Bringing up bare metal ExecuTorch on RISC-V

Jeremy Bennett is the founder and Chief Executive of Embecosm, a leading consultancy specializing in open-source compiler tool chains, processor modeling, and AI/ML tooling. A former academic, he is author of the standard textbook "Introduction to Compiling Techniques: A first course using ANSI C, Lex and YACC" (McGraW-Hill 1990, 1995, 2003). He remains an active, hands-on engineer with particular interests in deeply embedded systems and energy efficiency.

Shane Slattery is a member of Embecosm’s tool chain engineering team. He specialises in embedded software and compiler development, with expertise in LLVM compilers and Edge AI. His current focus is on porting ExecuTorch, PyTorch’s embedded framework, to bare-metal microcontrollers. By lowering the barriers to advanced AI capabilities, it is possible to run complex tasks such as image processing and audio streaming efficiently on highly resource-constrained hardware.

Pietra Ferreira is a member of Embecosm’s tool chain engineering team. She specialises in embedded software and compiler design, notably contributing to the Open Hardware Group’s GCC compiler tool chain. Her recent work focuses on Edge AI and constrained computing, where she has demonstrated how to adapt the PyTorch embedded framework, ExecuTorch, for bare-metal microcontroller-class processors. This project significantly lowers the barrier for running sophisticated AI applications, such as small image processing and streaming audio, on hardware with minimal resources.