Exploring Indium Distribution and Phase Stability in Epitaxially Grown In_xGa_1−xN with Machine Learning Potential

There is a growing need for high-resolution displays that can support virtual and augmented reality (VR/AR) effects. Among various display technologies, microLEDs based on In<i>_x</i>Ga₁_−<i>x</i>N/GaN have emerged as promising technology owing to various appealing characteristics such as high brightness, high-energy efficiency, short response time, and good durability. In addition, unlike LCD displays, microLEDs can show broader variation in color and brightness from one pixel to another because they include individual light emitters and do not require a backlight. Despite many advantages, microLEDs still have several technical problems that should be resolved for actual realization. In particular, the internal quantum efficiency (IQE) decreases significantly below 60% with increasing indium concentration, hindering the development of green and red LEDs. Compositional fluctuations and structural inhomogeneities in In<i>_x</i>Ga₁_−<i>x</i>N quantum wells have been suspected to be the cause of this efficiency reduction. However, the solubility of InN to GaN and cation distribution in In<i>_x</i>Ga₁_−<i>x</i>N quantum wells have not been satisfactorily elucidated so far. Moreover, the impact of strain and temperature, which are important process conditions, on the material quality of In<i>_x</i>Ga₁_−<i>x</i>N remains unverified.<br/>In this presentation, we investigate the phase stability and spatial cation distribution for epitaxially grown In<i>_x</i>Ga₁_−<i>x</i>N via neural network potential (NNP), which drastically accelerates molecular simulations compared to authoritative density functional theory calculations without losing accuracy. To this end, we develop a Behler-Parrinello type NNP using SIMPLE-NN package [1] for In<i>_x</i>Ga₁_−<i>x</i>N and perform semi-grand canonical ensemble Monte Carlo simulations using thousands-atom supercells to determine equilibrium phases for various compositions and temperatures. Our results show that strain-free In<i>_x</i>Ga₁_−<i>x</i>N has a wide miscibility gap even at high temperatures over 1000 K. On the other hand, compressive strains, applied to epitaxial In<i>_x</i>Ga₁_−<i>x</i>N thin films in LEDs due to the lattice mismatch between In<i>_x</i>Ga₁_−<i>x</i>N and the GaN substrate, make InN-GaN mixing thermodynamically favorable over a wide range of <i>x</i>. In epitaxial In<i>_x</i>Ga₁_−<i>x</i>N, indium atoms tend to be aligned along the <i>c</i>-axis of the wurtzite structure near <i>x</i>=1/3 at low temperatures (&lt; 350 K), as this arrangement is advantageous in terms of internal energy. However, the ordering gradually disappears with an increase of temperature to increase configurational entropy. This implies that In<i>_x</i>Ga₁_−<i>x</i>N, which is equilibrated sufficiently during the growth process using chemical vapor depositions (<i>T</i>~1000 K), can exhibit uniform cation distribution. Our simulations also reveal that In<i>_x</i>Ga₁_−<i>x</i>N phases can still form despite slight relaxation of the epitaxial strain, which may be too severe to avoid the creation of threading dislocations. This fact offers useful insights for strain engineering to achieve high-performance In<i>_x</i>Ga₁_−<i>x</i>N microLEDs.<br/><br/>[1] K. Lee, D. Yoo, W. Jeong, and S. Han, "SIMPLE-NN: An efficient package for training and executing neural-network interatomic potentials", Comp. Phys. Comm. <b>242</b>, 95 (2019)