Skip to main content
Video s3
    Details
    Presenter(s)
    Kanglin Xiao Headshot
    Display Name
    Kanglin Xiao
    Affiliation
    Affiliation
    Peking University Shenzhen Graduate School
    Country
    Country
    China
    Author(s)
    Display Name
    Kanglin Xiao
    Affiliation
    Affiliation
    Peking University Shenzhen Graduate School
    Display Name
    Xiaoxin Cui
    Affiliation
    Affiliation
    Peking University
    Display Name
    Xin Qiao
    Affiliation
    Affiliation
    Peking University
    Display Name
    Nanbing Pan
    Affiliation
    Affiliation
    Peking University
    Display Name
    Xin'An Wang
    Affiliation
    Affiliation
    Peking University Shenzhen Graduate School
    Display Name
    Yuan Wang
    Affiliation
    Affiliation
    Peking University
    Abstract

    We present a fully capacitive-coupling-based SRAM CIM macro aimed at improving the energy efficiency and throughput of edge devices. The proposed architecture is built around a customized 9T1C bit-cell in charge-domain computation in 28nm technology. The proposed design supports 8192 4b × 4b MAC operations simultaneously. 4-bit input is generated by DAC, while 4b weight is achieved by a hierarchical capacity attenuator array. To minimize the expensive AD conversion, an input sparsity sensing scheme is proposed, allowing to skip redundant comparators. The proposed design achieves energy efficiency of 666 TOPS/W and throughput of 4096 GOPS.

    Slides
    • A Computing-in-Memory SRAM Macro Based on Fully-Capacitive-Coupling with Hierarchical Capacity Attenuator for 4-b MAC Operation (application/pdf)