EinChip: A 4.6/1.5 TOPS/W 6/24b Log-compute Einsum-based Accelerator for Neuro Symbolic AI

Yao, Lingyun;Zhao, Shirui;Verhelst, Marian;Andraud, Martin
(2026) 2026 Custom Integrated Circuit Conference

Files

EinChip_A_4.6_1.5_TOPS_W_6_24b_Log-compute_Einsum-based_Accelerator_for_Neuro_Symbolic_AI_compressed.pdf
  • Open Access
  • Adobe PDF
  • 528.27 KB
Download

Details

Authors
  • Yao, LingyunAalto University,ELEC,Espoo,Finland,02150
    Author
  • Zhao, ShiruiKU Leuven,MICAS,Leuven,Belgium,3000
    Author
  • Verhelst, MarianKU Leuven,MICAS,Leuven,Belgium,3000
    Author
  • Andraud, MartinAalto University,ELEC,Espoo,Finland,02150
    Author
Abstract
Neuro Symbolic (NeSy) AI combines deep neural networks (DNNs) with symbolic reasoning, but existing DNN accelerators are ill-suited for NeSy workloads. We present EinChip, a unified NeSy accelerator that expresses both neuro and symbolic models in Einstein Summation Notation (Einsum) and executes them on an approximate 6/24-bit logarithmic processing unit. A 4mm2 16nm prototype achieves 4.6 TOPS/W on neuro tasks and 1.5 TOPS/W on symbolic tasks with only 0.1% accuracy loss.
Affiliations

Citations

  • APA
  • Chicago
  • FWB

Yao, L., Zhao, S., Verhelst, M., & Andraud, M. (2026). EinChip: A 4.6/1.5 TOPS/W 6/24b Log-compute Einsum-based Accelerator for Neuro Symbolic AI. Proceedings of 2026 Custom Integrated Circuit Conference, 1-4. https://doi.org/10.1109/CICC65509.2026.11509460 (Original work published 2026)