The research team led by Professor Wang Zhongfeng at Nanjing University has recently been awarded the prestigious 2025 IEEE Journal of Solid-State Circuits (VLSI Journal) Best Paper Award for their groundbreaking work titled "An Algorithm-Hardware Co-Optimized Framework for Accelerating N:M Sparse Transformers." This innovative paper introduces an algorithm-hardware co-optimized sparse acceleration framework specifically tailored for Transformers, which successfully extends the fixed 2:4 sparse pattern supported by NVIDIA GPU architecture to an adaptable integer ratio N:M sparse pattern. By synergistically optimizing both algorithms and hardware architecture, this framework achieves remarkable hardware acceleration, vastly improving the flexibility and computational efficiency of model deployment.

The Transformer model stands as a cornerstone in the architecture of deep neural networks, making this research finding particularly significant for the advancement and application of AI technologies. Since its publication in November 2022, the paper has garnered significant attention and acclaim from numerous universities and globally renowned corporations. The VLSI Journal, a leading publication in the realm of international integrated circuit design, bestows this award on no more than one paper annually, adhering to stringent selection criteria that underscore the exceptional quality and impact of the research.