닫기
Loading..

전자정보연구정보센터 ICT 융합 전문연구정보의 집대성

EIRIC 세미나

국내외의 전자정보 및 ICT 분야 연구자들이 학술정보 또는 연구와 관련된 교육 콘텐츠를 무료로 접할 수 있는 온라인 세미나입니다.

화상회의시스템을 활용하여 소규모 세미나, 워크숍 등 목적에 따라 다양하게 활용할 수 있습니다.

Webinar

홈 홈 > EIRIC 광장 > EIRIC 세미나
컴퓨터 전자/전기 통신 AI 융합
[신진연구자] Signal Integrity Design, Analysis and Optimization of 3D X-Point Memory using Reinforcement-Learning
  • 일시2021년 5월 13일 (목) 오후 4시
  • 연사손경준 박사과정KAIST Terabyte Interconnection & Package Lab.
  • 약력PDF
개최완료
세미나 개요

Recently, the demands for high bandwidth and high density memory have increased for the high performance server platforms In the memory/storage hierarchy, the memory performance gap between DRAM and NAND flash memory is widening because of the fast growth of semiconductor processing technology To fill the memory performance gap, the storage class memory (SCM) using new materials with high performance and non-volatility was developed. 3D cross-point (X-Point) memory, which is three-dimensionally stacked SCM, is the most promising memory solution. Because of the narrow space and long interconnection lines of the 20 nm-process technology, crosstalk and IR drop could be severe during memory operation. Therefore, it is important to analyze the signal integrity issues of 3D X-Point memory.
With conventional method, we designed, modeled and analyze the 3D X-Point memory and verified the trade-off relationship between crosstalk and IR drop. However, when trivial conditions such as the sizes of technology nodes change, the entire design should be remodeled and verified with repetitive processes. To overcome these issues, we propose a reinforcement-learning model to design an optimal 3D X-Point array structure considering signal integrity issues. The interconnection design problem is modeled to the Markov decision process (MDP). The proposed reinforcement-learning model designs the signal integrity optimized 3D X-Point array structure and we verified the scalability and sensitivity of the proposed model. Using the proposed model, we can easily design an optimal 3D X-Point array structure with a certain size, performance capabilities and specifications based on reward factors and hyperparameters.

댓글(0)