High gain ultraviolet (UV) light detectors are significant for various applications, such as medical imaging, optical communication, flame detection, chemical sensing, sterilization and purification, ozone sensing and detection, and various other emerging imaging applications for security. Currently, the dominant UV light detectors are based on photomultiplier tubes, which suffer from the high cost and low lifetime (Zheng et al., 2020). Solid-state-based avalanche photodiodes (APDs) look attractive for UV light detections since they can change the landscape of UV detection by offering more durability, lower cost, and surpassing the performance of a tube-based multiplier. The APD is a highly sensitive photodetector that converts a light signal to an electrical signal and amplifies it by taking advantage of the carriers’ multiplication during impact ionization and avalanche (Sze and Ng, 2006).

Impact ionization occurs in a semiconductor when one energetic carrier, either an electron or a hole, loses its energy after scattering with the crystal lattice and generates more energetic electrons and holes. The generated electrons and holes continue to travel through the high electric field region where they continue to accelerate and eventually lose their energy to scattering with the crystal lattice, generating more and more carriers. This phenomenon leads to the multiplication of carriers leading to an avalanche breakdown in the device. However, when properly controlled the impact ionization process can be of great use for enabling various high frequency, high power amplifiers, and detectors. Figure 1 schematically illustrates the impact ionization process in a reverse-biased p-i-n junction to highlight the multiplication process.

The condition of avalanche often needs to be defined quantitatively for practical use of the phenomenon in devices. For instance, the multiplication factor, M, can be mathematically written as the following equation: M = 1 1 − ( V B V ) n , (1) where V is the applied voltage and BV is the avalanche breakdown voltage.

In APDs, when the applied voltage is less than the BV, the device works in the linear mode. When the applied voltage equals to the BV, the multiplication factor tends to infinity, and the Geiger mode is triggered in an APD, which can be used appropriately for single-photon detections (Hu, 2008; Hadfield, 2009; Zhang, 2011).

Owing to its bandgap of 3.4 eV and large electric field tolerance, GaN has exceptional advantages as APDs, serving various applications, particularly those that require high temperature and harsh environment tolerance.

To enable a GaN APD with robust avalanche capability, we have noted from our studies (Ji D. et al., 2019; Ji et al., 2020a; Ji et al., 2020b; Ji and Chowdhury, 2020) the following two requirements: 1) presence of high-quality epitaxial layers, which is possible when grown on free-standing GaN substrates; 2) optimized edge terminations to eliminate both localized peak electric field and dark leakage in the device.

Due to the absence of natural GaN substrates, GaN film growths for devices were primarily developed on foreign substrates, such as sapphire, silicon, or silicon carbide (Amano, 2015; Liu and Edgar, 2002; Krost and Dadgar, 2002; Mion et al., 2006). Constrained by the lattice mismatch, as shown in Figure 2, the GaN films are amenable to a large number of defect densities in the order of 10⁹ or 10⁸ cm⁻² (Amano, 2015; Liu and Edgar, 2002; Krost and Dadgar, 2002; Mion et al., 2006). Such defects can severely inhibit the impact ionization process. When a reverse bias is applied, the carrier can lose energy due to the scattering with the defects, inhibiting the impact ionization. Table 1 summarizes the defect densities of different growth technologies. From various reports on the observation of avalanche, it can be concluded that the homogeneous growth technique based on free-standing GaN substrates enabling a lower defect density of 10³–10⁶ cm⁻² (Fawamura et al., 2006; Paskova and Evans, 2009) is essential for avalanche robustness.

Over the last decade, researchers have spent extensive efforts to develop the free-standing GaN substrates using methods such as hydride vapor-phase epitaxy (HVPE), Ammonothermal, and sodium flux enabling the homogeneous growth of GaN (Fawamura et al., 2006; Paskova and Evans, 2009). The issue of lattice mismatch could be solved by the homogeneous growth on free-standing single-crystalline GaN substrates. The overall quality of the epitaxial GaN was improved remarkably over the last decade, lowering the defect density in the material, making it suitable for impact ionization of carriers. On bulk GaN substrates, lo and behold, the impact ionization and avalanche breakdown were finally reported in p-i-n diodes in 2013 by Avogy Inc. (Disney et al., 2013). Since then, multiple research groups have reported the avalanche breakdown in GaN p-n diodes and extended the capability to Junction Field-Effect Transistors (JFETs) (Kizilyalli et al., 2013; Nomoto et al., 2016; Cao et al., 2018; Maeda et al., 2018; Ji D. et al., 2019; Maeda et al., 2019a; Fukushima et al., 2019; Ohta et al., 2019; Ji et al., 2020a; Ji et al., 2020b; Ji and Chowdhury, 2020; Liu et al., 2020; Liu et al., 2021). Our GaN APDs (Ji et al., 2020b), were also grown on the bulk GaN, which fulfilled the first requirement of high-quality- low defect density (LDD) epitaxial films.

The second requirement for a guaranteed robust avalanche capability is efficient edge termination (Baliga, 2012). An edge termination that can provide extra negative charges to terminate the electric field lines at the device boundary to reduce any localized peak electric field at any surface or corners is important for the realization of an APD. Figure 3 shows several edge terminations that were proposed for avalanche GaN p-n diodes, including the ion implantation-based edge termination (Kizilyalli et al., 2013; Ji et al., 2020a), etch-based edge termination (Fukushima et al., 2019; Maeda et al., 2018), and field-plated edge termination in which the field plate is connected to the anode electrode (Nomoto et al., 2016; Ohta et al., 2019). The schematic of the electric field distribution in the GaN p-n diodes with and without the ion-implanted edge termination is shown in Figure 4. The most common technique to form the mesa and isolation is the dry etching, which could lead to sharp corners and the dense electric field lines around the device, as shown in Figure 4A. Figure 4B shows the schematic of a GaN p-n diode with the ion-implanted moat etch termination (Ji et al., 2020a), where the corners are rounded during isolation and can effectively reduce the localized peak electric field. Moat termination combined with implantation can compensate for the dry etching-induced damages, eliminating the leakage current through the etched surface.

The most notable feature with avalanche breakdown in GaN APDs is the positive temperature coefficient of the breakdown voltage. Impact ionization is a process that energy transfers from the energetic carriers to the crystal lattice by scattering. When the temperature is elevated, the crystal lattice vibration becomes stronger, which scatters the carriers more frequently and reduces the impact ionization mean free path. Therefore, it is more difficult for carriers to gain the threshold energy to trigger the impact ionization. Experimental data also indicate that the impact ionization coefficients of both electrons and holes are decreasing when the measurement temperature increases (Ji D. et al., 2019; Maeda et al., 2019b; Cao et al., 2021). Figure 5A shows a temperature-dependent avalanche breakdown characteristics of a GaN APD reported by our group (Ji et al., 2020b). It indicates that the breakdown voltage of the device increases with temperature linearly, as shown in Figure 5B. The temperature-dependent breakdown voltage, BV(T), can be fitted to the following formula: B V ( T ) = B V R T ( 1 + θ ( T − 300 ) ) , (2) where BV _RT is the breakdown voltage measured in the room temperature, and θ is the temperature coefficient. According to the data presented in Figure 5B, the temperature coefficient, θ, of GaN can be extracted to be 3.85 × 10^–4 (Ji et al., 2020b). Table 2 summarizes the temperature coefficients of various semiconductors (Hall, 1967; Vassilevski et al., 2000; Groves et al., 2003; Tan et al., 2010), it indicates that all the semiconductors have positive temperature coefficients, whose values are in the order of 10⁻⁴ K⁻¹. We should note here that the positive temperature coefficient is an extremely important verification of avalanche. This verification is critical in determining if the gain in the APD is due to avalanche or any other mechanism. III-V (e.g GaAs) devices avalanche at a much lower electric field (0.5 MV/cm) and are more routinely observed. The measurement of the positive temperature coefficient is not often reported in III-As APDS (Hu et al., 1996; Anselm et al., 1997), perhaps due to its certainty. However, in GaN, often gain can be witnessed without avalanche, such as in heterojunction-based photodetectors, necessitating the importance of temperature measurements to confirm the avalanche.

Since GaN is a direct bandgap semiconductor, there are photons generated during the recombination of electrons and holes. When the GaN p-n diode is forward biased, there are electrons and holes injections into the p- and n-regions respectively, therefore an emission in the range of 370–420 nm light can be observed during the recombination of these carriers (Nakamura, 2015). When a reverse bias is applied to a GaN p-n diode and forces the device to an avalanche breakdown, the impact ionization process generates excess electrons and holes in the depletion region and should release photons during the radiative recombination of those carriers. Therefore, a 370–420 nm (violet/blue-violet) electroluminescence is expected to be observed in a GaN APD. This has been observed in GaAs where the 950 nm light is emitted (Lahbabi et al., 2004). Figure 6A presents the optical microscope image of a GaN APD layout. The image of the APD with avalanche induced electroluminescence is shown in Figure 6B (Ji et al., 2020b).

In GaN APDs, the impact ionization multiplication can be written as a function of impact ionization coefficients: M n = 1 1 − ∫ 0 W α   e − ∫ x w ( α − β ) d x ′ d x , (3) And M p = 1 1 − ∫ 0 W β   e − ∫ 0 x ( β − α ) d x ′ d x , (4) Where M _n is the electron-initiated multiplication factor and M _p is the hole-initiated multiplication factor, W is the width of the depletion region. The electron-initiated impact ionization multiplication is illustrated in Figure 7A, where UV light illuminates the p-GaN region and the photogenerated electrons are drift into the depletion region and initiate the impact ionization process. Figure 7B shows the hole-initiated impact ionization multiplication, where UV light illuminates the n-GaN layer and photogenerated holes initiate the impact ionization process.

For GaN APDs, the performance, such as gain and excess noise, depends on the impact ionization coefficients of both electrons and holes (Sze and Ng, 2006). Therefore, it is important to measure the impact ionization coefficients in GaN.

The reported estimated ionization coefficients in GaN are listed in Table 3 and are plotted in Figure 8.

The studies reflected in Figure 8, helps us realize that within the reasonably high electric field range (1 MV/cm—3 MV/cm), holes have a larger impact ionization coefficient compared to electrons in GaN, the hole-initiated multiplication APDs should have better performance compared to electron-initiated multiplication APDs, and this has been proved both theoretically and experimentally (McClintock et al., 2007; Minder et al., 2007; Zheng et al., 2016).

The early GaN APD studies belonged to electron-initiated devices where the structures were based on GaN-on-Sapphire (Verghese et al., 1998; Carrano et al., 2000; Cicek et al., 2010; Sun et al., 2010), as shown in Figure 9A. Large defect density, as well as the etching-induced damages, contributed significantly to the high dark current density, which is in the range of 10^–4 to 1 A/cm² (Verghese et al., 1998; Carrano et al., 2000; Cicek et al., 2010; Sun et al., 2010). With the introduction of GaN-on-GaN APDs, as shown in Figure 9B, dark current density of less than 10^–4 A/cm² could be realized, along with a gain of 10⁴ (Limb et al., 2006; Shen et al., 2007; Yoo et al., 2007; Zhang et al., 2008; Zhang et al., 2009).

Because the hole’s impact ionization coefficient is much larger than the electron’s in GaN, the hole-initiated APDs are more attractive due to the low excess noise factor and high gain (Zheng et al., 2016; McClintock et al., 2007; Minder et al., 2007). Therefore, backside-illuminated GaN-on-Sapphire APDs (p-type on top) were proposed and fabricated (McClintock et al., 2007; Minder et al., 2007), as shown in Figure 9C. Although the backside-illuminated GaN-on-Sapphire APDs achieved remarkable performance compared to the topside (aka frontside)-illuminated electron-initiated APDs, the backside-illuminated APDs structures will face challenges when transferred from its free-standing GaN substrate. An alternative path was attempted with a buried p-GaN layer. Ionization of acceptors in buried p-GaN layers remains to be a challenge in GaN. Mg is a deep acceptor in GaN with an energy difference of 0.17 eV from the valence band. When doped, about 1% Mg is active to provide free holes in the system. Innovative methods have increased this activation to 10% (Smorchkova et al., 2000; Nakayaa et al., 1996) and created more efficient p-n junctions. It has been vastly studied, that during the MOCVD growth, the Mg dopants in the p-GaN are compensated by H, forming the Mg-H complex (Nakamura et al., 1992a; Nakamura et al., 1992b). If the Mg-doped GaN is buried beneath the n-GaN, because of the limited H diffusion in n-GaN (Czernecki et al., 2018), Mg dopants cannot be activated. To activate the buried Mg-doped GaN, one must etch through the n-GaN layers, exposing the p-GaN to the atmosphere (Li et al., 2018; Narita et al., 2018). Figure 9D presents an n-p structure grown on a GaN substrate reported by Ji et al. (2020b), which is good for topside illumination hole-initiated APDs.

To enhance the pure hole-injection, the separated absorption and multiplication (SAM) design was utilized. In the regular p-i-n structure, the mixed-mode injection can be caused by the absorption of light in the multiplication layer, which may affect the device’s gain and noise. Therefore, by adding a charge layer to separate the absorption and multiplication layers is an effective way to enhance the gain and reduce noise. Figure 10A shows the schematic of a backside illuminated SAM APD structure (McClintock et al., 2009; Zheng et al., 2012; Wang et al., 2014; Shao et al., 2017; Ji et al., 2018; Zhang et al., 2020), and the corresponding electric field distribution along the depletion region. By inserting a Si-doped charge layer to separate the absorption and multiplication layers, the electric field in the absorption layer remains low that avalanche multiplication cannot be triggered in the absorption layer, while the electric field in the multiplication layer remains high where the avalanche can occur. Multiple groups have reported the structure shown in Figure 10A, however, the backside-illumination configuration cannot be achieved with bulk GaN substrates.

Figure 10B highlights a proposed SAM APD structure based on bulk GaN substrates, which is modified from our prior reported structure (Ji et al., 2020b). The structure includes a separated absorption and multiplication layers on top of a buried p-GaN layer. The corresponding electric field distribution is also shown in Figure 10B. Compared to the reported GaN APD structures, the structure shown in Figure 10B has three advantages: 1) homoepitaxial growth using bulk GaN substate; 2) the hole-initiation; and 3) the SAM structure. These combinations show a great potential for producing high-sensitivity UV detectors.

One other important feature of GaN-based APD is its high-temperature tolerance. Because of the wide bandgap, the intrinsic carrier density of GaN, which is proportional to T 2 / 3 e x p ( − E g 2 k T ) , remains low with elevated temperatures. Therefore, the dark current density of GaN-based APDs remains low under high-temperature environments. Our previous work indicated that the GaN APD can be well operated under a temperature up to 525 K (Ji et al., 2020b). GaN APDs thus outshines any competing technologies offering superior performance not only in terms of high gain and low dark current but also high-temperature tolerance. Due to the inert nature of GaN, these APDs are expected to be great candidates in harsh environments and future studies are expected to offer us those discussions.

Last but not least, compared to the photoconductor and phototransistor, p-i-n based photodetectors have the fastest response time, which is determined by three factors: 1) the diffusion time of carriers before they enter the depletion region, 2) the transient time through the depletion region, and 3) the R _L C _p-i-n time delay, where R _L is the load resistance, and the C _p-i-n is the junction capacitance. To minimize the response time, the n-GaN layer thickness should be minimized for the hole-initiated APD, and the p-GaN layer thickness should be minimized for the electron-initiated APD. Additionally, the device should be well designed to minimize the junction capacitance. Table 4 summarizes the reported response time of various GaN photodetectors, it indicates that GaN p-i-n diode and APD have the fastest response speed compared to other types of devices, enabling a short response time of 10^–8 s.

The reported GaN APD structures and performances are summarized in Table 5, which suggests that hole-initiated APD on GaN substrate enabled a record responsivity up to 60 A/W. This result has sparked interest in weak UV light detections.

In this article, we attempted to shed some light on GaN avalanche photodiodes for UV detection and the possibilities, which are of growing interest. GaN APDs are naturally suited to serve application that requires high temperature/harsh environment tolerance. The design space of GaN APDs was reviewed. It is quite clear from prior reports that a robust avalanche capability in GaN APDs are the result of at least two key requirements are discussed: 1) LDD in the epitaxial layers to offer the travel distance for the carriers to establish an avalanche process; and 2) optimized edge termination that can eliminate the localized peak electric field around the device edges. In the context of APD performance, we discussed the importance of the positive temperature coefficient of breakdown voltage, as well as the observation of avalanche-induced electroluminescence that ensures the gain mechanism is reliable. In addition, GaN’s impact ionization coefficients were discussed to highlight the scope of GaN APDs in applications. Since the ratio of hole’s and electron’s impact ionization coefficients (β/α) is so large, the hole-initiated GaN APDs with buried p-GaN layer are more attractive due to the low excess noise and high gain.