Publications
-
Precision-Scalable Microscaling Datapaths with Optimized Reduction Tree for Efficient NPU Integration
Stef Cuyckens, Xiaoling Yi, Robin Geens, Joren Dumoulin, Martin Wiesner, Chao Fang, and Marian Verhelst · 31st Asia and South Pacific Design Automation Conference (ASP-DAC 2026) — Invited Paper · 2026
Proposes a hybrid precision-scalable reduction tree for microscaling (MX) multiply-accumulate operations in neural processing units, enabling efficient mixed-precision accumulation. Integrated into the SNAX NPU platform, achieving energy efficiencies from 657 to 4065 GOPS/W and throughputs from 64 to 512 GOPS.