Unravelling the secrets of the resistance of GaN to strongly ionising radiation

The study investigates why GaN exhibits unusual resistance to strongly ionising radiation, a key issue for radiation-hard electronics in space and high-energy physics. While effects of low-to-medium energy ions in GaN are fairly understood, the high-energy regime involving swift heavy ions (SHIs) and HZE particles remains debated. Conflicting reports exist on whether SHIs amorphise GaN or leave it crystalline but defective. The authors aim to resolve this by combining a two-temperature model with molecular dynamics (TTM-MD) and experiments (TEM, EELS, RBS/C) to uncover the mechanisms—particularly recrystallisation—that govern GaN’s response to SHIs and its damage evolution with fluence.

Prior work shows SHIs deposit energy predominantly via electronic interactions, often producing ionisation spikes, latent tracks, and surface effects in semiconductors. Reports on GaN are contradictory: Sall et al. observed amorphisation in GaN thin films post-SHI irradiation via TEM, whereas Kucheyev et al. reported defective yet crystalline GaN at high SHI fluences via RBS/C, with behavior deviating from Poisson/Gibbons models, suggesting complex formation/annihilation mechanisms. SHI-induced recovery and recrystallisation have been observed in other materials (e.g., SiC), and ionisation-induced annealing can heal pre-existing defects. The literature also highlights challenges in RBS/C interpretation when extended defects and density variations are present, necessitating more advanced simulation tools for accurate spectra prediction.

- Materials: 3-µm thick c-GaN on Al2O3 (commercial Lumilog films), grown by MOCVD. - Irradiations: 185 MeV Au13+ (ε = 33 keV/nm by SRIM) at fluences 1×10^11, 1×10^12, 1×10^13 cm^-2 (room temperature) at ANU HIAF; energy-degraded 90 MeV Xe23+ to 70 and 45 MeV (ε ≈ 19.9 and 14.9 keV/nm) at GANIL with fluence 2×10^12 cm^-2; incidence normal to surface. Note SRIM ε does not include charge-state effects very near surface. - Imaging/analysis: High-resolution TEM/STEM-HAADF (JEOL ARM 200F, 200 kV; STEM inner/outer angles 68/280 mrad) on FIB-prepared plan-view and cross-sectional lamellae; EELS (collection angle 90 mrad, 1 eV/channel) for density mapping; RBS/C at IST-LATR using 2 MeV He+ with 165° detector, aligned along c-axis and random by 5° tilt and azimuthal rotation. - TTM-MD simulations: Two-temperature model with Te-dependent electronic heat capacity, thermal conductivity, and electron-phonon coupling derived from effective mass theory (I-valley) for GaN; electron relaxation time τe = 85 fs as the single free parameter. TTM validity assumed until ~90% of electronic energy is transferred (~80 fs). Lattice energy profile is then deposited into MD. - MD details: PARCAS code with Albe–Nord bond-order potential; simulations run to 155 ps post-impact to room temperature. Cells of 30×30×50 nm^3. Separate bulk (periodic in all directions) and surface simulations (non-periodic normal to trajectory); Berendsen thermostats at boundaries; fast thermostat at bottom of surface cell to avoid artificial surface effects. - TEM-HAADF simulation: Compute local average of Z^2 within 1 nm cutoff at each atom site and map to grayscale to compare with experimental contrast. - EELS simulation: Calculate atomic density in 0.5 Å slices across 30 nm depth; average over windows to emulate experimental resolution and compare with EELS. - RBS/C simulation (RBSADEC): Monte Carlo binary collision code to simulate RBS/C from arbitrary atom coordinates using final MD atoms. Construct large cells by tiling bulk and slicing/stacking surface MD cells to 1200 nm total depth; introduce fluence via Poisson statistics for number of impacts and nearest-neighbor spacing to capture direct/indirect overlap effects; account for depth inhomogeneities and surface region thickness distribution (50–320 nm, with weighting near maximum). No fitting beyond surface morphology inclusion.

- Single-track dynamics (185 MeV Au; ε = 33 keV/nm): An electronic spike melts a cylindrical track; peak lattice temperature up to ~8000 K. The molten radius reaches r1 ≈ 4.4 nm at 13 ps. High-pressure confinement triggers recrystallisation from the perimeter inward, yielding a highly crystalline shell with extended defects and a smaller amorphous core of radius ro ≈ 1.8 nm at room temperature. - Defects and density: Simulations and TEM-HAADF show extended defects—dislocation edges and stacking faults—producing misoriented domains matching experiment. A low-density core surrounded by a high-density shell is observed. Simulated amorphous core density ≈ 5.1 g/cm^3 agrees with DFT predictions and EELS-derived density profiles. - Surface effects: Near the surface, pressure relaxes upward causing sputtering and a hillock, plus formation of voids inside the track. Simulations match cross-sectional TEM in void presence, shape (elongated along the track), size (largest lateral ~3.6 nm simulated vs ~4 nm measured), and density (~0.35 voids/nm simulated; ~0.30–0.35 voids/nm measured near surface). - Overlapping impacts: A subsequent SHI impact substantially recrystallises the prior amorphous core even for indirect overlaps, implying a recrystallisation cross-section radius r ≈ 6.5 nm, larger than the initial melt radius. Void recovery is partial and more effective at depth, causing a depth gradient in void density. - Damage evolution with fluence: RBS/C shows increased aligned yield with fluence but remains well below random even at 1×10^13 cm^-2, indicating no complete amorphisation. RBSADEC spectra based on MD cells reproduce experimental spectra across depth, capturing the influence of extended defects, density gradients, and surface voids (not included in standard RBS/C models). Depth-dependent void distributions, due to overlap-induced recovery of deeper voids, significantly affect aligned spectra. - Lower-energy SHIs (GCR-relevant): For 70 and 45 MeV Xe (ε ≈ 19.9 and 14.9 keV/nm), tracks have smaller maximum radii rt ≈ 2.8 nm and 2.2 nm (at ~6 ps), fewer extended defects, and discontinuous tiny amorphous pockets (r < 1 nm) that recrystallise upon overlap. Voids are extremely shallow (∼18 nm for 70 MeV and ∼5 nm for 45 MeV Xe). RBS/C indicates nearly no damage at 2×10^12 cm^-2; 45 MeV Xe aligned yields are comparable to as-grown. Slight simulation overestimation likely arises from post-ps defect mobility not captured by MD.

The combined TTM-MD simulations and experimental validations (TEM, EELS, RBS/C) demonstrate that strong, pressure-assisted recrystallisation is the dominant mechanism underpinning GaN’s resistance to strongly ionising radiation. Recrystallisation recovers most of the initially melted region and can heal damage from prior ion impacts, including amorphous cores and some voids. This dynamic recovery significantly suppresses damage accumulation with fluence, preventing full amorphisation even under high overlap conditions. Incorporating realistic track morphologies—extended defects, density gradients, voids—into RBS/C simulations is essential for accurate interpretation and aligns simulations with experiment. For space applications, the model suggests that HZE-like energies (e.g., Fe with ε ≈ 14 keV/nm) will induce even milder damage profiles, informing failure rate estimates, dose thresholds, and shielding strategies for GaN-based devices.

This work identifies ion-induced recrystallisation as the key mechanism that confers GaN its unusual resilience to swift heavy ion irradiation. A physics-based TTM-MD framework, coupled with RBSADEC for RBS/C, quantitatively reproduces observed track structures (amorphous cores, extended defects, density gradients, voids) and their evolution with fluence. The approach predicts limited damage accumulation and explains the absence of amorphisation even at high fluences. The model, validated across energies relevant to galactic cosmic radiation, can guide the design, testing, and radiation-hardening of GaN devices and inform the use of SHIs in materials engineering. Future work should address longer time-scale annealing and defect mobility beyond the MD window, refine electronic parameters for TTM under non-equilibrium, and extend predictions to device-level performance under mixed radiation fields.

- TTM uses effective, temperature-dependent electronic parameters (C, K, g) derived from effective mass theory; the true electronic properties for GaN under far-from-equilibrium conditions are not fully known. The electron relaxation time τe (85 fs) is a fitted free parameter. - The TTM is applied only during the initial ~80 fs (90% electronic energy transfer), after which structural dynamics are handled by MD; non-equilibrium effects beyond this approximation may introduce deviations. - MD simulations are limited to ~155 ps, not capturing longer-term defect diffusion/annealing known to occur in GaN (e.g., point defect mobility at ~120 K), which likely explains small discrepancies (slight overestimation in simulated RBS/C yields). - SRIM-based electronic stopping values do not include charge-state effects near the surface, potentially underestimating near-surface energy deposition. - RBSADEC cell construction requires assumptions on overlap spacing (from Poisson/nearest-neighbor statistics) and surface depth distributions; while physically motivated, these introduce model dependence.