Trap characterization of high-growth-rate laser-assisted MOCVD GaN

Utilizing a CO₂ laser to more efficiently decompose the ammonia precursor in metal organic chemical vapor deposition (MOCVD) growth has demonstrated high growth rates and promises lower manufacturing costs for vertical GaN transistors and diodes for multi-kV applications.^1–5 Generally, vertical diode devices targeting a breakdown voltage greater than 4 kV require 10¹⁵–10¹⁶ cm⁻³ net n-type doping with thick drift layers to support the high electric fields.^6–9 A high growth rate is desired to achieve thick drift layers, but typical MOCVD growth rates are only 2–3 μm/h.^10–12 Laser-assisted MOCVD (LA-MOCVD) growth has achieved rates up to 84 μm/h and continues to improve as the optics and growth conditions are optimized.^4,5 Controllable low n-type net doping (N_D–N_A) requires n-type doping (N_D) and compensating trap concentrations (N_A) as low as possible, so the net doping is easily targeted and less sensitive to variations in N_D or N_A. Additionally, compensating acceptors and traps can change occupancy as the Fermi level is swept above and below the trap levels due to biasing, which leads to time-dependent transistor instabilities including dynamic on-resistance, threshold voltage instability, and current collapse.^13–16 Hence, identifying the traps in these materials, quantifying the trap concentrations, and identifying mechanisms to mitigate their formation or incorporation is critical to enable low-doped drift layers for vertical GaN transistors.

Trap characterization and growth studies of LA-MOCVD-grown GaN are very limited. Notably, Golgir et al. achieved a 25.8 μm/h growth rate and measured 5.9 × 10⁸ cm⁻² dislocation density and 369 cm² V⁻¹ s⁻¹ mobility.⁵ Zhang et al. recently achieved GaN growth rates up to 5.2 μm/h, characterized impurity incorporation, and achieved a 604 cm² V⁻¹ s⁻¹ mobility under the highest growth rate.¹ 

A higher growth rate with LA-MOCVD growth is possible with both pyrolytic and photolytic approaches.² In the pyrolytic approach, the laser is used to heat the substrate surface and aid the chemical reaction locally or globally. On the other hand, the photolytic approach uses the laser to crack the precursors, specifically ammonia, in the gas stream to provide higher concentrations of elemental nitrogen to the growth surface.¹⁷ Using optical emission spectroscopy, it was shown that higher concentrations of NH₂ and other NH₃-related radicals are observed using a 9.219 μm CO₂ laser beam in air, indicating enhanced cracking efficiency.⁴ The photolytic approach was chosen for the LA-MOCVD growth in this work as it allows higher growth rates while maintaining an independent control of growth conditions such as substrate temperature, growth pressure, etc.

In this work, an LA-MOCVD GaN sample utilizing the photolytic approach is compared with standard MOCVD samples with high and low growth rates to understand the effects of growth rate separately from that of the laser-assisted growth on the incorporation of deep levels. Trap energies and concentrations in these samples are characterized with deep-level transient and optical spectroscopy (DLTS/DLOS). Proton irradiation combined with DLTS characterization are employed to discern the intrinsic or extrinsic nature of the defects.

Three samples were grown by MOCVD and LA-MOCVD for this study. The first two were grown conventionally without laser assistance and were labeled low-growth-rate (LGR) and high-growth-rate (HGR) with growth rates of 2.0 and 5.2 μm/h, respectively. A third sample was grown with the laser at a growth rate of 2.8 μm/h and was labeled laser-assisted (LA). Growths for all three samples were done in a rotating-disk MOCVD reactor (Agnitron Technology Inc., Agilis MOCVD R&D System) at a growth pressure of 200 Torr and a substrate temperature of ∼960 °C, monitored by a thermocouple. The TMGa flow for the LGR sample was 68.6 μmol/min, and a higher TMGa flow of 203.9 μmol/min was applied for HGR and LA samples with an NH₃ flow rate of 4 slm for all three samples. The silane dopant flow rate was controlled at 8.9, 17.7, and 14.7 pmol/min for LGR, LA, and HGR samples, respectively, to keep the doping concentration consistent among all samples. The LA sample was grown with a PRC 1500 W tunable CO₂ laser system tuned to a wavelength of 9.219 μm and a power of 200 W for optimal ammonia cracking. The laser beam's shape was round with a diameter of 14 mm and was configured to be parallel to the substrate and wafer carrier to crack ammonia in the gas flow without impinging on the growth surface. The LA sample showed a significant reduction in growth rate compared to the HGR sample despite their same precursor flow rates, which resulted from significant gas phase parasitic reactions due to the round laser beam creating N radicals far from the growth surface. Since these growths, the laser beam has been shaped by focal lenses to have less vertical width so gas phase reactions would be suppressed,¹⁸ and defect studies of the latest growth method are ongoing. All three samples were grown on 2 in. unintentionally doped (UID) GaN-on-sapphire templates with similar growth structures, as shown in Fig. 1. A heavily doped n-type ([Si] ∼ 2 × 10¹⁸ cm⁻³) layer was grown on the substrate which enabled efficient lateral conductance and functioned as a base for good Ohmic contact. A lightly doped n-type ([Si] ∼ high-10¹⁶ cm⁻³) test layer was probed during the defect spectroscopy measurements with a thickness of 0.5 μm in the LGR sample and 2 μm in the HGR and LA samples. Since the defect spectroscopy measurements only measured the top ∼0.13 μm of the test layer, negligible influence of the test layer thickness has been observed in similar samples. For the contacts, 80 Å of Ni was deposited on the lightly doped test layer to form a 290 × 290 μm² semitransparent Schottky contact to enable DLOS measurements. For the Ohmic contact, reactive ion etching with Cl chemistry was used to etch down to the n⁺ GaN layer where Ti/Al/Ni/Au (100/2000/200/2000 Å) were deposited.

DLTS and DLOS were conducted to characterize traps throughout the GaN bandgap. DLTS was used to detect traps within ∼1 eV of the conduction band, and DLOS was used to detect deeper traps. For DLTS, the fill pulse was 0 V for 10 ms, and trap emission was measured at −0.5 V. DLTS was analyzed using the double boxcar method, where the majority traps cause positive peaks in a DLTS spectrum. For DLOS, the fill pulse was 0 V for 10 s, and then the device was biased to −0.5 V for the measurement phase with the shutter open for 300 s after a 100 s delay to allow thermal transients to subside. More details of the experimental apparatus and measurements are available in Refs. 19 and 20. Secondary ion mass spectrometry (SIMS) was conducted to detect extrinsic impurity density and aid the trap spectroscopy measurements.

To further understand the possible physical sources of the observed levels, proton irradiation was utilized to create native point defects. The LA sample was subject to 1.8 MeV proton irradiation with a 2 × 10¹³ cm⁻² fluence at Sandia National Laboratories. DLTS was conducted before and after the proton irradiation to explore the effect of radiation on trap concentrations and gain insights as to whether the as-grown traps are intrinsic or extrinsic.

Figure 2(a) shows three peaks in all samples. The high-temperature peak is asymmetric with a small shoulder near 350 K in the 4 s⁻¹ rate window indicating the existence of a fourth level. The DLTS peaks have some overlap, so peak fitting was used for each DLTS rate window in each sample to extract more precise trap energies and concentrations, where an example is shown in Fig. 2(c). From the fitting results, Arrhenius plot is plotted in Fig. 2(b) where the extracted trap energies are E_C-0.25 eV, E_C-0.57 eV, E_C-0.72 eV, and E_C-0.9 eV. The clustering of traps on the Arrhenius plot suggest they are the same among all growth conditions. Therefore, the increased growth rate and the presence of the laser-assisted growth did not result in the formation of new or different traps. Additionally, the traps in the LA-MOCVD sample are all consistent with previously reported traps plotted in Fig. 2(b). The E_C-0.25 eV and E_C-0.57 eV traps are nearly ubiquitous in GaN and have been attributed to a V_N-related defect and Fe_Ga, respectively.^22,23 Previous studies have shown that Fe and Fe-related trap incorporation can be minimized through improved wafer carrier cleaning and sample mounting, which has been used here to minimize Fe incorporation.²⁴ The other two traps at E_C-0.72 eV and E_C-0.9 eV have been previously observed, but the physical sources of these traps need to be better understood.^25–27 

Here, the trap concentrations are proportional to the DLTS peak height. Still, more accurate trap concentration extraction requires accounting for the volume where the traps are modulated, which differs from the depletion region volume and is referred to as Lambda effect correction.²¹ Depending on the trap level, doping, and measurement biases, the Lambda effect multiplier usually ranges from 1 to 10, and the concentration of deeper traps is more underestimated without this correction. The Lambda-corrected concentration of each trap is listed in Table I. The E_C-0.25 eV and E_C-0.57 eV traps had relatively low concentrations in the mid-10¹³ to low-10¹⁴ cm⁻³ range and varied less than 50% among all three samples. This indicated that the growth rate and laser-assisted growth had a minor impact on the incorporation of these traps. The E_C-0.9 eV trap concentration in the LA sample was approximately 2× higher than the LGR sample but approximately 2× lower than the HGR sample. This trap may be more sensitive to certain variations in growth conditions. Overall, the trap concentrations, excluding the E_C-0.9 eV trap, range from 100× lower to ∼10× lower than the necessary doping for vertical GaN devices (∼1 × 10¹⁶ cm⁻³), indicating these traps will have minimal impact on net doping/compensation and likely minimal impact on transistor stability. Because the E_C-0.9 eV trap concentration is relatively high, proton irradiation experiments were conducted to help elucidate its physical source.

DLTS measurements were conducted on the LA sample before and after 1.8 MeV proton irradiation with a fluence of 2 × 10¹³ cm⁻². Figure 2(d) shows the post-irradiation DLTS results where trap energies and concentrations were extracted by fitting. Radiation did not lead to any changes in trap energies as expected, and the trap concentrations were calculated again using the Lambda effect correction, which are summarized in Table II. Since the impact of irradiation on the E_C-0.25 eV trap has already been investigated,²⁸ the DLTS scan range was reduced starting at 200 K. From Table II, it is clear the E_C-0.57 eV, E_C-0.72 eV, and E_C-0.9 eV trap concentrations increased by 1 × 10¹⁴, 3.2 × 10¹⁴, and 8.5 × 10¹⁴ cm⁻³, respectively. As the E_C-0.57 eV level is likely an Fe_Ga defect, it is not expected to have radiation dependence though similar dependence has previously been observed.²⁸ Additionally, the E_C-0.57 eV trap concentration was also observed in transistors to increase after simultaneous temperature and bias stressing.^29,30 Others have suggested that Fe_Ga-H defects exist in the as-grown material, and that H can be disassociated by heat and electric fields.^31,32 This suggests local heating during the irradiation, and, likely to a lesser effect, knock-on displacement of hydrogens from the defects result in additional Fe_Ga defects and higher measured E_C-0.57 eV trap concentration. On the other hand, the increase in the concentrations of E_C-0.72 eV and E_C-0.9 eV traps is more significant. Their radiation sensitivity could indicate they are related to simple native point defects. Density functional theory (DFT) suggested that Ga_I, N_I, Ga_N, and N_Ga point defects would form trap levels near E_C-0.72 eV and E_C-0.9 eV.²⁶ Moreover, optical detection of magnetic resonance (ODMR) measurements detected an E_C-0.9 eV level and was associated with Ga_I,³³ which is consistent with the E_C-0.9 eV trap concentration increasing with proton irradiation. Though MOCVD growth will not result in theoretical equilibrium trap concentrations, these interstitial and antisite defects have high predicted formation energies (up to 9 eV), so there remains whether these are the correct potential sources.²⁶ Nonetheless, the strong radiation response indicates both the E_C-0.72 eV and E_C-0.90 eV traps are most likely native point defects.

To explore beyond the ∼1 eV limitation of DLTS, DLOS measurements were conducted. From Fig. 3(a), the steady-state photo-capacitance (SSPC) shows three onsets in each sample at similar energies, indicating that the same traps exist in all three samples. These traps are all commonly observed in n-type GaN. The E_C-1.35 eV trap has previously been attributed to C_I.¹⁹ There are two candidates for the 3.28 eV onset, namely E_C-3.22 eV and E_C-3.28 eV levels. The E_C-3.22 eV trap has been attributed to Mg_Ga.³⁴ The E_C-3.28 eV trap features a negligible Franck–Condon energy (d_FC), and its concentration is correlated with carbon concentration, which suggested it was due to C_N.¹⁹ In our case, the E_C-3.28 eV trap has a d_FC of 0.03 eV, so it is likely related to carbon. The E_C-2.6 eV trap also has two possible sources, which differ in d_FC. The carbon-related E_C-2.6 eV trap has a large d_FC (∼0.4 eV), while the V_Ga-related E_C-2.6 eV trap has a low d_FC of ∼0.1 eV.^19,25 Using Pässler's optical cross section model and fitting the extracted DLOS optical cross section in Fig. 3(b),³⁵ the extracted d_FC is 0.1 eV. Thus, the E_C-2.6 eV level is likely primarily a V_Ga-related trap that is a compensating center.²⁸ 

The trap concentrations are summarized in Table I. In the LGR sample, these trap concentrations are comparable to the best-reported trap concentrations.^36–40 In the LA sample, highest concentrations are observed from the E_C-0.9 eV, E_C-1.35 eV, E_C-2.6 eV, and E_C-3.28 eV traps. Compared with the LGR sample, the E_C-2.6 eV and E_C-3.28 eV trap concentrations both increased ∼2× to ∼2 × 10¹⁵ cm⁻³, and the E_C-0.9 eV trap increased ∼2× to 8.6 × 10¹⁴ cm⁻³. These trap concentrations in the LA sample are lower than the HGR sample suggesting that the laser-assisted growth may help reduce the trap concentrations, but the HGR sample also has a significantly higher growth rate. Thus, it is difficult to separate the effects of growth rate and laser for these traps, especially for the carbon-related traps, since the carbon concentration has been shown to depend on the growth rate, laser power, and laser beam shape.¹⁸ SIMS was conducted on similar samples under the same growth conditions as these three to determine the carbon incorporation dependence. SIMS indicates that the atomic carbon concentrations are 7 × 10¹⁵, 1.9 × 10¹⁶, and 2.6 × 10¹⁶ cm⁻³ for the LGR, LA, and HGR growth conditions, respectively, which is consistent with the DLOS trends. These numbers are higher than the carbon-related trap concentrations detected by DLOS, which is reasonable because carbon forms multiple levels and DLOS can underestimate the trap concentration when the trap energy is greater than half the bandgap.^41,42 Overall, E_C-1.35 eV, E_C-2.6 eV, and E_C-3.28 eV traps in the LA sample are comparable to the targeted doping (∼10¹⁶ cm⁻³), so work is needed to reduce these concentrations to allow more controllable doping through reduced carbon and gallium vacancy incorporation. Recently, it has been demonstrated that reduced gas phase reactions and lower carbon incorporation are possible through improved laser beam shaping, which enables both higher growth rates and suppressed carbon incorporation, so LA-MOCVD is continuing to improve.¹⁸ 

In summary, we characterized traps in laser-assisted MOCVD-grown GaN and compared them with conventional MOCVD-grown GaN. No new traps were introduced compared with the baseline MOCVD samples. Trap concentrations were higher in the laser-assisted sample primarily due to a gallium vacancy-related defect (E_C-2.6 eV), carbon defects (E_C-1.35 eV and E_C-3.28 eV), and a native point defect (E_C-0.9 eV) with trap concentrations up to 2 × 10¹⁵ cm⁻³. However, the laser-assisted sample's total trap concentration is only ∼2× higher than the baseline sample, which was comparable with the best previously reported trap concentrations. Proton irradiation and previous ODMR and DFT studies indicate that the E_C-0.9 eV trap is a native point defect likely related to Ga_I. Laser-assisted MOCVD growth is promising for high-growth-rate GaN applications such as high-voltage vertical GaN devices, but trap concentrations still need to be reduced for ideal device behavior.

The information, data, or work presented herein were funded in part by the Advanced Research Projects Agency-Energy (ARPA-E), U.S. Department of Energy (DOE), under Award No. DE-AR0001036, and the U.S. Department of Energy's Office of Energy Efficiency and Renewable Energy (EERE) under the Advanced Manufacturing Office, FY18/FY19 Lab Call, and the Office of Naval Research under Award No. N00014-20-1-2663. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof. The authors also acknowledge the assistance of Ed Bielejec and Andrew Armstrong for the proton irradiation of the sample. Proton beam implantation was performed at Ion Beam Laboratory (IBL) at Sandia National Laboratories. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology and Engineering Solutions of Sandia LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy's National Nuclear Security Administration under Contract No. DE-NA0003525. This paper describes objective technical results and analysis. Any subjective views or opinions that might be expressed in the paper do not necessarily represent the views of the DOE or the U.S. government.