SNN-MLIR: An MLIR Dialect for Compiling Neuromorphic SNNs from NIR to Bare-Metal C
Researchers have developed SNN-MLIR, a new MLIR dialect designed to compile spiking neural networks (SNNs) from a common intermediate representation (NIR) into C code for bare-metal deployment. This tool addresses the fragmentation of SNN training frameworks by providing a unified compiler representation that supports both floating-point and quantized data. The system includes a Python frontend to read NIR files and a lowering pass that generates self-contained C11 code, currently supporting feedforward, fully-connected networks on CPU targets. AI
IMPACT Enables more efficient deployment of spiking neural networks on diverse hardware platforms.