Professor Min-Soo Rhu’s research team has developed an AI accelerator 21 times faster than GPU-accelerated computing. The news has been reported to the leading media in Korea.
The research team developed the world’s first intelligent semiconductor system optimized for accelerating learning algorithms for artificial intelligence (AI) recommendation systems, which has applied the ‘Processing in Memory (PIM)’ technology. Intelligent memory semiconductor technology is one of the key technologies to dominate the global artificial intelligence market. This achievement of the research team is significant in that it implies the commercial viability and success potential of PIM technology.
Ph.D. student Young-Eun Kwon participated as the first author of the paper, and M.S. student Yun-Jae Lee participated as the second author. This paper will be published around February next year at the IEEE International Symposium on High-Performance Computer Architecture (HPCA), one of the best international journals in the computer system architecture field. The title of the paper is ‘Tensor Casting: Co-Designing Algorithm-Architecture for Personalized Recommendation Training’. This research was supported by the Samsung Science & Technology Foundation.