Reproduce “Eco-SoC: A Sustainable VLSI Architecture for Energy-Proportional Artificial Intelligence” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jatin Chopra
Eco-SoC is a VLSI architecture designed for energy-proportional AI, which reduces dynamic power consumption by 42.0% through dynamic precision scaling and extends hardware lifespan by over 100% using thermal-aware power gating. This approach enables a carbon payback period of 1.1 years for its increased embodied carbon, demonstrating a holistic strategy for sustainable edge AI.
Key insights
- Adopting a holistic sustainability framework, encompassing embodied carbon, operational energy, and device longevity, is essential for future VLSI design beyond traditional PPA metrics.
- Dynamically adapting hardware precision based on real-time workload requirements can effectively bridge the gap between algorithmic sparsity gains and actual silicon-level energy reductions.
- Thermal management is directly correlated with device reliability and Mean Time To Failure (MTTF), presenting a critical lever for mitigating electronic waste and enhancing overall system sustainability.
- The operational energy savings from sustainable hardware can quickly amortize its initial embodied carbon footprint, achieving a positive carbon payback within a short period.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.