Michael Rontionov MINDS CDT, 2023

Doctoral Researcher, University of Southampton

Michael Rontionov is a doctoral researcher working on hardware-software co-design for neuromorphic computing, a field that draws on how the brain processes information to build AI systems that are more compact and energy-efficient than conventional approaches.

The gap his research addresses is a practical one. Neuromorphic algorithms are typically designed and tested in software simulators. Moving from that simulated environment to actual hardware, without losing accuracy, flexibility or efficiency, is difficult and there is no standardised way to do it. That inconsistency slows the whole field down.

His work uses compiler techniques, including ideas from category theory and dependently-typed programming languages, to make this process more systematic. The first phase of his PhD produced NIR2FPGA, a tool that takes a neuromorphic algorithm defined in a simulator and compiles it into efficient hardware for an FPGA. The second phase uses Agda, a formal verification language, to build a mathematically rigorous framework for the same problem. This approach could eventually extend to support analogue computing devices as well as digital ones.

The aim is to make it straightforward for researchers and engineers to move from a working neuromorphic algorithm to working hardware, with confidence that the translation has been done correctly.

Michael is due to graduate in 2028 and is planning to move into industry R&D.