UoA Competitive Research Funding (Ref. 2023-26): Low-power Spiking Neural Network Solution for IoT and Edge devices (2023)

Project Summary

In this project, we develop a low-power solution for spiking neural networks using stacking 3D memory. Here, we split the weight of N-bit into several subsets and each subset is put into separated layers. We apply power-gating (dynamic quantization) or voltage scaling for LSBs to reduce the power consumption.

This project is funded by UoA Competitive Research Funding in AY 2023.

Project’s Output

Open-source

HeterGenMap: An Evolutionary Mapping Framework for Heterogeneous NoC-based Neuromorphic Systems

Patent

  1. Khanh N. Dang, A. Ben Abdallah, Nguyen Ngo Doanh, '’Neural Network Processor’’ [ニューラルネットワークプロセッサ], Japan patent, (to file).

Journal

  1. Khanh N. Dang, Nguyen Anh Vu Doan, Ngo-Doanh Nguyen, Abderazek Ben Abdallah, '’HeterGenMap: An Evolutionary Mapping Framework for Heterogeneous NoC-based Neuromorphic Systems’‘, IEEE Access, vol. 11, pp. 144095-144112, 2023. [DOI: 10.1109/ACCESS.2023.3345168]/[PDF]/[Source Code].
  2. Ngo-Doanh Nguyen, Akram Ben Ahmed, Abderazek Ben Abdallah, Khanh N. Dang, '’Power-aware Neuromorphic Architecture with Partial Voltage Scaling 3D Stacking Synaptic Memory’‘, IEEE Transactions on Very Large Scale Integration Systems (TVLSI), vol. 31, no. 12, pp. 2016-2029, Dec. 2023. [DOI: 10.1109/TVLSI.2023.3318231]/[PDF].
  3. Ngo-Doanh Nguyen, Xuan-Tu Tran, Abderazek Ben Abdallah, Khanh N. Dang, '’An In-situ Dynamic Quantization with 3D Stacking Synaptic Memory for Power-aware Neuromorphic Architecture’‘, IEEE Access, vol. 11, pp. 82377-82389, 2023. [DOI: 10.1109/ACCESS.2023.3311031]/[PDF].

Conference

  1. Ngo-Doanh Nguyen, Khanh N. Dang, '’A Novel Yield Improvement Approach for 3D Stacking Neuromorphic Architecture’‘, 2023 IEEE 16th International Symposium on Embedded Multicore*Many-core Systems-on-Chip (MCSoC), Dec. 18-21, 2023.

Demonstration

  1. See this slide for the demo: 2023_Demo_FPGA.pdf.
  2. Video of the SNN SoC on FPGA: