Gallium nitride (GaN) micro- and nano-LEDs promise high efficiency, robustness, and high pixel densities for next- generation displays, yet their internal quantum efficiency (IQE) degrades as device dimensions shrink. A central cause is nonradiative S...
Gallium nitride (GaN) micro- and nano-LEDs promise high efficiency, robustness, and high pixel densities for next- generation displays, yet their internal quantum efficiency (IQE) degrades as device dimensions shrink. A central cause is nonradiative Shockley–Read–Hall (SRH) recombination driven by surface defects formed during top-down fabrication. However, the atomistic origins of the efficiency-deteriorated recombination centers at GaN surfaces, as well as the process mechanisms that create them, have remained unresolved.
This dissertation develops an integrated, multiscale framework that couples hybrid functional density functional theory (DFT) with machine-learning interatomic potentials (MLIPs) and large-scale molecular dynamics (MD) to connect processing, structure, and electronic properties at GaN surfaces. In particular, two complementary research thrusts are pursued. First, the electronic structure and nonradiative capture characteristics of sidewall point defects are quantified on the non-polar m surface using a hybrid functional. Second, process simulation based on a Behler–Parrinello-type neural-network potential (BPNNP), trained and iteratively refined on DFT datasets, is used to perform reactive MD and enhanced sampling simulations of alkaline wet etching in KOH on the ±c, m, a planes under temperature and pressure accelerated conditions. Together, these studies establish concrete, atomistically grounded links between etch chemistry, resultant surface terminations and defects, and their impact on SRH recombination.
On the electronic side, nitrogen vacancies (VN) on the m plane sidewall are identified as the dominant SRH centers. Unlike their bulk counterparts, VN at the surface stabilizes a single deep charge- transition level near midgap. In addition, configuration coordinate analysis reveals strong electron–phonon coupling with a small electron-capture barrier and a moderate hole-capture barrier; the resulting hole capture coefficient at typical LED operating temperatures is on the order of 10−7cm3/s , large enough for VN to dominate SRH recombination. These results rationalize the observed size-dependent efficiency degradation in micro-LEDs and point to sidewall VN suppression or passivation as improving device performance.
On the processing side, the MLIP-MD simulations reproduce experimentally consistent morphological evolution during alkaline wet etching and quantify the reaction steps that control etchability across facets. Ga dissolution emerges as the kinetic bottleneck and proceeds via OH– adsorption that weakens and breaks Ga−N bonds, and nearby N centers are proton passivated to form soluble NH3. Facet -dependent free energy surfaces explain the widely observed anisotropy: the −c surface etches readily through moderate barriers; the +c surface exhibits a substantially higher initial barrier associated with constrained Ga displacement and repulsive pre-adsorbed hydroxyls; and the nonpolar m and a surfaces display intermediate barriers consistent with their observed lateral etch -pit propagation. Beyond rates and morphologies, the simulations uncover Ga−O−Ga bridge species that form when oxygen substitutes nitrogen or transiently occupies interstitial positions during etching. This implies that VN can be passivated during the alkaline wet etching process. In addition, substitutional O bridges are stable against dissolution until neighboring atoms are removed and can therefore survive post-etch, creating electrically active surface motifs that are plausible nonradiative centers or precursors to such centers.
Collectively, these results yield an end-to-end, atomistic picture of how surface point defects and process-induced oxygen bridges, in particular, arise during wet etching and how they control SRH recombination. The findings suggest practical guidelines for fabrication: (i) minimize the creation and retention of sidewall VN via nitrogen- rich anneals or nitride passivation layers; (ii) alkaline wet etching can eliminate VN by substituting the site with oxygen or hydroxide ion, (iii) optimize alkaline etch conditions to discourage formation and survival of Ga−O−Ga bridges. Methodologically, the work demonstrates how hybrid functional accuracy and accelerated MLIP reactive dynamics can be combined to quantify both defect physics and processing kinetics. The framework is extensible to alternative chemistries (e.g., TMAH), doped or alloyed nitride surfaces, and more advanced long-range-aware MLIPs, offering a route to predictive, processing-informed design of high-efficiency GaN micro-LED.