Abstract:In this paper, two methodologies are used to speed up the maximization of the breakdown volt-age (BV) of a vertical GaN diode that has a theoretical maximum BV of ~2100V. Firstly, we demonstrated a 5X faster accurate simulation method in Technology Computer-Aided-Design (TCAD). This allows us to find 50% more numbers of high BV (>1400V) designs at a given simulation time. Secondly, a machine learning (ML) model is developed using TCAD-generated data and used as a surrogate model for differential evolution optimization. It can inversely design an out-of-the-training-range structure with BV as high as 1887V (89% of the ideal case) compared to ~1100V designed with human domain expertise.

What problem does this paper attempt to address?

The problem this paper attempts to address is improving the breakdown voltage (BV) of vertical Gallium Nitride (GaN) diodes. Specifically, the authors propose two methods to accelerate the search process for high BV designs: 1. **Fast TCAD Simulation Method**: - Developed an accurate TCAD simulation method that is 5 times faster than traditional methods, allowing for the discovery of more high BV designs (over 1400V) in the same amount of time. - This method achieves significant acceleration by simplifying the simulation model, such as removing the avalanche breakdown model and monitoring the peak electric field. 2. **Machine Learning (ML) Surrogate Model**: - Utilized data generated by TCAD to train a machine learning model, which serves as a surrogate model for differential evolution optimization. - Through inverse design, this method can find structures beyond the training range, achieving up to 1887V BV, which is 89% of the ideal value. These two methods together improve the efficiency and accuracy of the search for high BV designs, thereby providing an effective means to enhance the performance of vertical GaN diodes in high voltage, high power applications.

Vertical GaN Diode BV Maximization through Rapid TCAD Simulation and ML-enabled Surrogate Model