Physics-Informed Multi-Parameter Optimization of Oxide-Based Memristor Characteristics Using Genetic Algorithms

We present a multiobjective, physics-informed optimization framework for oxide-based memristive devices that jointly minimizes forming voltage (V _f) and maximizes memory contrast (HRS/LRS). The proposed approach integrates COMSOL Multiphysics with MATLAB through a genetic algorithm (GA), enabling self-consistent electrothermal simulations and systematic exploration of a high-dimensional design space encompassing geometrical, electrical, thermal, and operational parameters. By optimizing 13 key variables, the GA efficiently identifies optimal trade-offs between switching efficiency and resistance ratio. The optimization results demonstrate a substantial enhancement in device performance, with the high-resistance-state to low-resistance-state ratio (HRS/LRS) improved by up to 21× relative to the reference device, while alternative optimized configurations achieve a reduction in V _f down to 0.7× of the reference value. A parameter-importance analysis based on a random forest model highlights oxide thickness and effective electrical and thermal conductivities as influential parameters. Importantly, the optimization further reveals nontrivial multiparameter trade-offs, demonstrating that reductions in V _f and enhancements in HRS/LRS can be simultaneously achieved through coordinated tuning of geometrical, material, and pulse-related parameters beyond simple intuitive design rules. Overall, the MATLAB-COMSOL co-optimization strategy provides a scalable and generalizable computational design methodology for next-generation low-power, high-contrast RRAM devices, and offers quantitative insights into the coupled physical mechanisms underlying filament formation and rupture.