Since semiconductor manufacturing consists of many processes, process optimization is complicated due to its complexity and long lead times. Furthermore, in recent years, it has become clear that to improve device performance, the distribution of defect density inside the Si wafer must be appropriately tailored to the device process. Therefore, process co-optimization across wafer and device manufacturers, rather than single-process optimization, is necessary. In this study, the digital twins of each process from the wafer process to the device process were created, and by combining them in a virtual space, total process optimization across companies was implemented. The digital twins are based on the total process simulator that combines a simulator to calculate the evolution of the distribution of defects in Si wafers and a TCAD ion implantation simulator. Then we made the simulation 1,000 times faster by machine learning, and it enabled optimization of process conditions in 8 hours, which would take one year with the simulation. Moreover, the device properties were improved by approximately 70% in the implementation of the obtained optimal solutions on the actual production lines.
Seki et al. (Wed,) studied this question.