3rd place at the Grand Challenge ICIP 2025 won by the team formed by Alessio Caviglia, Filippo Marostica, Alessio Carpegna, Alessandro Savino e Stefano Di Carlo

The team formed by Alessio Caviglia, Filippo Marostica, Alessio Carpegna, Alessandro Savino and Stefano Di Carlo won 3rd place at the Digit Recognition Low Power and Speed Challenge, on the occasion of the IEEE International Conference on Image Processing - IEEE ICIP 2025, held from September 14 to 18, 2025, in Anchorage (AK, USA).
The team, belonging to the DAUIN research group reSilient coMputer archItectures and LIfE Sciences SMILIES, won the study award titled SFATTI: Spiking FPGA Accelerator for Temporal Task-driven Inference - A Case Study on MNIST.

ABSTRACT: The ICIP 2025 Grand Challenge "Digit Recognition – Low Power and Speed" measures the speed, energy efficiency, and accuracy (≥ 97.5%) of FPGA accelerators for MNIST. The project, SFATTI, is an SNN accelerator developed using the Spiker+ framework. It uses surrogate gradients for software training, applies hardware-aware quantization, and automatically generates optimized HDL. The result demonstrates how SNNs on FPGAs can deliver high throughput and low power consumption.

BIO: The SMILIES group at the Department of Control and Computer Science (DAUIN) works on neuromorphic computing, Spiking Neural Networks and FPGA accelerators. This activity integrates training algorithms and the development of digital architectures for edge applications.

More information
- Preprint (arXiv): https://arxiv.org/abs/2507.10561
- GitHub: https://github.com/smilies-polito/sfatti