講演情報
[14p-P10-12]Research on Differential Type STT-MRAM Cell for Accelerator Based on Low-Power Digital CiM Architecture
〇Yongcheng Wang1,2, Tao Li1,3, Tetsuo Endoh1,3,2 (1.Tohoku Univ., 2.TU GP-Spin, 3.TU CIES)
キーワード：
hardware design、STT-MRAM、Computing-in-Memory
Recent artificial intelligence (AI) technologies make mobile devices more and more intelligent. However, the tremendous amount of data and frequent data transfer between memories and processing elements during processing bring high power consumption to AI chips. Using the Computing in Memory (CiM) architecture to design AI chips is an effective way to reduce power consumption. Spin-Transfer Torque Magnetoresistive Random Access Memory (STT-MRAM) is one of the most promising emerging verified technologies because of its non-volatile, high-speed, high-density, and high-reliability characteristics compared to conventional non-volatile memories. However, most designs of STT-MRAM-based CiM architecture are suffering from the limited accuracy and access speed due to the limited TMR. The goal of this research is to design a differential type STT-MRAM-based CiM cell that can avoid these problems for AI chips.