Search papers, labs, and topics across Lattice.
This paper presents Eco-SoC, a scalable VLSI architecture designed for energy-proportional artificial intelligence, addressing the environmental impact of semiconductor manufacturing and operation. By implementing a Dynamic Precision-Scaling Logic (DPSL) framework, Eco-SoC reduces switching activity by up to 42% while maintaining performance, and it achieves a favorable Life Cycle Assessment that offsets its carbon footprint within 1.1 years of deployment. Additionally, the architecture incorporates a thermal-aware power gating mechanism that significantly enhances the Mean Time To Failure (MTTF), contributing to e-waste mitigation in future computing systems.
Eco-SoC achieves a remarkable 42% reduction in switching activity while offsetting its carbon footprint in just over a year of deployment.
In an era defined by escalating climate change and the pervasive deployment of edge intelligence, the environmental cost of semiconductor manufacturing and operation has reached a critical threshold. As Deep Learning (DL) accelerators dominate System-on-Chip (SoC) die area, achieving true sustainability requires a paradigm shift from static worst-case efficiency to dynamic energy-proportionality. This paper introduces Eco-SoC, a highly scalable VLSI architecture co-designed specifically for sustainable artificial intelligence. We propose a hardware-level Dynamic Precision-Scaling Logic (DPSL) framework that adaptively modulates bit-width precision based on real-time activation sparsity, successfully reducing switching activity by up to 42% on a commercial 7nm FinFET process node. Furthermore, we transcend traditional Power-Performance-Area (PPA) metrics by providing a comprehensive Life Cycle Assessment (LCA) using the Architectural Carbon footprint Tool (ACT). Our synthesis demonstrates that Eco-SoC offsets its increased embodied carbon footprint (a marginal 4.8% area overhead) within 1.1 years of edge deployment. Finally, by introducing a thermal-aware power gating mechanism that mitigates localized hotspots, Eco-SoC doubles the projected Mean Time To Failure (MTTF) of the silicon, providing a tangible, scalable strategy for electronic waste (e-waste) mitigation in next-generation computing systems.
Microsoft Research