A Current-Mode Multiply-Accumulate Macro in Sensing-Computing Fusion System for Feature Extraction and Redundancy Reduction

This brief presents a current-mode sensing-computing fusion system for advanced Internet of Things (IoT) machine vision. The key contributions of our work are: (a) a low-voltage of 0.9V multiply-accumulate (MAC) computing macro with reconfigurable weights is proposed, enabling efficient on-sensor feature extraction; (b) a weight-flipping method for processing negative signals is employed, reducing both power consumption and circuit complexity; (c) a convolutional horizontal shifting technique with fixed weights is equipped, eliminating power consumption associated with weight updates. A <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$16\times 16$ </tex-math></inline-formula> CCIS prototype, fabricated using a 0.18<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mu $ </tex-math></inline-formula>m CMOS process, achieves a power efficiency of 27.7pJ/frame<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\cdot $ </tex-math></inline-formula>pixel at 2000fps. Experimental evaluations in edge feature extraction demonstrate a 32% reduction in power consumption, highlighting the efficiency gains of our approach.