18.2 A 22nm 1.87ms/Frame Streaming Multi-Speaker ASR Accelerator Leveraging Contextual-Aware Redundancy Skipping with 2D-Writable Microscaling Compute-in-Memory and Similarity-Aware TCAM Design

Digest of technical papers/Digest of technical papers - IEEE International Solid-State Circuits Conference 2026

View at publisher
Authors: Wenjie Ren, Mingxuan Li, Zhenghao Jin, Yifan Ding, Ruohuang Xu, Xuelian Ye, Yufei Ma, Tianyu Jia, Le Ye
DOI: 10.1109/isscc49663.2026.11409100
Cited by: 0
References: 5

Abstract

This paper presents a DCIM-based accelerator for on-device streaming multi-speaker ASR (MS-ASR), featuring: 1) a context-aware redundancy skipping scheme with online sparse block prediction, 2) a 2D-writable CIM with static pages and dynamic pages, 3) a similarity-aware TCAM for rapid search of similar speakers and N-best semantic results. The test chip achieves a system energy efficiency of 37.50TFLOPS/W in MXFP8 and shows superior performance on MS-ASR tasks with$1.87-7.51 \text{ms} /$frame and$0.158-1.26 \text{mJ} /$frame efficiency.

Fields of Study & Subjects

Speech Recognition and Synthesis Advanced Neural Network Applications Advanced Memory and Neural Computing

Publication details

Journal / source
Digest of technical papers/Digest of technical papers - IEEE International Solid-State Circuits Conference
Pages
310–312
Publication date
15 February 2026
Language
EN