Gallium nitride (GaN) has attracted a great deal of attention as a next-generation power electronic material to replace silicon (Si) due to its excellent characteristics such as high electron mobility and high breakdown field (Chowdhury et al., 2013; Chowdhury and Mishra, 2013). In general, GaN thin films are grown using a metal–organic chemical vapor deposition (MOCVD) system, and Mg doping for p-type GaN (p-GaN) remains as a last step on the top of epitaxial layers due to the large diffusivity of Mg from the p-GaN surface and the reactor wall of MOCVD (Ran et al., 2006). Evidently for the Ga-polar p-GaN thin-film growth (Benzarti et al., 2008), these residual Mg dopants unintentionally diffuse into Al _x Ga _1-x N-based thin films that are regrown on p-GaN, even though Mg source (Cp₂Mg) injection into a reactor was ceased (Fichtenbaum et al., 2008). Nevertheless, embedding p-GaN has recently emerged as an important research topic to create more efficient and higher-performance electronic devices, e.g., lateral AlGaN/GaN high-electron-mobility transistors (HEMTs) including a p-GaN buffer layer (Lee et al., 2014), vertical super-junction metal-oxide-semiconductor field-effect transistors (MOSFETs) using a field-stop p-GaN layer (Xiao et al., 2019; Zhou et al., 2019), and current aperture vertical electron transistors (CAVETs) with a current blocking p-GaN layer (Chowdhury et al., 2008; 2011).

To date, various solutions have been suggested to suppress Mg out-diffusion such as ex situ wet chemical treatment using hydrofluoric acid (Xing et al., 2003), the in situ dry etching process with H₂ carrier gas (Soman et al., 2019), alternative regrowth using molecular beam epitaxy (MBE) instead of using MOCVD (Chowdhury et al., 2011; Brown et al., 2013; Wu et al., 2021), and the introduction of low-temperature aluminum nitride (LT-AlN) or LT-GaN insertion layers as an Mg suppression layer (MSL) (Tomita et al., 2008; Agarwal et al., 2017). Agarwal et al. (2017) recently reported the LT flow-modulated epitaxy (referred to as pulsed LT-GaN) technique for high crystallinity of intrinsic GaN (i-GaN) regrown on p-GaN with improved surface morphology as well as suppression of Mg out-diffusion. As the demand for embedding p-GaN in the Al _x Ga _1-x N thin film without Mg diffusion increases explosively, it is also important to explore the characteristics of MSLs through an electronic device embedding p-GaN.

In this study, we systematically investigate the efficacy of LT-GaN regrown on p-GaN as the MSL and its dependence on MOCVD growth conditions. Two types of MSL with and without perturbation (or pulsed injection) of a trimethylgallium (TMGa) metal–organic source were employed to investigate Mg out-diffusion: (1) an MSL with a continuous injection (CLT-GaN) and (2) an MSL with a pulsed injection (PLT-GaN) of the TMGa source. Subsequently, Al_0.3Ga_0.7N (40 nm)/i-GaN (140 nm) thin films for prototype HEMTs were regrown on MSLs to explore the decay rate of Mg diffusion and the two-dimensional electron gas (2DEG) characteristics. Secondary ion mass spectrometry (SIMS) depth profiles were measured to analyze the Mg diffusion into regrown Al_0.3Ga_0.7N/i-GaN according to the MOCVD growth condition of LT-GaNs. Capacitance–voltage (C–V) and current–voltage (I–V) measurements were also taken to understand 2DEG and the output/transfer characteristics of Al_0.3Ga_0.7N/i-GaN HEMTs with our LT-GaN MSLs. This work demonstrates the importance of suppressing Mg out-diffusion in device performance and shows the correlation between the degree of Mg diffusion and 2DEG density according to various growth conditions of existing LT-GaN.

A 300-nm-thick p-GaN and 3 µm-thick i-GaN buffer layer were grown at 930°C and 1,040°C, respectively, by using the MOCVD system. The Al_0.3Ga_0.7N (40 nm)/i-GaN (140 nm) thin films including the MSL were regrown with the ex situ process for the HEMTs. There were two types of samples grown in this study, as shown in Supplementary Figure S1: (A) HEMT structures on a sapphire substrate with various LT-GaN techniques for the C–V measurement using a Hg-probe tool and (B) the same structures on a GaN substrate for the device characteristics of lateral HEMTs. Figure 1A shows a schematic illustration of Al_0.3Ga_0.7N/i-GaN with LT-GaN MSLs regrown on the p-GaN layer. First, two p-GaNs with different Mg doping concentrations (1.0E19 and 4.0E19 cm⁻³) were grown on sapphire substrates. After that, two types of the MSL, CLT-GaN and PLT-GaN, were introduced to investigate Mg out-diffusion from p-GaN. The CLT-GaN MSL was regrown under a growth pressure of 600 mbar at 750°C with a consistent flow rate of TMGa and NH₃ sources into a H₂ atmospheric MOCVD reactor. Meanwhile, 50 cycles of (i) TMGa interruption (5 s) and (ii) TMGa injection (3 s) were implemented for the PLT-GaN MSL at the same growth pressure and temperature as the CLT-GaN MSL. Figure 1B shows the in situ reflectance and growth temperature during the MOCVD process for the Al_0.3Ga_0.7N/i-GaN/PLT-GaN thin-film regrowth on p-GaN. The part of the PLT-GaN growth step in Figure 2B is shown enlarged in Figure 1C, where two different growth rates repeated 50 times in one cycle can be observed: (i) a TMGa interruption step (5 s) and (ii) a TMGa injection step (3 s). On the other hand, the growth behavior of CLT-GaN is continuous. To investigate the suppression of Mg out-diffusion by applying these LT-GaN MSLs, the MOCVD growth temperature and thickness of LT-GaN were controlled as summarized in Table 1. Subsequent regrowth was conducted for 140-nm-thick i-GaN and 40-nm-thick Al_0.3Ga_0.7N at 1,025°C and 1,080°C, respectively, to form the 2DEG structure. An XRD 2theta-omega scan of samples was performed, as shown in Supplementary Figure S6. For device fabrication, the mesa etching was performed using a dry etching process with inductively coupled plasma reactive ion etching (ICP-RIE). For the source electrodes, Ti, Al, Ni, and Au were deposited on Al_0.3Ga_0.7N followed by post-annealing using a rapid thermal annealing (RTA) system at 850°C for 45 s to form ohmic contacts. Al₂O₃ (12 nm thick) was deposited as a gate dielectric through atomic layer deposition (ALD). Ni/Au was deposited for the formation of the gate contact.

Supplementary Figure S2 shows the AFM images of the grown 2DEG channel layer on sapphire substrates. RMS roughness shows significant differences between CLT- (∼1.4 nm) and PLT LT-GaN (∼0.40 nm). Figure 2 shows the SIMS depth profile for the regrown Al_0.3Ga_0.7N/i-GaN to evaluate Mg out-diffusion. The black dotted lines in Figure 2 indicate the p-GaN borderline or the place where the regrowth of Al_0.3Ga_0.7N/i-GaN with the MSL starts. The fitting lines for extracting the Mg suppression rate are overlapped with the actual SIMS results. For the as-grown sample without MSL (Figure 2A), the decay rate of Mg diffusion in the regrown Al_0.3Ga_0.7N/i-GaN region was estimated to be ∼230 nm/decade. This result generally shows the diffusivity of Mg into regrown thin films on p-GaN. On the other hand, 10-nm-thick CLT-GaN as an MSL dramatically reduces the Mg diffusion rate (∼27 nm/decade) in a regrown film. It was also found that CLT-GaN with a thickness of 100 nm further reduces Mg diffusion to ∼21 nm/decade. The lowest Mg concentrations at 150-nm thicknesses of the regrown Al_0.3Ga_0.7N/i-GaN layer were 6.69E18 cm⁻³ for as-grown, 9.75E17 cm⁻³ for 10-nm-thick CLT-GaN grown at 750°C, and 5.94E17 cm^−-3 for 100-nm-thick CLT-GaN grown at 750°C. As Agarwal et al. (2017) previously reported, the PLT-GaN MSL regrown on p-GaN not only improves surface morphology but also suppresses Mg out-diffusion. Figure 2B shows that the 100-nm-thick PLT-GaN MSL grown at 750°C has a similar Mg suppression rate (∼26 nm/decade) compared to the CLT-GaN MSL under the same growth conditions, but the lowest Mg concentration at 150 nm is reduced by almost half (3.67E17 cm⁻³). Generally, these LT growth methods introduce various crystallography defects as well as impurities (C, O, and H) from byproducts of organic sources. The regrowth at a temperature lower than 750°C showed poor crystallinity of the thin films (not shown in this study) regardless of Mg suppression. For the other case, when the growth temperature is increased from 750°C to 900°C, 100-nm-thick PLT-GaN loses most of its role as the MSL, which indicates that Mg diffusion is very sensitive to the p-GaN surface temperature. Additionally, SIMS measurements, as shown in Figures 2C, D, obviously displayed the influence of growth temperature for the LT-GaN MSLs on the impurity concentration and the suppression of Mg diffusion. The concentration of O and C in the regrown Al_0.3Ga_0.7N/i-GaN layer decreases by more than one order of magnitude as the growth temperature of the MSL increases up to 900°C, but the suppression of Mg diffusion is significantly reduced. Since the impurity level strongly depends on the growth temperature, the impurities contained when inserting CLT-GaN and PLT-GaN grown at the same temperature are similar. Because of this reciprocal relationship, further studies on the LT-GaN MSL are required to simultaneously achieve Mg diffusion suppression and impurity reduction.

To investigate the 2DEG characteristics of regrown Al_0.3Ga_0.7N/i-GaN, the C–V measurement was obtained, as shown in Figure 3. Figure 3A shows the simulated energy band diagram of our Al_0.3Ga_0.7N/i-GaN. Bandgaps and offset values were obtained by the Silvaco TCAD (Sharbati et al., 2021). The calculated ΔE _c and ΔE _v were 377 and 171 meV, respectively. As shown in Figure 3B, the C–V measurements were directly obtained from the unprocessed samples at 1 MHz using a mercury (Hg)-probe system (MDC Model 802), which is commonly used for the C–V characterization of HEMTs without device fabrication (Cordier et al., 2008; Tucker et al., 2022). The Hg C–V system provides a non-destructive electrical characterization and employs a unique dot-ring configuration by forming a Schottky contact (dot contact) and an ohmic contact (ring contact) with the non-metalized planar semiconductor surface (Enisherlova et al., 2008). The C–V profiles show the measured capacitance with the bias voltage on the Schottky contact swept from 1 to −10 V, and the 2DEG sheet carrier density is calculated using the following equation (Kakanakova-Georgieva et al., 2007): n s = 1 q A ∫ C   d V , where q, A, C, and V are the elementary charge of electron (1.6E-19 coulomb), contact area (∼4.57E-3 cm⁻²), capacitance, and bias voltage, respectively. For data accuracy, the background capacitance value in C–V curves was removed, as shown in the Supplementary Material. The 2DEG density of Al_0.3Ga_0.7N/i-GaN regrown on a 300-nm-thick i-GaN without the LT-GaN MSLs was 4.27E12 cm⁻² (the black curve in Figure 3). When 300-nm-thick p-GaN with Mg doping concentrations of 1.0E19 and 4.0E19 cm⁻³ was introduced, the 2DEG density abruptly decreased to 2.84E10 and 1.76E10 cm⁻², respectively, as shown in Figure 3C. This result indicates that the 2DEG is damaged by the Mg diffusion into Al_0.3Ga_0.7N/i-GaN. Figure 3D shows that the 2DEG density of Al_0.3Ga_0.7N/i-GaN slightly improved by 5.21E10 cm⁻² when 10-nm-thick CLT-GaN regrown at 750°C was inserted onto p-GaN (Mg: 4.0E19 cm⁻³). As the thickness of CLT-GaN was increased to 100 nm (Figure 3E), the 2DEG density increased noticeably to 1.58E12 and 4.39E11 cm⁻² when the Mg doping concentrations of p-GaN were 1.0E19 and 4.0E19 cm⁻³, respectively. To improve the crystallinity and surface properties of regrown Al_0.3Ga_0.7N/i-GaN thin films, PLT-GaN was also introduced as the MSL (Figure 3F). With the reduction in the growth temperature from 900°C to 750°C for 100-nm-thick PLT-GaN, the 2DEG density of Al_0.3Ga_0.7N/i-GaN regrown on p-GaN (Mg: 4.0E19 cm⁻³) increased from 1.74E11 to 3.13E12 cm⁻². This is due to the increased suppression efficiency against Mg diffusion in accordance with lowering the growth temperature of MSLs, which could be easily inferred from the SIMS results (Figure 2B). The efficacy of PLT-GaN to suppress Mg diffusion also depends on its thickness. As shown in Figures 3G, H, the 2DEG density increased significantly when the thickness of the PLT-GaN was increased from 10 to 100 nm. With the 100-nm-thick PLT-GaN grown at 750°C, the highest 2DEG density of Al_0.3Ga_0.7N/i-GaN regrown on p-GaN (Mg: 1.0E19 cm⁻³) was 4.04E12 cm⁻², which should be related to the crystalline quality of regrown thin films with the PLT growth method. It is worth mentioning that our previous bandgap simulation result showed different electron densities in 2DEG according to the Mg diffusion into the Al_0.3Ga_0.7N/i-GaN regrown on p-GaN (Lee et al., 2022). Additionally, we confirmed that there was no significant difference in C–V results before and after p-GaN activation, indicating that the electron density of 2DEG formed on p-GaN changes more sensitively depending on the degree of Mg diffusion than p-GaN activation (details not shown).

Figure 4 presents the output and transfer characteristics of the fabricated lateral HEMTs regrown on p-GaN with 10- and 100-nm-thick PLT-GaN MSLs. Figure 4A shows the cross-section device schematics and the plan-view optical microscope image. In the case of HEMTs without the MSL, no 2DEG characteristics were observed (not shown). The gate voltage was swept from 0 to −4 in −1 V steps. The threshold voltage (V _th) was defined as the gate voltage (V _G) when the drain current reaches 1 mA/mm at a drain bias of 5 V from the I _D–V _GS curve (Hwang et al., 2013). The on-state resistance is extracted from the gradient of I _D–V _DS curves at V _GS = 0 V and V _DS from 0 to 0.5 V. A maximum drain current density (I _max) increases from 41.5 to 84.5 mA/mm as the thickness of the PLT-GaN MSL increases, as shown in Figures 4B, C. The I _D–V _GS hysteresis was 0.21 and 0.14 V for the 10-nm- and 100-nm-thick MSL, respectively. From the SIMS and C–V analysis, it is shown that increasing the thickness of PLT-GaN MSLs could further suppress the Mg diffusion into the 2DEG. Therefore, I _max can be increased by maintaining a high-electron density in the 2DEG region. The V _th of these HEMTs is −4.12 V, and the on-state resistance is 52 and 31 mΩ∙mm with 10- and 100-nm-thick PLT-GaN, respectively. The I _D–V _GS transfer characteristics show well-behaved curves, as plotted in Figures 4D, E. The average gate leakage (I _G) of HEMTs using 10- and 100-nm-thick PLT-GaNs was similar to ∼1.45E-8 A/mm. The I _D off-current representing the leakage current characteristics related to defects was 5.74E-11 and 5.36E-11 A/mm, respectively, in the case of using 10- and 100-nm-thick PLT-GaNs. The two values do not show a large difference, and when PLT is used, defect density, one of the main issues of LT-GaN growth, shows no sensitivity according to the thickness of PLT-GaN. The subthreshold slopes were also calculated as 99.23 and 83.19 mV/dec for the 10- and 100-nm-thick MSLs, respectively.

We addressed the PLT-GaN and CLT-GaN inserted between Al_0.3Ga_0.7N/i-GaN and p-GaN to suppress Mg out-diffusion. In SIMS depth profiling, PLT-GaN improved the Mg decay rate up to ∼21 nm/decade in regrown Al_0.3Ga_0.7N/i-GaN on p-GaN, presenting a much higher decay rate compared to the case without any suppression layers (∼230 nm/decade). In addition, the C–V characteristic showed a high electron density of 3.13E12 cm⁻² in 2DEG of Al_0.3Ga_0.7N/i-GaN with the PLT-GaN MSL on p-GaN (Mg: 4.0E19 cm⁻³), which is the almost same electron concentration (4.27E12 cm⁻²) as that in 2DEG grown without p-GaN. In addition, we successfully demonstrated the electrical characteristics of the lateral HEMTs regrown on p-GaN with 10- and 100-nm-thick PLT-GaN MSLs. This study provides a pathway to solve a critical bottleneck caused by Mg out-diffusion and enables diverse applications in high-efficiency and high-output power electronics that require GaN-based epi-structures embedding p-GaN layers.