EUVIMEDEUVIMED entdecken
Crossref

A 28 nm 128-kb Exponent- and Mantissa-Computation-In-Memory Dual-Macro for Floating-Point and INT CNNs

An Guo, Xueshan Dong, Fangyuan Dong, Dongqi Li, Yiran Zhang, Jingmin Zhang, Jun Yang, Xin Si

IEEE Journal of Solid-State Circuits · 2025-10 · 60 (10) · 3639-3654

DOI: 10.1109/jssc.2025.3558322 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Journal of Solid-State Circuits, 2025-10. Verfasst von An Guo und 7 weiteren Autor:innen. Die Arbeit wurde bisher 13-mal zitiert (Crossref) und nennt 32 Literaturangaben. Gefördert von: National Natural Science Foundation of China, National Science and Technology Major Project, Key Research and Development Program of Jiangsu Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Floating-point computing-in-memory (FP-CIM) has emerged as a promising approach to enhance the energy efficiency of cloud and edge artificial intelligence (AI) devices. However, recent advancements in large language models (LLMs) and complex neural networks demand efficient support for both FP and integer (INT) multiplication-and-accumulation (MAC) operations to achieve high performance. This work presents a 28 nm 128 kb exponent-computing-in-memory (ECIM) and mantissa-computing-in-memory (MCIM) dual-macro, which effectively supports both true FP and INT MAC operations. The contributions of this work include: 1) a dual-macro architecture enabling both INT and true FP MAC operations; 2) an ECIM unit for digital exponent (EXP) addition and analog INT MAC; 3) an EXP-mantissa alignment (EMA) unit with early exit capability for outlier situations; and 4) an MCIM unit featuring two-level approximate adder trees and offline LUT-based retraining. The fabricated chip achieves 45.3 TFLOPS/W in energy efficiency for FP MAC with over 91.44% hardware utilization, demonstrating significant performance and energy efficiency improvements compared to existing solutions.

Förderung

  • National Natural Science Foundation of China
  • National Science and Technology Major Project
  • Key Research and Development Program of Jiangsu Province
  • Postgraduate Research and Practice Innovation Program of Jiangsu Province
  • Fundamental Research Funds for the Central Universities

Zitieren

APA

Guo, A., Dong, X., Dong, F., Li, D., Zhang, Y., Zhang, J., Yang, J., & Si, X. (2025). A 28 nm 128-kb Exponent- and Mantissa-Computation-In-Memory Dual-Macro for Floating-Point and INT CNNs. IEEE Journal of Solid-State Circuits, 60(10), 3639-3654. https://doi.org/10.1109/jssc.2025.3558322

Vancouver

Guo A, Dong X, Dong F, Li D, Zhang Y, Zhang J, et al. A 28 nm 128-kb Exponent- and Mantissa-Computation-In-Memory Dual-Macro for Floating-Point and INT CNNs. IEEE Journal of Solid-State Circuits. 2025;60(10):3639-3654. doi:10.1109/jssc.2025.3558322

BibTeX
@article{Guo2025,
  author = {Guo, An and Dong, Xueshan and Dong, Fangyuan and Li, Dongqi and Zhang, Yiran and Zhang, Jingmin and Yang, Jun and Si, Xin},
  title = {A 28 nm 128-kb Exponent- and Mantissa-Computation-In-Memory Dual-Macro for Floating-Point and INT CNNs},
  journal = {IEEE Journal of Solid-State Circuits},
  year = {2025},
  volume = {60},
  number = {10},
  pages = {3639--3654},
  doi = {10.1109/jssc.2025.3558322}
}
Kostenlos starten →