Unveiling p-type doping strategies in β-Ga2O3: insights from machine learning and first-principles calculations

Advancements in optoelectronics necessitate effective p-type doping strategies for β-Ga2O3, yet traditional experimental and computational methods have shown significant limitations. This study introduces a machine-learning methodology to screen 88 elements for potential p-type dopants for β-Ga2O3. Utilizing random forest regression, chosen for its robustness from five evaluated models, we identified 7 features influencing Fermi level (EF). Valence electron count and electron affinity showed the strongest correlation with the EF. Mg, Zn, and Cd are the top candidates for p-type doping, with Be, Cu, N, and Hg as secondary contenders. First-principles calculations confirm that these dopants introduce impurity levels just above the valence band maximum, favorably modifying the electronic and optical properties of β-Ga2O3. Notably, Cd doping stands out by achieving the most substantial EF reduction, with lower activation energy than Mg, lower cohesive energy than Cu, and lower formation energy than Zn, while maintaining strong bond stability in the Cd-OII direction. Cd has been underexplored experimentally, presenting a promising avenue for further study. The interstitial H passivates the characteristics of the dopants, reducing the maximum number of photogenerated carriers. This computationally efficient approach accelerates dopant discovery for β-Ga2O3 and holds promise for broader applications in other oxide semiconductors.