For the intelligent perception nodes at the edge, analog Compute-in-Memory (CIM) is promising for its speed and power benefits, whereas Analog-to-Digital Converters (ADCs) are the bottleneck of performance and efficiency. This brief presents a CIM architecture with resolution shifting SAR ADC, namely RS-CIM. The multiply-accumulate (MAC) operation of signed numbers is completed in the form of charge-sharing, which significantly improves energy efficiency. By truncating and quantifying the dynamic range of the bit line, which represents the result of MAC, the signal margin of the least significant bit (LSB) expands. Meanwhile, a configurable shift operation restores the data range after quantification to compensate for the loss of accuracy caused by low-resolution ADC. Based on TSMC 40nm process, the RS-CIM is designed and verified. The simulation results show that the energy efficiency of RS-CIM can reach 326.5TOPS/W, and end-to-end classification accuracy on CIFAR-10 achieves 83.37%.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2023) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: