Design and Simulation of Reversible Neuromorphic Architectures

This DC will study the applicability of RC to non-von Neumann architectures, namely neuromorphic systems. In conjunction with DC 8, the DC will start surveying the state-of-the-art reversible architectures, but this will focus on neural simulations and will target neuromorphic computing. Neuromorphic computing aims to develop unconventional low energy architectures inspired by biological neural networks with emphasis on connectivity, recurrence and higher time resolutions. These platforms can in principle require less energy and target ML and AI. This DC will collaborate on developing the fundamentals of the simulation platform for RC architectures accelerating neural networks simulations and ML applications. The interaction between the standard chiplet and novel reversible neuromorphic accelerators would be the main research topic. This DC will work on architectures that will be beneficial for cyber-physical systems (collaborating with DC 12 on RC algorithms for aerial robots) and for ML applications (targeting platforms which can accelerate techniques developed by DC 13), enabling reversible on-chip online learning paradigms and efficient neural networks simulations.

1) Simulation platform for RC neuromorphic systems; 2) Energy simulation platform for RC neuromorphic architectures; 3) Novel insights in combining standard processing with new reversible computing-based neuromorphic systems.

M14, SDU, U. P. Schultz, requirements for neuromorphic architectures for cyber-physical systems; M31, AGH, I. Ulidowski, neuromorphic architectures and neural networks.