DPIM: A 19.36 TOPS/W 2T1C eDRAM Transformer-in-Memory Chip with Sparsity-Aware Quantization and Heterogeneous Dense-Sparse Core

  • Kim, Donghyuk; 
  • Kim, Jae Young; 
  • Cho, Hyunjun; 
  • Yoo, Seungjae; 
  • Lee, Sukjin; 
  • 외 14명
Citations

WEB OF SCIENCE

4
Citations

SCOPUS

9

초록

This paper presents DPIM, the first 2T1C eDRAM Transformer-in-memory chip. Its high-density eDRAM cell supports large-capacity processing-in-memory (PIM) macros of 1.38 Mb/mm(2), reducing external memory access. DPIM adopts a sparse-aware quantization scheme to entire layers of Transformer, which quantizes the model to 8-bit integer (INT8) with a minimal accuracy drop of 2% in the BERT-large model on the GLUE dataset while increasing the bit-slice sparsity ratio of both weight and activation from dense matrices to 83.3% and 88.4%, respectively. Its heterogeneous PIM macro supports intensive dense matrix multiplications with an extreme to moderate range of sparse matrix multiplications with a peak throughput of 3.03-12.12 TOPS, enhancing the efficiency up to 4.84-19.36 TOPS/W.

키워드

Transformer; 2T1C cell; Heterogeneous processor; Processing-in-Memory
제목
DPIM: A 19.36 TOPS/W 2T1C eDRAM Transformer-in-Memory Chip with Sparsity-Aware Quantization and Heterogeneous Dense-Sparse Core
저자
Kim, Donghyuk; Kim, Jae Young; Cho, Hyunjun; Yoo, Seungjae; Lee, Sukjin; Yune, Sungwoong; Jeong, Hoichang; Park, Keonhee; Lee, Ki-Soo; Lee, Jongchan; Han, Chanheum; Koo, Gunmo; Han, Yuli; Kim, Jaejin; Kim, Jaemin; Lee, Kyuho; Cha, Joo-Hyung; Cho, Kunhee; Kim, Joo-Young
DOI
10.1109/ESSERC62670.2024.10719539
발행일
2024
유형
Proceedings Paper
저널명
2024 50TH IEEE EUROPEAN SOLID-STATE ELECTRONICS RESEARCH CONFERENCE, ESSERC 2024
페이지
141 ~ 144