TY  - RPRT
TI  - W4A16 Mixed-Precision Matrix Multiplication on Decoupled Architecture: Kernel Design and Memory Bottleneck Analysis for Ascend NPUs
AU  - Yuanhong He
AU  - Peiyu Niu
AU  - Jun Chen
AU  - Chenchen Zhang
AU  - Chao Yang
PY  - 2026
UR  - https://arxiv.org/abs/2601.16536
ID  - 2601.16536
ER  -