Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Neural Network Acceleration on MPSoC board: Integrating SLAC's SNL, Rogue Software and Auto-SNL

Created by
  • Haebom

Author

Hamza Ezzaoui Rahali, Abhilasha Dave, Larry Ruckman, Mohammad Mehdi Rahimifar, Audrey C. Therrien, James J. Russel, Ryan T. Herbst

Outline

This paper introduces the SLAC Neural Network Library (SNL), developed at SLAC to address the 1 MHz T80-ray pulse data processing challenge of the LCLS-II FEL. SNL is a specialized framework for real-time machine learning inference on FPGAs, supporting dynamic updates of model weights to provide flexibility in adaptive learning. We also present Auto-SNL, which converts Python-based neural network models to SNL-compatible code, and demonstrate SNL's competitive latency and FPGA resource-saving efficiency through a performance comparison with hls4ml. We suggest potential applications in various fields such as high-energy physics, medical imaging, and robotics.

Takeaways, Limitations

•
Takeaways:
◦
SNL provides an efficient framework for FPGA-based real-time machine learning inference.
◦
Support for adaptive learning through dynamic model weight update.
◦
Provides Auto-SNL, a Python extension for improved usability.
◦
Demonstrated competitive performance and resource efficiency compared to hls4ml.
◦
Suggests applicability to various fields requiring high-speed data processing.
•
Limitations:
◦
Lack of specific mention of long-term maintenance and community support systems for SNL and Auto-SNL.
◦
Further research is needed on scalability to different FPGA architectures and other ML models.
◦
The benchmark results presented in this paper are limited to a specific FPGA (Xilinx ZCU102) and require further verification of generalizability.
👍
Made with Slashpage