Late Breaking Results: Leveraging Approximate Computing for Carbon-Aware DNN Accelerators

2025-02-01

AI Summary

Deep neural network (DNN) accelerators incur substantial embodied carbon emissions during manufacturing and operation, yet lack systematic carbon-aware hardware design methodologies. Method: This paper proposes a sustainable hardware design framework optimizing the Carbon-Delay Product (CDP), integrating approximate computing with carbon-aware design for the first time. It introduces a gate-level pruned and precision-scaled area-, energy-, and carbon-sensitive approximate multiplier, and employs a genetic algorithm for architecture-level co-optimization. Contribution/Results: The approach maintains target performance and model accuracy while significantly reducing embodied carbon across both fabrication and operational phases. Experimental evaluation shows an average CDP reduction of 37.2%, establishing a quantifiable, optimization-driven paradigm for low-carbon AI hardware.

Technology Category

• Machine Learning: Hardware-aware ML
• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Sustainability and carbon-aware systems for Web, mobile, and WoT
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Responsible Web: Sustainability and climate impact of web technologies

Abstract

The rapid growth of Machine Learning (ML) has increased demand for DNN hardware accelerators, but their embodied carbon footprint poses significant environmental challenges. This paper leverages approximate computing to design sustainable accelerators by minimizing the Carbon Delay Product (CDP). Using gate-level pruning and precision scaling, we generate area-aware approximate multipliers and optimize the accelerator design with a genetic algorithm. Results demonstrate reduced embodied carbon while meeting performance and accuracy requirements.

Problem

• Carbon Emission Reduction
• Deep Learning Accelerators
• Environmental Sustainability

Innovation

• Approximate Computing
• Genetic Algorithm Optimization
• Energy Efficiency