CARMEN: CORDIC-Accelerated Resource-Efficient Multi-Precision Inference Engine for Deep Learning

2026-05-07

AI Summary

This work addresses the challenge of balancing accuracy and hardware efficiency in deep learning inference by proposing a runtime-adaptive multi-precision vector engine. For the first time, it dynamically links CORDIC iteration depth with computational precision, enabling flexible switching between approximate and exact modes without any hardware modifications. The architecture features CORDIC-based iterative multiply-accumulate (MAC) units, a time-multiplexed activation function module, and configurable 8/16-bit precision support. Implemented in 28nm ASIC, the design reduces computation cycles by 33% and power consumption by 21% per MAC stage. A 256-PE configuration achieves 4.83 TOPS/mm² and 11.67 TOPS/W. FPGA-based evaluation demonstrates a target detection latency of 154.6 ms at a power consumption of only 0.43 W.

Abstract

This paper presents CARMEN, a runtime-adaptive, CORDIC-accelerated multi-precision vector engine for resource-efficient deep learning inference. The key insight is that CORDIC iteration depth directly governs computational accuracy, enabling dynamic switching between approximate and accurate execution modes without hardware modification. The architecture integrates a low-resource iterative CORDIC-based MAC unit with a time-multiplexed multi-activation function block, supporting flexible 8/16-bit precision and high hardware utilization. ASIC implementation in 28 nm CMOS achieves up to 33% reduction in computation cycles and 21% power savings per MAC stage; a 256-PE configuration delivers 4.83 TOPS/mm2 compute density and 11.67 TOPS/W energy efficiency. FPGA deployment on PynqZ2 validates 154.6 ms latency at 0.43 W for real-time object detection.

Problem

• deep learning inference
• multi-precision
• resource-efficient
• hardware acceleration
• energy efficiency

Innovation

• CORDIC
• multi-precision
• resource-efficient
• runtime-adaptive
• deep learning inference