• MLIR support for Neuromorphic workflows Hi all, I´m an electronics engineer working on SNNs (Spiking Neural Networks) on the edge for the last years. • Seems that right now the Neuromorphics community has an heterogeneity problem for back-ends, hardware architectures and a lack of compilers. • True that this is increasing with the adoption of Intermediate Representations (such as NIR) but still there is a gap for low level compilers. • In this context we’ve thought about creating a SNN dialect based on MLIR to help improving standardisation. • But before, I wanted to ask the official community if there is someone already working on something similar or if there are any plans ahead for supporting SNNs. • We’ll be happy to collaborate together.

Article Summaries:

  • Alex, an electronics engineer focused on edge spiking neural networks (SNNs), highlights a growing heterogeneity issue in the neuromorphic community-diverse back‑ends, hardware architectures, and a shortage of compilers. While intermediate representations like NIR are gaining traction, low‑level compiler support remains limited. To address this, Alex proposes developing an SNN dialect built on MLIR to standardize workflows and improve compiler interoperability. He invites the community to share any existing efforts or future plans for SNN support and offers collaboration on the initiative.
  • Hi all, I´m an electronics engineer working on SNNs (Spiking Neural Networks) on the edge for the last years. Seems that right now the Neuromorphics community has an heterogeneity problem for back-ends, hardware architectures and a lack of compilers. True that this is increasing with the adoption of Intermediate Representations (such as NIR) but still there is a gap for low level compilers. In this context we’ve thought about creating a SNN dialect based on MLIR to help improving standardisation. But before, I wanted to ask the official community if there is someone already working on somethin

Sources: