Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
September 27, 2025ACM Transactions on Design Automation of Electronic SystemsOpen Access

Towards Floating Point-Based AI Acceleration: Hybrid PIM with Non-Uniform Data Format and Reduced Multiplications

View Full Paper
Ask AI
Bookmark
Share

Authors

LGLidong GuoHeilongjiang University of Chinese MedicineZZZhenhua ZhuTsinghua UniversityXNXuefei NingNational Engineering Research Center for Information Technology in Agriculture

Discussion

Loading...

Member takes

Overview

Experimental analysis shows improved inference accuracy in neural networks, indicating the effectiveness of hybrid PIM.

Key Points

  • Achieving up to 99.4× speedup and 5697.7× energy efficiency improvement compared to GPU.
  • The proposed PN format allows for flexible adjustment and maintains floating-point accuracy in operations.
  • Hybrid PIM architecture reduces hardware overhead while improving memory utilization.
  • Quantization error analysis reveals method adaptations for different operations, enhancing neural network performance.

Cite This Study

Guo et al. (2025) studied this question.

synapsesocial.com/papers/68d7cc66eebfec0fc5238857https://doi.org/10.1145/3769304
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1DeepOTF: Learning Equations-constrained Prediction for Electromagnetic Behavior2024 · 1 citations
  2. 2ImageNet: A large-scale hierarchical image database2009 · 63,551 citations
  3. 3FP8 Quantization: The Power of the Exponent2022 · 13 citations
  4. 4ISOAcc: In-situ Shift Operation-based Accelerator For Efficient in-SRAM Multiplication2024 · 1 citations
  5. 53D SRAM Macro Design in 3D Nanofabric Process Technology2023 · 6 citations