Exploration of Unary Arithmetic-Based Matrix Multiply Units for Low Precision DL Accelerators

Prabhu Vellaisamy 1 , Harideep Nair 1 , Di Wu 2 , Shawn Blanton 1 , and John Paul Shen 1 · 2026

Read Exploration of Unary Arithmetic-Based Matrix Multiply Units for Low Precision DL Accelerators by Prabhu Vellaisamy 1 , Harideep Nair 1 , Di Wu 2 , Shawn Blanton 1 , and John Paul Shen 1 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-00838.txt.

Description

Prabhu Vellaisamy and colleagues show how unary GEMM can serve low-power compute in future edge AI accelerators. They evaluate unary matrix-multiply designs for low-precision deep learning inference against conventional binary hardware. They compare three recent unary proposals, uGEMM, tuGEMM and tubGEMM, with a conventional binary GEMM, using post-synthesis results across bit-widths and matrix sizes. They also analyse weight sparsity across eight pretrained convolutional networks and the LLaMA2 large language model.

Catalog record

CREATORPrabhu Vellaisamy 1 , Harideep Nair 1 , Di Wu 2 , Shawn Blanton 1 , and John Paul Shen 1
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE24773
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.00838

Browse the full library · About the Tower of Babel