AlGaN-Based Ultraviolet PIN Photodetector Grown on Silicon Substrates Using SiN Nitridation Process and Step-Graded Buffers

Abstract

The integration of aluminum gallium nitride (AlGaN) with silicon substrates attracts significant attention due to the superior UV sensitivity of AlGaN and the cost-effectiveness as well as mechanical robustness of silicon. A PIN ultraviolet photodetector with a peak detection wavelength of 274 nm is presented in this paper. By employing a SiN nucleation layer and a step-graded buffer, a high-quality AlGaN-based photodetector structure with a dislocation density of 2.4 × 10⁹/cm² is achieved. A double-temperature annealing technique is utilized to optimize the Ohmic contact of the n-type AlGaN. The fabricated UV photodetector attains a dark current of 0.12 nA at −1 V and a peak responsivity of 0.12 A/W.

1. Introduction

Ultraviolet (UV) photodetectors are indispensable in various applications, including environmental monitoring, flame detection, biological research, and space exploration [1,2,3,4]. Historically, photomultiplier tubes and silicon photodetectors were the primary technologies for ultraviolet photodetectors. But they have limitations such as achieving high solar blindness and rejecting near-UV and visible spectra [5]. Aluminum gallium nitride (AlGaN) is a suitable semiconductor material ideal for ultraviolet (UV) photodetectors, especially in the solar-blind region (below 290 nm), due to its tunable wide bandgap. This property is crucial for applications requiring the detection of UV radiation while filtering out visible and near-UV wavelengths [6,7,8,9]. The tunability of AlGaN’s bandgap, achieved through the modulation of the aluminum content, enables a variation in bandgap energy ranging from 3.39 to 6.024 eV, effectively encompassing the ultraviolet spectrum and particularly the entire solar-blind wavelength region. This characteristic is highly advantageous for the development of efficient solar-blind photodetectors.

While sapphire has traditionally been the preferred substrate for AlGaN-based photodetectors due to its UV transparency, the use of silicon (Si) substrates is becoming increasingly popular because of their lower cost, compatibility with standard semiconductor processes, and availability in large diameters [10]. However, significant challenges arise when growing AlGaN on Si, including a substantial lattice mismatch (~17%) and thermal expansion differences that lead to high dislocation densities and potential cracking. These issues can severely deteriorate the performance of the photodetectors, making it necessary to develop advanced techniques for enhancing the crystal quality of the AlGaN layers grown on Si [11,12,13].

One successful approach to address these challenges is the utilization of AlN template layers, which serve as a buffer to alleviate lattice mismatch effects and reduce dislocation densities. The crack density decreases with the increase of AlGaN layer thickness. When the thickness of the AlGaN layer reaches approximately 250 nm, the crack density will tend to zero. However, the cracks reappear when the film thickness increases further [14]. The occurrence of cracks can be reduced by introducing an about 70 nm thick AlN nucleation layer as well as a GaN/AlGaN superlattice [15]. According to the former researchers’ reports, as noted in ref. [16,17], cracking can be avoided during sample cooling by preventing the formation of amorphous Si₃N₄ on the Si surface in the early growth stages and ensuring uniform AlN nucleation [16,17]. Step-graded AlGaN layers are also used to prevent stress relaxation in GaN on Si layers, acting as a dislocation filter and achieving a smooth surface morphology for crack-free epitaxial AlGaN layers [18,19,20,21,22]. Techniques like maskless lateral epitaxial overgrowth have also been utilized to further improve the crystalline quality by decoupling the epilayers from the substrate, enabling the growth of thick, crack-free layers [23].

The research group Rzhanov Institute of Semiconductor Physics in Russia, has been investigating the effect of a monolayer SiN film on the surface states of the AlN/GaN heterostructures grown by molecular beam epitaxy since 2015 [24,25,26]. Al atoms deposited on top of a highly ordered (8 × 8) structure can help the formation of a graphene-like AlN (g-AlN) layer, which greatly enhances the growth quality of Si/AlGaN materials.

In this paper, we report a high performance AlGaN-based ultraviolet PIN photodetector grown on a Si (111) substrate using a graphene-like AlN nucleation layer and step-graded AlGaN buffers. The tested threading dislocation density (TDD) of the PIN material is 2.4 × 10⁹/cm². Through the application of the double-temperature annealing technique, the resistance of the n-type Ohmic contact can be reduced to 3.4 × 10⁻⁵ Ωcm². After an optimization of the fabrication process, the detectors achieve a dark current of 0.12 nA@−1 V and a peak responsivity of 0.12 A/W@274 nm.

2. Materials and Methods

The AlGaN-on-Si photodetector materials were grown by molecular-beam epitaxy on 2-inch diameter Si (111) substrates in Riber CBE-32 system. The Si (111) substrates were pretreated using the wet chemical etching procedure to form the hydrogen-saturated surface [27,28,29]. After the etching, the substrates were dried by high purity N₂ with 5-nines (5N) purity and introduced into the vacuum chamber for 15–20 min. Subsequently, the outgas procedure was carried out in a preparatory vacuum chamber at 600 °C for two hours. The substrate temperature was controlled by an infrared pyrometer. The high-purity ammonia (8N) utilized as the active nitrogen underwent an additional purifying process via a filter. The aluminum cold lip source with cold neck pyrolytic boron nitride (PBN) crucible and gallium double zone source with pyrolytic graphite (PG) crucible were employed as the gallium and aluminum sources, respectively.

In order to completely remove the surface oxide, the silicon substrates underwent high-temperature annealing up to 1100 °C, resulting in the formation of a 7 × 7 reconstruction on the reflection high-energy electron diffraction (RHEED) pattern of the Si (111) surface upon cooling below 830 °C. The de-oxidation process occurred during the annealing procedure, forming a clean and smooth surface for the nitridation of the silicon substrates.

After the surface annealing process, a standard initial nitridation process in NH₃-MBE was performed. The Si substrate was heated to 950 °C and exposed to the ammonia flux of 10 sccm. Within a few seconds, the SiN-8 × 8 structure fractional reflections appeared in the RHEED patterns. It is believed that the dominant role in the formation of the SiN-8 × 8 structure was played by mobile Si adatoms that were in equilibrium with the Si surface [30]. After 40 s of nitridation, graphene-like SiN covered the silicon surface and the intensity of the 8 × 8 refractive patterns reached a maximum. Further increasing the nitridation time would form an amorphous Si₃N₄ layer on the silicon surface, and the diffraction pattern would disappear after several minutes.

An AlN flat ultrathin nucleation layer was prepared by depositing Al while the ammonia flux was off and achieving a background ammonia pressure of ~10⁻⁷ to 10⁻⁸ Torr. The ammonia flux was turned off when the optimal 8 × 8 RHEED pattern with sharp and bright eightfold fractional spots appeared. During the Al deposition, the source flux was set to match an AlN growth rate of ~0.1 ML/s. After approximately 5 MLs g-AlN growth, an ~150 nm thick AlN bulk layer was grown as a buffer layer for the step-graded buffer and the full photodetector structure growth.

The step-graded buffer consisted of 3 different composition AlGaN layers. After the 150 nm AlN layer, 150 nm Al_0.85GaN, 150 nm Al_0.75GaN, and 200 nm Al_0.6GaN buffer layers were grown sequentially to enhance the strain relaxation and dislocation annihilation.

The entire photodetector structure, which has been verified as effective by the Center for Quantum Devices in Northwestern University (CQD) [31], was grown following the step-graded buffer. On top of the last step of the step-graded buffer, the Al_0.6GaN buffer layer, a 600 nm thick Si doped n-type Al_0.5GaN conduction layer was grown. This layer also served as a dislocation annihilation layer to further reduce the dislocation density. Following this highly conductive layer, the p-i-n active region consisted of 35 nm Si-doped n-type Al_0.45GaN with a doping concentration of 7.2 × 10¹⁷/cm³, followed by 200 nm not intentionally doped (NID) intrinsic absorber region, and 50 nm Mg-doped p-type Al_0.38GaN with a doping concentration of 2.4 × 10¹⁶/cm³ was grown. To facilitate the formation of the Ohmic contacts, a 100 nm Mg doped p-type GaN with a doping concentration of 2.1 × 10¹⁸/cm³ was grown as a cap layer. The scanning electron microscopy (SEM) cross-sectional images of the entire epitaxial structure and the microscopic images of the grown material’s surface under 100× and 1000× magnifications are shown in Figure 1.

The epitaxial wafer was cleaned in acetone and ethanol before the subsequent processing procedure to remove surface-organic and pollutants. A photodetector mesa was etched with inductively coupled plasma (ICP) to the n-contact layer and protected with SiO₂ grown by plasma-enhanced chemical vapor deposition (PECVD) to suppress the leakage current on the mesa side. P- and n-type contact windows were opened on SiO₂ with photolithography and reactive ion etching (RIE). After that, the p-type Ti/Pt/Au and n-type Ti/Al/Ti/Au were deposited with magnetron sputtering and evaporation, respectively, and protected with SiO₂ grown by PECVD again. Finally, two electrode pads were symmetrically deposited following the opening contact pad windows. The fabrication procedure for the photodetectors is illustrated in Figure 2.

3. Results and Discussion

The substrate preparation process, including high-temperature annealing for 7 × 7 reconstruction and pre-nitridation of the Si substrate for 8 × 8 reconstruction with a SiN-like intermediate layer has been extensively studied previously [16,17,24,25,26,30]. Figure 3a presents a transmission electron microscope (TEM) high-angle annular dark field (HAADF) image depicting the Si/AlN interface along with the step-graded buffers. At the Si/AlN interface, the nucleation layer functions as a dislocation filter, localizing stress on the interface by generating misfit dislocations. This effect is accomplished by introducing both SiN and AlN ultrathin flat nucleation layers during the nitridation process. Additionally, the TEM dark field image of the interface is also shown in Figure 3b, where a regularly arranged stress field can be seen, indicating periodic stress release at the Si/AlN interface.

The dark field image corresponding to step-graded buffers is also presented in Figure 3c. The orange dash dot lines indicate interfaces within these graded structures. By finely tuning the growth parameters across the steps, it becomes possible to terminate dislocation lines at these interfaces, as illustrated by red arrows. To further assess the effectiveness of these step-graded buffers, additional TEM HAADF images were employed to capture inclined dislocation configurations between AlGaN layers as shown in Figure 4.

Former analysis has elucidated mechanisms underlying stress relaxation and their correlation with dislocation inclinations [32,33]. As observed in Figure 4a, dislocations exhibit abrupt inclinations at the interface between different graded AlGaN layers. The projected length of the inclined dislocation acts as a misfit dislocation segment to relax the compressive strain [33]. It is posited that greater lattice mismatch strains induce larger bend angles of dislocation inclination. This can be proved in Figure 4b, which reveals that significant compositional differences between Al_0.75GaN and Al_0.6GaN yield larger inclination angles compared to those observed between Al_0.85GaN and Al_0.75GaN, thereby enhancing opportunities for interactions among dislocations.

The dislocation density in the step-graded buffer layers can be expressed using an empirical formula proposed by Fitzgerald et al. [34,35]:

$${\mathsf{\rho }}_{\mathrm{t}}={\displaystyle \frac{2{\mathrm{R}}_{\mathrm{g}}{\mathrm{R}}_{\mathrm{g}\mathrm{r}}{\mathrm{e}}^{\frac{\mathrm{U}}{\mathrm{k}\mathrm{T}}}}{\left|\mathrm{b}\right|\mathrm{B}{\mathrm{Y}}^{\mathrm{m}}{\mathsf{\epsilon }}_{\mathrm{e}\mathrm{f}\mathrm{f}}^{\mathrm{m}}}}\propto {\displaystyle \frac{{\mathrm{R}}_{\mathrm{g}}{\mathrm{R}}_{\mathrm{g}\mathrm{r}}}{\mathrm{V}}}$$

(1)

where r_t is the threading dislocation (TD) density, B is a constant with a unit of velocity, Y is Yang’s modulus, ε_eff is the strain reduced by the dislocation flow, m is an exponent between 1 and 2, R_g is the growth rate, R_gr is the variation rate (lattice mismatch per unit thickness), U is the dislocation glide activation energy, k is the Boltzmann constant, T is the absolute temperature, and V is the dislocation slip velocity.

According to the empirical formula, the TD density exhibits a positive correlation with both growth rates and variations between respective grading structures. This is evidenced in Figure 4c, where an increased number of dislocation lines is generated at the interface between AlN/Al_0.85GaN and Al_0.75GaN/Al_0.6GaN. In contrast, at the interface between Al_0.85GaN/Al_0.75GaN, dislocation lines tend to merge, resulting in a reduction of TDs (indicated by yellow circles).

Considering these mechanisms, the careful design of step-graded buffer layers is essential to enhance the bend angle of dislocation lines, thereby facilitating TDs annihilation through increased compositional differences between step layers while simultaneously reducing interface dislocation generation by decreasing either growth rate or compositional variation rate across steps. Following systematic optimization of growth parameters, the TD density for the entire photodetector structure was estimated to be 2.4 × 10⁹/cm² based on XRD FWHM and TEM analyses.

An important factor limiting the performance of the solar-blind UV devices is achieving effective Ohmic contacts between high Al composition AlGaN and the Ohmic contact metals. It is well known that the increase in the aluminum fraction raises both the band gap and the election affinity of AlGaN, worsening the Ohmic contact barriers, causing nonlinear behavior, and elevating the device resistivity [36,37,38,39]. To optimize the contact resistivity, we introduced a double-temperature annealing technique to improve the fabricated 30 nm-Ti/100 nm-Al/30 nm-Ti/30 nm-Au Ohmic contact performance of the n-type AlGaN.

L. Wang et al. [40] conducted a comprehensive analysis of the n-type Ohmic contact in GaN over a temperature range of 400–950 °C, determining that the optimal annealing temperature for Ti/Al/Mo/Au is 850 °C. We have adopted this finding and established our primary annealing temperature at 850 °C. Figure 5 illustrates the IV curve of the metal pad used for the circular transmission line method (CTLM) Ohmic contact test under different annealing conditions. And the annealing condition and the calculated resistivity are listed in

Table 1. Resistivity of the n-type Ohmic contact under different annealing conditions.

Annealing Condition	Resistivity (Ω·cm2)	Profile
As grown	0.1087	Nonlinear
850 °C 10 s	0.0660	Nonlinear
850 °C 60 s	0.0075	Linear
900 °C 30 s	0.0032	Linear
350 °C 1 min + 900 °C 30 s	0.0011	Linear
350 °C 10 min + 900 °C 30 s	3.43 × 10−5	Linear

. As depicted in Figure 5, the as-grown sample exhibits pronounced modulation characteristics indicating Schottky contact behavior at the metal/Al_0.5GaN interface. Annealing at 850 °C improves the Ohmic behavior of the contact. However, if the annealing time is too short, the effectiveness of annealing will be affected; if the annealing time is too long, the surface morphology of the contact pad will be greatly deteriorated (as shown in the inset of Figure 5, deteriorated surface of the metal pad after annealing at 850 °C for 60 s). Further increasing the annealing temperature or the time optimizes Ohmic contact characteristics within a limited range but compromises surface quality.

To further decrease the resistivity of the Ohmic contact, a low-temperature annealing procedure was introduced before high-temperature annealing. As shown in Table 1, after 10 min at 350 °C and 30 s at 900 °C, the resistivity of the n-type Ohmic contact reduced to 3.43 × 10⁻⁵. It is believed that the reaction between Ti and Al starts at lower temperatures (250–300 °C), forming primary products such as Al₃Ti and α-Ti [39]. Also, Au is mobile at this low annealing temperature, allowing it to diffuse into the interior of the metal contacts and react with Al and Ti, forming AlAu_xTi alloys. The formation of this alloy layer hinders the out diffusion of Al/Ti and prevents the oxidation of aluminum from generating Al/Ti-oxide at a high temperature and reduces conductivity. Annealing at 900 °C after low-temperature annealing promotes the further in-diffusion of Ti to AlGaN, form TiN at the metal/AlGaN interface and generate nitrogen vacancies in the AlGaN layers that help form 2DEG at the metal-semiconductor interface and finally reduce the resistivity of the Ohmic contact [40].

Figure 6 illustrates the responsivity and the dark current of the fabricated photodetectors. The dark current of the 100 um diameter device is tested to be 0.12 nA at −1 V. The responsivity of the photodetectors was measured using a high-intensity xenon arc lamp, monochromator, and reference calibrated UV-enhanced silicon photodetector. The tested peak responsivity of the photodetector is 0.12 A/W at −5 V and the peak detection wavelength is 274 nm. A low photo response persists because of the weak Schottky-photodetector-like behavior of the p-type contact and should be eliminated by optimizing the p-type Ohmic contacts. To validate the performance of this device, we also compared it with previously reported similar devices as presented in

Table 2. Comparison of AlGaN UV photodetectors.

Year	Substrate	Type	Wavelength	Dark Current	Responsivity
2001 [41]	Si	PN	310 nm	200 uA@5 V	6 mA/W
2007 [42]	Si	MSM	297 nm	7.5 × 10−9 A/cm2	110 mA/W
2013 [5]	Saphire	PIN	275 nm	2 × 10−9 A/cm2@10 V	176 mA/W
2015 [43]	Si	PIN	290 nm	1.6 × 10−8 A/cm2	18.3 mA/W
2017 [44]	Si	nanorods	276 nm	-	115 mA/W
2017 [45]	Si	MSM	362 nm	0.43 nA@15 V	0.183 A/W
2020 [46]	Si	MSM	365 nm	3.3 × 10−7 A/cm2	-
2020 [47]	Si	V-Pit MSM	315 nm	0.7 nA	125 A/W@5 V
2020 [48]	Saphire	PIN	270 nm	-	190.96 mA/W
2020 [49]	Si	HEMT	365 nm	2.9 × 10−8 mA/mm	2 × 104 A/W
2020 [50]	Si	Disk	240 nm	10 nA@6 V	-
2020 [51]	Si	Schottky	~278 nm	3 × 10−8 A/cm2	-
2021 [52]	Si	HEMT	360 nm	~1 × 10−7 mA/mm	3.5 × 105 A/W
This work	Si	PIN	274 nm	0.12 nA	120 mA/W

.

4. Conclusions

An AlGaN-based UV PIN photodetector fabricated on Si(111) substrates is reported in this study. The incorporation of silicon nitridation, in conjunction with an AlN nucleation layer and step-graded buffer layer growth, enhances the quality of the AlGaN/Si structure and reduces the dislocation density of the whole photodetector structure to 2.4 × 10⁹/cm². The application of a double-temperature annealing technique effectively lowers the resistance of the n-type Ohmic contact by facilitating the formation of AlAu_xTi alloys at relatively low temperatures and TiN at elevated temperatures. Consequently, the optimized n-type resistivity for the metal/AlGaN contact has been reduced to 3.43 × 10⁻⁵ Ω·cm². Following process optimization, the detectors exhibit a dark current of 0.12 nA at −1 V and a peak responsivity of 0.12 A/W at 274 nm.