A Physics-Informed Neural Network Model for Body Potential Distribution in MOSFETs Down to 50 K

A potential model using the physics-informed neural network (PINN) is developed for the channel potential profile in the temperature range of 50K to 300K. By parameterizing the PINN, the voltage and spatial dependence can be incorporated into the model from accumulation to the inversion region. The results show that the PINN model achieves high accuracy for calculations of the gate-controlled charge density. Compared to iterative solutions, the method proposed in this paper solves the device's potential more quickly. Further, the model considers the incomplete ionization effects under low temperatures, beyond the capability of the full-depletion approximation. The PINN-based potential model could serve as a basis for cryogenic device modeling.