TREA: Low-precision T ime-Multiplexed, R esource-Efficient E dge A ccelerator for Object Detection and Classification

Vijay Pratap Sharma, Mukul Lokhande, Ratko Pilipović, Omkar Kokane, Santosh Kumar Vishvakarma · 2026

Read TREA: Low-precision T ime-Multiplexed, R esource-Efficient E dge A ccelerator for Object Detection and Classification by Vijay Pratap Sharma, Mukul Lokhande, Ratko Pilipović, Omkar Kokane, Santosh Kumar Vishvakarma (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-2605-07321.txt.

Description

Vijay Pratap Sharma and four co-authors design TREA, an edge chip for spotting and sorting objects. Edge vision platforms face tight limits on area, power, and delay. TREA uses a dual-precision multiply unit that computes most-significant digits first, supports runtime truncation, and avoids the usual multiplier cost.

A pruning strategy co-designed with the datapath keeps usage high at near half sparsity, and an adjustable unit handles several activation types through time sharing. One hundred such units are reused layer by layer across the network. Experiments show large gains in speed, usage, and energy use over fixed-width, non-adjustable chips.

Catalog record

CREATORVijay Pratap Sharma, Mukul Lokhande, Ratko Pilipović, Omkar Kokane, Santosh Kumar Vishvakarma
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE43 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.07321

Browse the full library · About the Tower of Babel