Solar-blind ultraviolet (SBUV) detectors can be widely applied in the scopes of short-range communication, corona discharges in transmission line and high-voltage equipment, flame detection, missile plume detection, biological imaging, etc., since they can operate on the Earth in broad daylight without any background signal from the Sun [1–5]. However, currently available SBUV detectors are mostly based on high-voltage and high-vacuum devices, which feature large volume, high energy consumption, fragility, etc. [6–10]. Therefore, there is an urgent need to renovate SBUV detection technologies. AlGaN is one of the most promising materials for SBUV detection due to its tunable ultrawide bandgap and efficient absorption in the SBUV band. The development of AlGaN materials gives rise to the fabrication of various UV detectors, including photoconductors, Schottky diodes, metal-semiconductor-metal (MSM) structures, p-n and p-i-n diodes, etc. [11–15]. However, the device performance will sharply decrease with the increase of Al content due to crystal quality deterioration and p-type doping difficulty.

Several approaches have been applied to conquer the bottlenecks. AlN/sapphire template and p-GaN were used to replace the sapphire substrate and the p-AlGaN to improve the crystal quality and p-doping efficiency [16]. Although a high peak external quantum efficiency (EQE) was achieved, the peak responsivity (R) was low due to the absence of internal gain and the existence of p-GaN damaged the SBUV detection characters. To avoid the p-doping difficulty, an asymmetric electrode MSM structure was proposed [17]. A self-powered SBUV detector was achieved, but the zero-bias EQE was low, too. Afterward, surface modifications were introduced to improve the MSM photodetector performance. Al nanoparticles were employed to modify the AlGaN MSM SBUV detector, and it was found that the R could be enhanced due to the surface plasmon resonance (SPR) effect [18,19]. However, the SPR effect can hardly be applied inside the AlGaN structure since it decayed exponentially in the direction perpendicular to the surface and only took effect within few dozens of nanometers. Therefore, the unique 2D electron gas (2DEG) at the interface of AlGaN heterostructure was noticed [20–23]. An SBUV peak R higher than 103A/W could even be achieved based on the 2DEG [20,21,23], but the high visible-blind ultraviolet (VBUV) R and unspecified response speed may limit their wide-range application. Hence, the performance of the AlGaN SBUV detector is still less than expected, and further improvement is required.

III-nitrides possess strong spontaneous and piezoelectric polarization [24,25]. The ionic bond component and the noncentrosymmetrical crystal structures result in spontaneous polarization [Fig. 1(a)]. The tensile or compressive strain can alter the bond angle and further result in piezoelectric polarization [Figs. 1(b) and 1(c)]. When the AlGaN materials form a heterostructure, the fixed charges will appear at the interface according to the Gauss law [Fig. 1(d)]. This is also the underlying mechanism that the high-concentration 2DEG can be formed at the interface. The polarization effect has significant impacts on the performance of III-nitrides-based devices. As for light-emitting diodes (LEDs), the polarization will separate the electrons and holes in the quantum wells and, thus, cause a sharp reduction of wave function overlap between the electron and hole (quantum-confined Stark effect, QCSE), which severely deteriorates the recombination of carriers [26]. Therefore, nonpolar III-nitrides are being developed to eliminate the polarization effect [27]. However, the polarization is not always harmful to devices. For example, the success of high electron mobility transistor (HEMT) strongly depends on the polarization-induced 2DEG [28]. Therefore, the role of the polarization effect in devices depends on the design. For AlGaN SBUV detectors, if the built-in field in the absorption layer can be enhanced by controlling the polarization [Figs. 1(e) and 1(f)], the photogenerated electron-hole pairs can be separated more efficiently and thus the carrier collection efficiency can be improved.

In this work, the polarization effect was utilized to improve the performance of AlGaN SBUV detectors. AlGaN heterostructures were designed to enhance the built-in field in the absorption layer. Two effects were generated: 1) the photogenerated electron-hole pairs were effectively separated, thus obtaining a longer carrier lifetime; 2) an ultrahigh electron mobility conduction channel was formed, thus achieving a faster electron transport speed. As a result, an SBUV detector with a high peak R of 1.42 A/W at 10 V was achieved, being 50 times higher than the nonpolarization-enhanced one. It is worth mentioning that the polarization effect did not severely affect the response speed. Furthermore, a high work function metal (Ni/Au) was adopted as an electron reservoir to modify the polarization effect. A higher peak R of 3.1 A/W at 30 V and a better signal-to-noise ratio (SNR) were finally realized. The investigation provided feasible approaches to improve the performance of AlGaN SBUV detectors by taking advantage of its inherent property.

Simulations were employed to predict the effects of polarization on the device performances. The polarization-enhanced detector in simulations is labeled as device SA [Fig. 1(g)]. For comparison, a nonpolarization-enhanced one was designed as device SB [Fig. 1(h)]. The Al content of the absorption layers was designed as 45%. The commercial APSYS software was used in simulations. The simulations were conducted by self-consistently solving the Poisson’s equation, continuity equation, and drift-diffusion equation with proper boundary conditions. Important factors include polarization level of 60%, defect-induced carrier lifetime of 3 ns [29], hole and electron mobility of 5 and 1200cm2/(V·s) [30,31], Schottky barrier heights of Ni/Al0.60Ga0.40N and Ni/Al0.45Ga0.55N contacts of 1.96 and 1.73 eV (Appendix A), and the energy band offset ratio for AlGaN heterojunction of 50:50 [32], respectively. The optical absorption coefficients for AlGaN layers were extracted from Refs. [33,34].

The epilayers of devices SA and SB were grown on an AlN/sapphire template by high-temperature metal–organic chemical vapor deposition (HT-MOCVD). The trimethyl aluminum (TMAl), trimethyl gallium (TMGa), and ammonia (NH3) were used as the Al, Ga, and N precursors, and the hydrogen (H2) was used as the carrier gas, respectively. An AlN epilayer was regrown on the AlN/sapphire template at 1300°C before AlGaN epilayers to improve the surface morphology. Then, AlGaN hetero-epilayers were grown on the as-grown AlN/sapphire template at 1180°C. Through tuning the Al/Ga source ratio, the designed Al content of each layer can be realized.

Standard photolithography processes and e-beam evaporation were adopted to fabricate the devices SA and SB. Interdigital grooves were first made on the photoresist mask by standard photolithography. 20/60 nm Ni/Au films were then deposited on the photolithograph-completed epitaxial wafers by e-beam evaporation. After lift-off process, the fabricated devices were rapidly annealed at 450°C for 3 min in N2 atmosphere to make Schottky contact. Moreover, a device with an electron reservoir structure was fabricated, labeled as device SC. First, 20/60 nm Ti/Al electrodes were deposited, followed by annealing at 600°C for 30 s in N2 atmosphere to form the source and drain. Then, semitransparent Ni/Au films (Tr∼60%, Fig. 7 in Appendix A) were deposited between the source and drain as an electron reservoir, followed by annealing at 550°C for 5 min in N2 atmosphere to form a higher Schottky barrier.

The in situ reflectance curves were used to monitor the epilayer growth rate. A high-resolution X-ray diffractometer (HRXRD, Bruck D8 discover) was adopted to estimate the Al content and crystal quality of the epilayers. An FL920 spectrophotometer was utilized to measure the absorption and transmission spectra of the as-grown epilayers. The IV curves were measured by a Keithley 6487. The spectral responses were measured by a DSR100 system with a xenon lamp, chopper, monochromator, Keithley 6487, SR830 lock-in amplifier, and standard Si detector. All response spectra were calibrated by the standard Si detector (Appendix A). The transient response spectra were measured by a RIGOL DS6104 digital oscilloscope. A pulse laser with emission wavelength of 266 nm, average power of 20 mW, repetition rate of 3 kHz, and pulse width of 10 ns was used to measure the photocurrent and transient responses.

The energy band, electric field profile, and carrier distribution of devices SA and SB in dark and on illumination have been simulated to analyze the performance. The energy band of device SA in the equilibrium state bends heavily due to the polarization-effect-induced interface charges [upper panel in Fig. 2(a)]. The energy band of the Al0.45Ga0.55N absorption layer bends down toward the Al0.45Ga0.55N/Al0.60Ga0.40N interface from the Al0.70Ga0.30N/Al0.45Ga0.55N interface, which indicates there exists a strong built-in field in the absorption layer from the Al0.45Ga0.55N/Al0.6Ga0.4N interface to the Al0.70Ga0.30N/Al0.45Ga0.55N interface. The built-in field in the absorption layer in the equilibrium state is calculated to be as high as 0.2–0.25MV/cm [Fig. 2(b)]. Such a high electric field can efficiently separate the electrons and holes to different areas. The holes and electrons separately accumulate at the Al0.70Ga0.30N/Al0.45Ga0.55N and Al0.45Ga0.55N/Al0.60Ga0.40N interfaces [Fig. 2(c)], confirming this conception. In this case, the electrons mainly contribute to the current, promising the fast response speed. When device SA is illuminated at 280 nm, the energy band bends less [lower panel in Fig. 2(a)]. It is because most of the photogenerated electron-hole pairs in the absorption layer are separated and swept toward the Al0.45Ga0.55N/Al0.6Ga0.4N interface and Al0.70Ga0.30N/Al0.45Ga0.55N interface, respectively, and the polarization-effect-induced interface charges are partially screened. Despite the screening effect, the built-in field still reaches as high as 0.1–0.2 MV/cm [Fig. 2(b)], guaranteeing the sustained photogenerated carriers’ separation ability when illuminated. As expected, the total carrier concentration becomes higher, and the electrical conducting layer extends deeper on illumination than in dark [Fig. 2(c)].

For the nonpolarization-enhanced device SB, the energy band does not bend as severely as device SA [upper panel in Fig. 2(d)]. The slight band bending beneath the surface originates from the Schottky contact barrier in the simulation. Actually, the surface states can also cause such band bending in practical devices [35–38]. According to the band bending, it can be deduced that the built-in field points to the surface and that the value is much lower than that of device SA. The simulated average electric field within 200 nm beneath the surface is merely 0.02 MV/cm in dark [Fig. 2(e)], which is an order of magnitude smaller than that of device SA, indicating the weaker carriers’ separation ability of device SB. This is verified by the smaller concentration difference between the hole and electron around the surface of device SB [Fig. 2(f)]. In this case, both electrons and holes can contribute to the current. On illumination, the energy band bending becomes less, too [lower panel in Fig. 2(d)]. The average built-in field within 200 nm beneath the surface reduces to 0.01 MV/cm [Fig. 2(e)]. At such a low field, the photogenerated electron-hole pairs cannot be effectively separated [Fig. 2(f)]. The carrier lifetime will be shorter because of the huge overlap between electron and hole profiles and both types of carriers will contribute to the current, resulting in a poor R and SNR. Hence, it can be predicted that the polarization effect can handle positive effects when improving AlGaN SBUV detector performance.

To confirm our above analyses, the current-voltage (IV) and response characters of devices SA and SB have been calculated. The dark current of device SA is several magnitudes lower than that of device SB, especially at low biases [Fig. 3(a)], which come from the higher Schottky barrier and the single type of carrier conduction of device SA. When illuminated at 280 nm, the currents sharply increase and reach a similar order of magnitude for both devices. However, the difference between the photocurrent and dark current of device SA is significantly higher than that of device SB [Fig. 3(a)]. It can result in higher R and better SNR, which is proved by the simulated response spectra [Fig. 3(b)]. At biases from 5 to 20 V, the R of device SB is always lower than that of device SA and so does the SBUV/VBUV ratio (Rpeak/R350nm). The performance variation trends of devices SA and SB demonstrate that the performance of the polarization-enhanced detector can surely exceed that of the nonpolarization-enhanced one.

Experiments have been performed to confirm the superiority of the polarization-enhanced detector over the nonpolarization-enhanced one. Epilayers of devices SA and SB are grown on an AlN/sapphire template [Fig. 4(a)]. According to the reflectance curves, the regrown AlN layer is about 500 nm for both epitaxial wafers, the Al0.70Ga0.30N, Al0.45Ga0.55N, and Al0.60Ga0.40N layers of device SA are about 1 μm, 200 nm, 20 nm, and the Al0.45Ga0.55N layer of device SB is about 1.2 μm, respectively, which satisfies the design requirements [Fig. 4(b)]. The HRXRD 2θ−ω scanning of the epitaxial wafer for device SA exhibits three peaks near 36°, 35.5°, and 35.3°, which correspond to AlN, Al0.70Ga0.30N, and Al0.45Ga0.55N [39–43], respectively [upper panel in Fig. 4(c)]. The diffraction signal of 20 nm Al0.60Ga0.40N layer is too weak to be distinguished from the two peaks of Al0.70Ga0.30N and Al0.45Ga0.55N layers. The epitaxial wafer for device SB exhibits two HRXRD 2θ−ω scanning peaks near 36° and 35.3°, corresponding to AlN and Al0.45Ga0.55N, respectively [lower panel in Fig. 4(c)]. The HRXRD results indicate the Al contents of the epitaxial wafers for devices SA and SB satisfy the design requirements. The full widths at half-maximum (FWHMs) of (0002) plane HRXRD rocking curve scanning (XRC) of the absorption layer of the epitaxial wafers for devices SA and SB are 379 and 418 arcsec, respectively, implying the similar crystal quality between devices SA and SB [Fig. 4(d)]. Absorption spectra of the epitaxial wafers for devices SA and SB show a 280 nm cutoff wavelength, ensuring the SBUV detection character [Fig. 4(e)]. The optical microscopy image illustrates the electrodes of the fabricated devices SA and SB. The length, width, and spacing of the interdigital electrodes of devices SA and SB are 190, 10, and 10 μm. The devices are front-illuminated when working [Fig. 4(f)]. The effective photosensitive area of devices SA and SB is the region between the two interdigital electrodes [Fig. 8(a) in Appendix A].

The IV curves of devices SA and SB in dark and on illumination have been measured. The difference between photocurrent and dark current of device SA is much larger than that of device SB [Fig. 5(a)], which coincides with our simulations, indicating the polarization-enhanced detector has better response, including R and SNR, than the nonpolarization-enhanced one. However, at biases between ±4V, the dark current of device SA is higher than that of device SB, which deviates from the simulations [Figs. 5(a) and 3(a)]. It is likely because the interface-defect-induced tunneling current and surface-defect-induced surface current cannot be considered in the simulations, leading to the underestimation of dark current of device SA [Fig. 5(c)]. Meanwhile, since device SB has a traditional back-to-back Schottky-type MSM structure, the two Schottky junctions are always oppositely biased. At low biases, the forward-biased junction cannot be completely turned on, and there still exists a built-in field to suppress the thermal emission current. When the bias increases, the forward-biased junction is fully turned on, and the thermal emission current will remarkably increase [Fig. 5(b)]. The phenomena that the dark current of device SB keeps low within ±3V and rapidly rises with the increasing voltage, and that the photocurrent exhibits the same trend as the dark current, verify it [Fig. 5(a)].

From the IV curves in dark, it can be deduced that the Schottky barrier of device SB is lower than that of device SA. On illumination, the current of device SA is lower than that of device SB within ±2V. The higher Schottky barrier of device SA should be responsible for it because the carriers cannot overcome the barrier at low voltage. The photocurrent of device SA exceeds that of device SB with the increasing bias because the applied voltage can overcome the higher Schottky barrier. The rapidly raising dark current made the photocurrent of device SB exceed that of device SA again, indicating that the SNR of device SA will become better than that of device SB with increasing bias [Fig. 5(a)].

The calibrated spectral response can more intuitively reflect the final device performance. Both devices produce a long-wavelength cutoff edge of 280 nm, demonstrating that the devices retain SBUV detection ability [Fig. 5(d)]. Device SA exhibits a peak response at around 257 nm. At biases of 5 V and 10 V, the R can reach 1.18 A/W and 1.42 A/W with the SBUV/VBUV ratios of about 260 and 220. At even higher bias of 20 V, the R can reach as high as 1.78 A/W, which is a competitive value even in real applications [Fig. 5(d)]. For comparison, device SB shows a peak response at around 255 nm. With the bias increasing from 5 V to 10 V, the R rises from 0.022 A/W to 0.0285 A/W. The SBUV/VBUV ratios at 5 V and 10 V are about 90 and 60, respectively [inset of Fig. 5(d)]. These results obviously reflect that the polarization effect has enhanced the R by about 50 times at the same bias and the SNR can also be improved. As shown, the results such as peak R, peak wavelength, and SBUV/VBUV ratios of devices SA and SB are different from the simulated ones [Fig. 3(b)]. It is mainly because the optical absorption coefficients and light illumination power density of the practical devices are different from those used in the simulations.

For the traditional Schottky-type MSM detector, the applied electric field separates the photogenerated electron-hole pairs and drives the electrons and holes to move in opposite directions. In this case, the electrons and holes transport in the same areas, and the carrier lifetime will be shorter because they are easy to recombine [Fig. 5(e)]. Although both types of carriers can contribute to the photocurrent, the current gain (G), which is determined by the carrier lifetime (τ) and transit time (t), can be very low (G∝τ/t), resulting in a poor response [44,45]. Fortunately, the polarization effect brings about effective solutions. First, the polarization-induced 2DEG with ultrahigh electron mobility of over 1000cm2/(V·s) is the only conduction channel [Fig. 5(f), and Fig. 9 in Appendix A]. It can tremendously reduce the carrier transit time. Second, the polarization-enhanced built-in field in the absorption layer is vertical to the carrier transport channel. It separates the photogenerated electron-hole pairs and sweeps the electrons into the conduction channel and the holes to the opposite direction [Fig. 5(f)]. The inherent existence of the built-in field makes electrons of less possibility recombine with the hole and, hence, obtain a longer lifetime, leading to better response. Moreover, the strong polarization effect prevents the hole from contributing to the current, which is helpful in reducing noise. It is also a reason why device SA can exhibit higher SNR than device SB.

Because of the opposite carrier lifetime dependence of gain and carrier recombination time, there is usually an opposite trend between responsivity and response speed. Longer carrier lifetime usually leads to higher responsivity and slower response speed due to the higher gain and longer carrier recombination process. To value the effect of the polarization on response time, the transient response spectra at different load resistances have been measured. The detector is connected to an adjustable load resistance and inserts into a circuit with sustainable and stable terminal voltage of 8 V. The pulse laser directly illuminates on the detector, and the oscilloscope simultaneously measures the voltage variation of the load resistance [Fig. 5(h)]. The response times are extracted by exponentially fitting the transient response spectra [inset of Fig. 5(g)]. Since the polarization-effect-enhanced built-in field prolongs the carrier lifetime, it can be speculated that device SA will have slower response speed than device SB. The response time at load resistance from 100Ω to 100kΩ has been measured for devices SA and SB [Fig. 5(g)]. It can be seen that the response time of device SB is always smaller than that of device SA, confirming the polarization effect. We have extrapolated the response time at zero load resistance according to the response time variation trends. The extrapolated response times of devices SA and SB are 7.6 and 3.5 μs, respectively, which can be considered as their inherent response times. Although the response speed of device SA has been slowed down compared with device SB, it still maintains a relatively high response speed due to the short transient time.

It has been verified that the enhanced polarization effect can improve the detector performance above. However, it is also noted that the dark current level is not low enough because of the high concentration 2DEG. Besides, the polarization-induced 2DEG will conversely screen the polarization-induced built-in field, lowering the separation efficiency of photogenerated electron-hole pairs. Therefore, if the 2DEG concentration can be effectively controlled, the device performance such as R and SNR ought to be further improved. As is known, metals can usually act as electron reservoirs to provide or store electrons. By contacting a high work function metal to a semiconductor, the electrons can transport to the metal through tunneling or thermal emission if the metal work function is higher than that of the semiconductor [Fig. 6(a)]. Therefore, we employ a structure in which a metal electron reservoir is deposited between the source and drain electrodes (SC). The optical microscopy image illustrates the electrodes and metal electron reservoir of device SC. The spacing between the source and drain electrodes is 20 μm, and the source and drain electrode widths are both 100 μm. The electron reservoir locates at the center between the source and drain and the width is 10 μm. Device SC is also front-illuminated when working [Fig. 6(b)]. The effective photosensitive area is estimated to be 80% of the area between the source and drain electrodes due to the semitransparent electron reservoir structure [Fig. 8(b) in Appendix A].

The IV curves of device SC have been measured in dark and on illumination to investigate the device performance [Fig. 6(c)]. Apparently, the difference between the photocurrent and dark current of device SC is further increased compared with devices SA and SB. It is found that the dark current is reduced by over three orders of magnitude at the same biases. Considering the effective area difference among the devices, it can be deduced that the dark current density of device SC has been decreased by at least one order of magnitude. Besides, the photocurrent reaches about 10 μA, which is the same magnitude as that of devices SA and SB. Taking the effective device area into consideration, we can also speculate that the collection efficiency of the photogenerated carriers has been improved. Spectral responses at different biases have been measured to examine the performance [Fig. 6(d)]. The cutoff wavelength still locates at around 280 nm. At biases of 5, 10, and 20 V, the peak R are 1.75, 2.31, and 2.51 A/W, respectively, which are remarkably higher than the values of device SA at the same biases. By further elevating the bias to 30 V, the peak R can reach as high as 3.1 A/W, thus demonstrating the excellent performance of device SC. Besides, the SBUV/VBUV ratios of device SC at different biases are calculated to be over 5×102, which is much higher than that of devices SA and SB, also indicating the better performance of device SC.

In summary, AlGaN SBUV detectors were fabricated and investigated. The inherent polarization effect was utilized to improve the device performance. AlGaN heterostructures were designed to modify the polarization effect and to enhance the built-in field in the absorption layer. The polarization-enhanced detector realized a 50 times improvement on the peak R compared with the nonpolarization-enhanced one. A peak R of 1.16 A/W at 5 V and a better SNR were achieved. The longer carrier lifetime at the absorption layer and the ultrahigh electron mobility conduction channel induced by the polarization effect were responsible for the improvements. Elimination of the hole current also partially contributed to the improvements. Besides, the polarization effect did not severely postpone the response speed due to the fast carrier drift speed in the conduction channel. Furthermore, an electron reservoir structure was proposed to improve the response character. A higher peak R of 3.1 A/W at 30 V and a better SNR were achieved since the electron reservoir structure could modulate the polarization-induced 2DEG concentration in the conduction channel to suppress the dark current and enhance the separation efficiency of photogenerated electron-hole pairs. We believe that the investigations here provide alternative and feasible approaches for the community to improve the AlGaN SBUV detector performance.