Accuracy-Configurable Floating-Point Multiplier Design for SRAM-Based Compute-in-Memory

2026-06-06

AI Summary

This work addresses the lack of compiler integration and configurable-precision floating-point support in existing SRAM-based in-memory computing systems, as conventional IEEE 754 floating-point units incur substantial area and power overheads. Within the OpenACM framework, the study presents the first compiler-integrated, precision-configurable floating-point multiplier. It begins by establishing an IEEE 754-compliant baseline design and then introduces a mantissa-segmentation-based approximate multiplication strategy that significantly reduces hardware costs without adding latency. Post-layout results demonstrate a 69% reduction in logic area and a 72% decrease in power consumption, while maintaining negligible accuracy loss on image processing and ResNet-18 inference tasks, thereby validating the approach’s efficiency and practicality.

Abstract

Digital Compute-in-Memory (DCiM) reduces data movement and has become a promising solution for energy-efficient edge AI. However, most existing DCiM frameworks still primarily target integer or fixed-point arithmetic, and provide limited support for compiler-integrated and accuracy-configurable floating-point computation. Directly integrating conventional IEEE 754 floating-point units into dense SRAM-based DCiM arrays, however, incurs high area and power overhead. To address this challenge, this work presents an accuracy-configurable floating-point multiplier integrated into the OpenACM framework for SRAM-based DCiM. An exact IEEE~754-compliant multiplier is first implemented as a baseline, and a mantissa-segmentation-based approximate multiplier is then proposed to reduce hardware cost while preserving numerical fidelity. Post-layout results show up to 69% logic area reduction and 72% power savings over exact floating-point designs without delay overhead. Evaluations on image processing tasks and ResNet-18 inference further demonstrate negligible accuracy degradation. These results indicate that compiler-integrated approximate floating-point multiplication is a practical approach for enabling efficient and configurable floating-point support in SRAM-based DCiM systems. The Floating-Point Multiplier is available on https://github.com/ShenShan123/OpenACM

Problem

• Compute-in-Memory
• Floating-Point Multiplier
• Accuracy Configurability
• SRAM
• Hardware Overhead

Innovation

• accuracy-configurable
• floating-point multiplier
• Compute-in-Memory
• approximate computing
• SRAM-based DCiM