High-sensitivity ultraviolet (UV) detectors are imperative in many key fields, such as non-line-of-sight communication, missile plume detection, environmental monitoring, corona detection, UV astronomical research, and biological molecule detection^[1–3]. Benefiting from high avalanche gain factor, high spectral responsivity, long lifetime, and compact size, the avalanche photodiode (APD) is an attractive candidate to replace conventional photomultipliers in the field of weak UV photon detection^[4–6]. Compared with other wide bandgap semiconductors for UV APD development, 4H-SiC is very competitive, owing to its lower defect density and relatively mature processing technologies^[7,8]. 4H-SiC APDs exhibit promising performance, including low dark current, high multiplication gain, and high quantum efficiency (QE).

Along the wide UV wavelength region ranging from 200 to 400 nm, the solar-blind band (240–280 nm) is what people are mostly interested in, which is due to its ultra-low background noise. In literature, although 4H-SiC APDs with both n-i-p and p-i-n structures have been reported with certain success, there is no detailed avalanche mechanism comparison between the two device structures for deep UV detection^[9–14]. Since deep UV light has short penetration depth in 4H-SiC, photo-carriers are mainly excited in the upper region of the APD structure. Then, electrons and holes would have different drift paths for n-i-p and p-i-n APDs. Considering the fact that the impact ionization coefficient of holes in 4H-SiC is much larger than that of electrons at the same electrical field strength, the performance of n-i-p and p-i-n APDs could be quite different. In this work, 4H-SiC n-i-p and p-i-n APDs are designed and fabricated with similar termination structure and processing technology. A detailed comparison in terms of the avalanche mechanism is conducted. It is determined that due to hole-initiated ionization, n-i-p APDs have better performance for deep UV detection.

The SiC APDs are grown on 4 inch (1 inch = 2.54 cm). $n$-type 4H-SiC substrates. As shown in Figs. 1(a) and 1(b), the epi-structure of the n-i-p APD from top to bottom consists of a 0.15 µm $n+$ contact layer ($N_D=1\times {10}^{19}{\mathrm{cm}}^{-3}$), a 0.2 µm $n$ transition layer ($N_D=1\times {10}^{18}{\mathrm{cm}}^{-3}$), a 0.78 µm $n$ avalanche multiplication layer ($N_D=1\times {10}^{15}{\mathrm{cm}}^{-3}$), and a 10 µm $p$ buffer layer ($N_A=3\times {10}^{18}{\mathrm{cm}}^{-3}$). Comparatively, the epi-structure of the p-i-n APD consists of a 0.2 µm $p+$ contact layer ($N_A=2\times {10}^{19}{\mathrm{cm}}^{-3}$), a 0.2 µm $p$ transition layer ($N_A=1\times {10}^{18}{\mathrm{cm}}^{-3}$), a 0.6 µm $p$ multiplication layer ($N_A=1\times {10}^{15}{\mathrm{cm}}^{-3}$), and a 10 µm $n$ buffer layer ($N_D=3\times {10}^{18}{\mathrm{cm}}^{-3}$). Figure 1(c) shows the top view image of the fabricated APD. The first step in the fabrication process of both device structures is mesa etching by using inductively coupled plasma etching. In order to enhance the fill factor, the mesa is etched down to the surface of the multiplication layer (∼0.5 µm). In order to prevent mesa edge breakdown, the photoresist reflow technique is employed to obtain a small positive beveled mesa (∼6°)^[15,16]. As shown in Fig. 2, the electric field profile of a beveled mesa SiC p-i-n APD is simulated under avalanche breakdown voltage ($V_B$), which indicates that the beveled mesa termination is effective for suppressing the peak electric field around the mesa edge through increasing depletion width. Subsequently, the APD surface is passivated by a thermal oxidation layer and followed by a ${\mathrm{SiO}}_2$ layer deposited by plasma-enhanced chemical vapor deposition (PECVD). Both $n$ and $p$ ohmic contact metals use Ni/Ti/Al/Au deposited by $e$-beam evaporation, where the thickness of Ni/Ti/Al/Au is 35 nm/50 nm/100 nm/100 nm, respectively. The contact metals are annealed by rapid thermal annealing at 850°C for 3 min in ${\mathrm{N}}_2$ ambient. Finally, the top contact pad based on a Ti/Au bi-layer is deposited by $e$-beam evaporation. The mesa diameter of both APDs is ∼150 µm, which refers to the top edge of the beveled mesa. Here, it should be noted that the n-i-p APD fabricated by the above process is actually an n-i-p-n diode, in which its avalanche current flows through a forward-biased $p–n$ junction formed between the $p$ epi-layer and the $n$-type substrate. As a result, it is not necessary to expose its $p$ contact layer by deep mesa etching. In a past study, it has been determined that the n-i-p-n APD has similar current-voltage (I-V) and avalanche characteristics to those of conventional n-i-p APDs^[17]. The advantage of adopting this n-i-p-n scheme in current work is that both the n-i-p APD and the p-i-n APD under study could be terminated by the same beveled mesa structure, so that similar high fill factors (∼80%) and electrical field modulation effects could be achieved for the two APDs, which is desired for performance comparison.

Figure 3(a) and 3(b) show the room-temperature (RT) I-V and the gain-voltage curves of the SiC n-i-p and p-i-n APDs, respectively. Both devices exhibit a stable dark current of ∼0.1 pA level before avalanche breakdown occurs. The avalanche gain (M) is calculated by the following equation:$$M=\frac{I_{\mathrm{MP}}{-I}_{\mathrm{MD}}}{I_P{-I}_D},$$(1)where $I_{\mathrm{M}\mathrm{P}}$ and $I_{\mathrm{M}\mathrm{D}}$ are the multiplied photo-current and dark current after avalanche multiplication, respectively, while $I_P$ and $I_D$ are photo-current and dark current before avalanche multiplication, respectively. In this work, $I_P$ and $I_D$ are measured at 50 V reverse bias. If $V_B$ is defined at the multiplication gain of 100, the $V_B$ is found at ∼257.5 V and ∼201.2 V for the n-i-p and p-i-n APDs, respectively. At higher overbias, the multiplication gains can reach over ${10}^5$.

In the spectral response measurement, a high-power xenon lamp is used as the light source and a Horiba iHR320 monochromator is employed to sort monochromatic light. A UV-enhanced Si photodiode (Hamamatsu S1226-8BQ) is used to calibrate the incident light power density. The zero-bias spectral response characteristics of two APDs are plotted in Fig. 4, which exhibit similar shape and magnitude. The peak responsivity of the n-i-p APD is ∼0.114 A/W (QE ∼50%) at 285 nm. Comparatively, the p-i-n APD has a maximum QE of ∼48% at 285 nm. The UV/visible (285 nm/400 nm) rejection ratio of both devices is higher than ${10}^4$. In addition, the overall QE of the p-i-n APD is slightly lower than that of the n-i-p APD over the whole UV wavelength region, which should be caused by its slightly thicker $p+$ contact layer.

Next, the single photon detection performance of the n-i-p and p-i-n APDs is compared, which is characterized by a passive quenching circuit [the inset in Fig. 5(a)]. The APDs are biased above $V_B$, and the avalanche events are quenched by a 50 kΩ quenching resistor. The voltage pulse signals are recorded by a high-speed oscilloscope, which is connected with a 100 Ω sampling resistor in parallel. A 265 nm UV LED is used to evaluate the detection capability of devices, and the calibrated incident UV photon flux is $\sim 1.2\times {10}^7$ photons/s. A typical avalanche voltage pulse signal is shown in Fig. 5(a). The RT dark count rate (DCR) of the two APDs is shown in Fig. 5(b) as a function of overbias. Since the $V_B$ of the two APDs is different due to the small thickness difference of their multiplication layers, here, overbias is intentionally normalized to respective $V_B$ for fair comparison. A good exponential dependence of DCR on overbias is observed, which agrees with past reports that trap-assisted tunneling is the main source of DCR in current SiC APDs^[18]. The DCR of the n-i-p APD is lower than that of the p-i-n APD. For example, at 1% overbias the DCR of the n-i-p APD is ∼13 kHz, while the DCR of the p-i-n APD is ∼28 kHz. The single photon detection efficiency (SPDE) is calculated by using the following equation:$$\mathrm{SPDE}=\frac{\mathrm{PCR}-\mathrm{DCR}}{n}\times 100\%,$$(2)where $n$ is the total number of incident UV photons of a specific wavelength per second, and PCR is photon count rate. At the same overbias, the SPDE of the n-i-p APD is higher [see Fig. 5(c)]. For example, at 1% overbias, the SPDE of the n-i-p APD is ∼2 times higher than that of the p-i-n APD. With the increase of overbias, the SPDE increases due to the enhanced photo-carrier avalanche multiplication probability at a high electric field. At the same DCR, the SPDE of the n-i-p APD is considerably higher [see Fig. 5(d)], which certainly cannot be explained by the slightly higher zero-bias QE of the n-i-p APD.

The capacitance of the APDs is calculated by the following equation:$$C=\frac{{\text{εε}}_{0}S}{d},$$(3)where $\epsilon$, ${\epsilon }_0$, $S$, and $d$ are relative permittivity, vacuum permittivity, effective area, and depletion layer width of the APD, respectively. Based on this equation, capacitance of the n-i-p and the p-i-n APDs is estimated at ∼1.9 pF and ∼2.5 pF, respectively. The experimental capacitance of the n-i-p and p-i-n APDs is ∼3–4 pF, which is larger than that of the calculated geometric values due to the parasitic effects of measurement cables and probes during on-wafer testing. The time from avalanche onset to avalanche quenching is called the quenching time, which can be calculated by the following equation:$$T_q\approx {(C}_{\mathrm{APD}}{+C}_s{)\times R}_{\mathrm{APD}}.$$(4)

The time taken to bring the APD back to its original state is called the reset time, which is calculated by the following equation:$$T_r{=(C}_{\mathrm{APD}}{+C}_s{)\times R}_L,$$(5)where $C_{\mathrm{APD}}$ and $C_s$ are the junction capacitance of the APD and stray capacitance, respectively, while $R_{\mathrm{APD}}$ is the resistance of the APD. $R_{\mathrm{APD}}$ is ∼150 Ω derived from the linear region of the forward I-V curve. From Fig. 5(a), the quenching time and reset time are ∼1 ns and ∼300 ns, respectively, so the stray capacitance of the passive quenching circuit is ∼ 5 pF. During the quenching process of avalanche multiplications, the time during which the APD is not responsive to further incoming photons is called dead time. The dead time includes the avalanche quenching time and the APD reset time. However, when the APD is biased beyond $V_B$, it is able to detect the next photon prior to being fully reset. Thus, apparent fluctuations in the reset waveform can be observed [see Fig. 5(a)].

To understand the detailed avalanche mechanisms of SiC APDs, the photon-induced multiplication characteristics are measured at various wavelengths ranging from 220 nm to 330 nm. Figures 6(a) and 6(b) show the selected gain-voltage curves of the n-i-p and p-i-n APDs obtained under 260 nm, 280 nm, and 300 nm UV illumination, respectively. It is found that with increasing illumination wavelength, the gain-voltage curves of the n-i-p APD shift towards higher voltage with a corresponding higher $V_B$. Meanwhile, an opposite trend is observed for the p-i-n APD. The detailed evolution behavior of $V_B$ as a function of wavelength is summarized in Fig. 6(c). A $V_B$ shift up to 2 V can be observed, which is fairly large compared with normal overbias applied during Geiger mode operation.

The following physical picture can explain the obvious difference in terms of single photon detection performance and $V_B$ shift trend versus wavelength for the two APDs. As illustrated in Fig. 7, electron–hole pairs are mainly excited in the upper region of the depletion layer under front deep UV illumination due to the large absorption coefficient of 4H-SiC for high-energy photons^[19]. For n-i-p APDs under reverse bias, most holes would drift across the whole depletion layer towards the $p$ contact layer, having an average long acceleration and multiplication distance. On the other hand, most electrons would be quickly swept back to the $n$ contact layer, having much less chance for impact ionization. Thus, for n-i-p APDs in the deep UV region, the dominated carriers triggering avalanche events are photo-excited holes. At a higher wavelength, the light penetration depth would increase, which leads to more electron–hole pairs being excited in the deeper region of the depletion layer. Then, some electrons would gain enhanced impact ionization probability due to their longer drift and acceleration distance. Therefore, with increasing wavelengths from deep UV to more than 300 nm, the avalanche events gradually vary from dominant hole-initiations to mixed carrier-initiations. The situation of the p-i-n APD is opposite to that of the n-i-p APD. At the deep UV wavelength region, the avalanche events in p-i-n APDs are mainly electron-initiations, while the chance of mixed carrier initiations increases at higher wavelengths. Specifically, according to the simulated absorption spectrum of 4H-SiC, the penetration depths of 240 nm and 300 nm UV light are ∼50 nm and ∼2.5 µm, respectively^[19]. Thus, under 240 nm UV illumination, the avalanche events in SiC n-i-p APDs should be mainly hole-initiations, while the dominated mechanism for avalanche multiplications in SiC p-i-n APDs is the electron-initiation. As the UV wavelength increases to 300 nm, avalanche events in both types of APDs should be mixed carrier-initiations. Considering the fact that the ionization coefficient of holes in 4H-SiC is considerably larger than that of electrons^[20–23], dominant hole initiation means smaller critical electric field and higher gain factor. Because of this, the gain-voltage curves are strongly UV wavelength dependent. In addition, the lower critical electric field also explains why the n-i-p APD with dominant hole-initiated avalanche multiplications has lower DCR.

In summary, vertical Geiger mode 4H-SiC n-i-p and p-i-n APDs have been fabricated and characterized. The n-i-p APD exhibits an overall better single photon counting performance for deep UV detection, which can be explained by its dominant hole-initiated avalanche characteristics. Comparatively, the avalanche events in p-i-n APDs are mainly electron-initiations in the deep UV wavelength region. Since the impact ionization coefficient ratio of holes to electrons is much larger than one, 4H-SiC APDs with dominant hole-initiated avalanche would feature relatively lower critical electric field strength and then fewer dark counts. In addition, this is the first time, to the best of our knowledge, to report single photon detection characteristics of vertical SiC n-i-p-n APDs with an enhanced fill factor.