Robert Kleijnen - DR project
Intra- and inter-chip communication of a neuromorphic accelerator
Doctoral Researcher: Robert Kleijnen
Topic: Neuromorphic Computing
Research Field: Information
Neuromorphic computing
As opposed to most traditional computers, which rely on the von Neumann architecture, neuromorphic systems are based on the structure of the mammal brain. These so called neural networks show great potential for efficient computing, especially for applications such as pattern recognition and the processing of noisy data.
Another field of interest for these systems is the use in computational neuroscience. They offer the potential to investigate learning processes in the human brain at an accelerated rate, something even the most powerful super computers to date are not able to do.
Inter- and intra- chip communication
The fan-out of neurons in our brains is between 50 and 50.000 times larger than the typical fan-out of a VLSI logic gate, resulting in a design challenge to facilitate the communication traffic in a neuromorphic system. Aim of this research is to develop a VLSI-based concept for a communication network facilitating faster than real-time simulation of multi-scale spiking neural networks with natural density connectivity.
In the last decade, a number of system have been developed. However, none of these current systems is able to simulate neural networks comparable in scale, and connectivity density and at an accelerated rate, all at the same time. The first step in this research is to determine the limiting factors preventing the up-scaling of the current systems by performing network simulations.