报告题目：ReRAM for Neuromorphic Computing

报告人：高滨 副教授 (清华大学)

邀请人：陈杰智 教授

时  间：2019年8月29日（周四）14:00-16:00

地  点：js33333金沙线路检测N5楼128室

报告摘要:

Resistive switching memory based neural-network processing unit (NPU) is promising for the next generation AI hardware. In this work, an end-to-end simulator for RRAM NPU with an integrated framework from device to algorithm is developed. The complete design of circuit and architecture for RRAM NPU is provided. General purpose neural network algorithms are processed for the study of device-circuit interaction. Device optimization methodology for analog type RRAM device is proposed. A complete hardware system with integration analog RRAM arrays are fabricated. Different deep neural network algorithms are demonstrated on the developed hardware system. The measurement results on the analog RRAM array validates the significant performance improvement of neuromorphic computing based on RRAM.

报告人简介:

高滨，2008年毕业于北京大学物理系，获得物理学专业理学学士学位；2013年获得北京大学信息科学技术学院微电子学与固体电子学专业理学博士学位。2015年加入清华大学微纳电子系，2017年晋升准聘副教授。现主要从事新型存储器制备、表征和模型等方面的研究。已发表论文100余篇，其中31篇发表在微电子顶级会议IEDM、VLSI和ISSCC，SCI他引超过2000次；已获得发明专利36项。