Acceleration of Three-Dimensional Device Simulation with the 3D Convolutional Neural Network

We propose to use a 3D convolutional neural network to accelerate three-dimensional device simulation by generating an electrostatic potential profile. In the training phase, the deep neural network is trained with the simulation results for various 3D MOSFETs in a supervised manner. The generated potential profile is used as an initial guess at a non-equilibrium condition, while carrier densities are estimated by the frozen field simulation. By numerical examples for three-dimensional MOSFETs, we show that the proposed method significantly reduces the number of the Newton iterations.