Neural Drift-Diffusion Model Based on Operator Learning in Fourier Space
Kyeyeop Kim, Sanghoon Myung, Yunji Choi, Gijae Kang, Kyungmi Yeom, Songyi Han, Jaehoon Jeong, Dae Sin Kim · 2024
We present a novel neural drift-diffusion model that aims to emulate the process of solving the drift-diffusion equation. By utilizing Fourier neural operators, the model learns the underlying physics of the drift-diffusion equation and accurately predicts device characteristics. Our approach stands out from previous studies by representing the learned principle as a generalized Green's function in Fourier space, which maps between the solutions of the initial and the next biases. This capability allows our model to provide precise solutions even with unseen inputs in terms of doping profiles or biases.