LEGO: Spatial Accelerator Generation and Optimization for Tensor Applications

Explore a 16-minute talk from HPCA 2025 titled "LEGO: Spatial Accelerator Generation and Optimization for Tensor Applications," presented by Zhekai Zhang from MIT HAN Lab. This presentation, co-authored by Yujun Lin, Zhekai Zhang, and Song Han, introduces the LEGO framework for generating and optimizing spatial accelerators specifically designed for tensor applications. Learn about cutting-edge approaches to hardware acceleration for tensor operations that are crucial for modern machine learning workloads. For additional information and resources about the LEGO project, visit the official website at https://hanlab.mit.edu/projects/lego.