The numerical optimization of semiconductor devices requires robust multiphysics models capable of handling strong nonlinear couplings while remaining stable under geometrical and parametric variations [1]. In this work, a two-dimensional drift–diffusion framework for semiconductor junction devices is presented, combining Poisson electrostatics, carrier transport and recombination mechanisms within a finite element formulation tailored for global optimization. Device performance is evaluated by computing complete current–voltage (JV) and power–voltage (PV) characteristics. The optimization problem is formulated by maximizing the electrical output power, ensuring that current density and voltage contributions are consistently embedded in a single objective functional. Global optimization is performed using an advanced hybrid genetic algorithm, which enables an efficient exploration of the design space and avoids convergence towards local optima in the presence of strong nonlinearities [2,3]. Special emphasis is placed on the numerical treatment of surface recombination effects. Surface recombination velocity is incorporated through ultra-thin volumetric layers with effective properties, avoiding boundary-condition-based formulations and preserving compatibility with volume discretizations. Doping profiles are defined using analytical spatial functions, ensuring thickness-independent descriptions and preventing artificial distortions during geometrical updates within the optimization loop. Representative results are presented, illustrating the influence of device thickness, junction positioning, surface recombination and doping profiles on the optimized power output. The proposed multiphysics and optimization framework provides a flexible and physically consistent numerical platform for semiconductor device optimization and constitutes a solid intermediate step towards journal-level publication. REFERENCES [1] Shiba, K., Okada, Y., Sogabe, T., et al., Drift–Diffusion Simulation of Intermediate Band Solar Cells, Journal of Nanomaterials, Article ID 5578627 (2023). [2] M. A. Londe et al., Biased Random-Key Genetic Algorithms: A Review. European Journal of Operational Research 321, 1, 16 (2025). [3] P. Ferrada et al., Optimization of N-PERT Solar Cell under Atacama Desert Solar Spectrum. Nanomaterials, 12, 20, 3554 (2022)