Our paper “Sparrow ECC: A Lightweight ECC Approach for HBM Refresh Reduction towards Energy-efficient DNN Inference” was awarded the Best Paper Award at the ACM/IEEE International Symposium on Low Power Electronics and Design (ISLPED) 2024.
About the Paper
We proposed Sparrow ECC, a lightweight error-correcting code approach designed to reduce HBM refresh overhead while maintaining reliability for DNN inference workloads. The key insight is that not all memory regions require the same level of protection, and by adapting ECC strength to actual error patterns, we can significantly reduce energy consumption.
The Conference
ISLPED is one of the premier venues for low-power electronics and design research. This year’s conference featured excellent talks on energy-efficient computing, from circuit-level techniques to system-level optimizations.
Key Takeaways
- Growing interest in memory-centric energy optimization for AI workloads
- HBM reliability and refresh reduction is an increasingly important topic
- Great discussions with researchers from industry and academia on future directions
This is a sample blog post. Replace with your actual trip report and photos!