Hon Hai Precision Industry Co., Ltd. (TWSE:2317), the world's largest technology manufacturer and service provider, has achieved another breakthrough at its research institute! The Semiconductor Research Institute, in collaboration with the Artificial Intelligence Research Institute and leveraging the superior manufacturing capabilities of Hon Hai Semiconductor, has successfully combined AI learning models with reinforcement learning technology, significantly accelerating the R&D process of silicon carbide power semiconductors and achieving remarkable results. These research findings have been published in the internationally renowned journal *IEEE Open Journal of Power Electronics (OJPEL)*, with an impact factor exceeding 5.8, demonstrating its high academic value in the field of power electronics.
In this study, Foxconn Research Institute employed policy optimization methods from reinforcement learning. Through the Proximal Policy Optimization (PPO) algorithm in policy gradient technology and the Actor-Critic (A2C) architecture combining policy and value functions, they explored and optimized the process parameters and component design of silicon carbide materials to improve performance. Unlike traditional methods that predict results using multiple parameter values, this approach uses AI for reverse prediction. Given target values, the corresponding design parameters are directly identified. In practical applications, this reduces the number of trial and error attempts by designers, improving efficiency.
This technology can not only simulate and adjust complex process parameters, but also significantly shorten device development time and reduce R&D costs, injecting new momentum into silicon carbide technology in the power semiconductor field. For example, in the research on guard rings for high-voltage, high-power silicon carbide devices, the research team conducted process simulation and device characteristic simulation of key parameters of the guard ring (as shown in Figure 1), and input the results into an AI model, successfully establishing an AI model of the guard ring. This model can provide parameter feedback based on the required device characteristics, further improving the performance and process efficiency of silicon carbide devices through accurate data analysis and prediction, and finally verifying it through actual processes (as shown in Figure 2). In addition to "design optimization," this research result can be extended to "process improvement" and "fault diagnosis" in the future, expanding its application scope.
(Figure 1: Process simulation of the guard ring)
(Figure 2: Cross-sectional view of the optimized protective ring)
Silicon carbide (SiC) power semiconductors, due to their ultra-wide bandgap and resistance to high temperatures and pressures, have become key materials in high-power applications such as new energy electric vehicles, smart grids, and aerospace electronic systems. Through reinforcement learning optimization, Foxconn Research Institute has been able to accurately predict the performance of SiC devices under different voltages and temperatures, further improving their stability and reliability. This technological breakthrough not only drives the development of SiC power semiconductors but will also accelerate their application in high-end markets.
Looking ahead, Foxconn Research Institute will continue to dedicate itself to the development of cutting-edge technologies, deeply integrating AI technology with semiconductor R&D to create more application possibilities for high-power components and further enhance Taiwan's competitiveness in the global semiconductor field.
