Invited Paper: A Ping-Pong Analog Delta Generator for Delt-Sigma Computing-In-Memory SRAM Macro for Edge AI Processing

The delta-sigma computing in memory ($\triangle$ CIM) is a promising approach for edge computation, which is beneficial for energy efficiency. Prior research utilized a digital-delta-digitalsigma or digital-delta-analog-sigma computing-in-memory scheme to mitigate this redundant processing, thereby reducing power consumption. This article introduces a dynamic analog ping-pong delta generation to construct an analog-delta-analog-sigma CIM macro. This approach employs a switched capacitor and dynamic amplifier to boost the throughput while maintaining energy efficiency. The simulation demonstrates that our proposed technique can improve throughput and energy efficiency by $\mathbf{2 4 \%}$ and 40%, respectively. The enhanced energy efficiency and computational speed are significant for edge computation devices.