimadzu). The surface morphology of the perovskite films was observed by a scanning electron microscope (SUPRA55 SAPPHIRE, Zeiss). The absorption spectra were characterized by a UV–vis spectrophotometer (UV-2600i, Shimadzu). The PL spectra of perovskite and GaN films were obtained by a photoluminescence spectrometer (FLS 1000, Edinburgh) with an excitation wavelength of 325 nm. The thickness of perovskite films was measured by a stylus profilometer (Dektak XT-A, Bruker). The energy-level measurements of perovskite and GaN films were carried out using ultraviolet photoelectron spectroscopy (ESCALAB Xi+, Thermo Fisher Scientific). The COMSOL software was used to simulate the electric field distribution and I–V curves of the devices.

Optoelectronic measurements for the photodetectors: The photoresponse performance was scanned by a source meter (2636B, Keithley) under dark and illumination. The light intensity was calibrated by a standard photodiode sensor (818-UV, Newport). The response speed of the photodetectors was measured by a transient photocurrent test. The light-emitting diode with a wavelength of 365 nm was modulated by a square wave at the frequency of 37 Hz from an arbitrary function generator (AFG-2225, GW instek) and was used to illuminate the devices. The current was recorded by an oscilloscope with a sampling resistor of 50 Ω (InfiniiVision DSOX4024A, Keysight). The EQE spectra were measured by a homemade external quantum measurement system (Omni-λ 300i, Zolix). The noise current spectral density of photodetectors was recorded by SR770 (FFT Analyzer, Stanford Research Systems) and the devices were biased by a battery case and sealed in a metal shielding box. The distance between the LED and the device used in the transient response and the LDR characterizations was 20 cm. The power density parameter is the light density illuminated on the device surface.

Indoor optical communication system demonstration: Two open-source electronics platforms (STM32F103ZET6) were, respectively, adopted as the driver at the transmitter and at the receiver end. Data from one computer were transferred into high/low-level voltage using the driver to drive the on/off state of the UV LED. The modulated visible-blind light from UV LED can be received by our device to produce the photocurrent, which is converted to high/low-level voltage by SR570 (low-noise current preamplifier, Stanford Research Systems) as the input signal at the receiver end. Then, the driver transfers the voltage level into data to be decoded by another computer. The signal coding method is pulse width modulation (PWM).

The formulas to calculate the figures of merit for photosensing are as follows:

The responsivity (R_λ) is defined as R_λ = I_ph/P_in, where I_ph is the photocurrent of the device when the incident light power on the device sensing area is P_in.

The EQE_λ is expressed as where h is the Planck constant, c is the light speed, and q is the elementary charge.

The noise equivalent power (NEP) is calculated by NEP = I_n/R_λ, where I_n is the current noise spectral density measured with the described method.

Therefore, the specific detectivity D* is calculated by where A is the sensing area of the device.