Communication systems for neuromorphic massive multicore systems
Background
Neuromorphic computing denotes brain-inspired systems for artificial intelligence (AI). It promises vast performance improvements and power savings compared to conventional von-Neuman architectures. These systems are researched at RWTH within multiple large-scale projects.
One major bottleneck of neuromorphic systems is their communication infrastructure. They determine, to a large part, the overall system performance. In this thesis, their architecture will be improved.
Description
The goal of this thesis is to extend a SystemC simulator for neuromorphic computing to model NoCs with a fast and abstract performance model. Communcation patterns for common AI benchmarks are analyzed. Then, an optimized architecture is proposed to improve latency.
Tasks
- Develop a methodology to model NoC architectures
- Evaluate the impact of the NoC on the overall system performance
- Include the model into nueromorphic simulation
Supervisor
Requirements
Must have:
- Good C++ and Python knowledge
- Interest in computer architectures