Circuit design cuts memory compute comparisons to save energy
Circuit-Architecture-Training Co-Design with Regenerative-SA Similarity Sensing for Aggressive SAR Skipping in Analog Compute-in-Memory
Hardware Architecture
Summary
Computers that perform many calculations inside their memory often spend time comparing similar values, which wastes energy. The authors found a way to detect when values are similar using special circuits, so the computer can skip unnecessary steps. They trained the system to handle minor errors that happen in these circuits. This approach saves nearly half of the comparison operations, reducing energy use while keeping accuracy almost the same. The work was tested on known computer vision models to show practical benefits.
What this means in practice
- •For hardware designers: Build energy-efficient analog compute-in-memory circuits by integrating similarity sensing to skip comparisons and reduce power.
- •For embedded systems engineers: Develop low-power neural network accelerators by incorporating regenerative sense-amplifier detection to improve energy efficiency.
Authors
Yufei Liu, Shuang Liu, Junjie Wang
Abstract
This work presents a circuit-architecture-training co-design framework that exploits sense-amplifier (SA) regeneration to detect analog-output similarity and reduce SAR comparisons in compute-in-memory (CIM) systems. Hardware-aware training incorporates circuit-characterized SA disturbance and encoding errors caused by prefix reuse, enabling aggressive comparison skipping. The detector is characterized through 55-nm CMOS schematic simulations, with system-level evaluation on WRN-28-10, ResNet20, and DeiT using an ISAAC-based W4A4 CIM model. On WRN-28-10, the proposed approach achieves 77.3% Top-1 accuracy (W4A4 baseline: 78.4%) while reducing SAR comparisons by 48.19% across the evaluated layers. Energy-budget analysis estimates a 27.18% reduction in reference ADC energy after detector overhead, leaving 0.52 pJ per conversion to accommodate additional control and peripheral costs.