Professor Minkyu Je from our department was recognized for his contributions to the advancement of the AI semiconductor industry through the development of innovative neural network computation circuits and system technologies based on embedded Magnetoresistive Random Access Memory (eMRAM) Process-In-Memory (PIM) technology. He was awarded a commendation from the Ministry of Science and ICT at the Future Technology Conference on AI Semiconductors held on December 20.
Since April 1, 2022, Professor Je’s research team at the Integrated Memory & Processor Architecture for Compact Systems (IMPACT) Lab has been working on the “Development of Core Technologies for High-Efficiency AI Semiconductors Utilizing eMRAM-based PIM Technology” project, supported by the Ministry of Science and ICT and supervised by the Institute for Information & communication Technology Planning & evaluation. Through this project, the team has achieved remarkable results. These include the development of neural network computation circuit technology that consumes low power, allows for adjustable computational precision, and improves memory utilization efficiency by more than twofold compared to existing technologies. They also developed a weight representation method and supporting circuit architecture that reduce computational errors caused by analog variations during in-memory analog computation, while simultaneously reducing power consumption. Furthermore, they introduced an efficient analog neural network accelerator technology capable of processing the entire neural network computation within a PIM-based System-on-Chip (SoC). This technology is robust against computational errors due to analog variations and can be applied to large-scale neural network models without retraining.
