Accelerated neural network simulations: novel neuromorphic system architecture approaches

Summary

In this project, we are exploring hard- and software architectural approaches to identify opportunities to accelerate simulations of complex neural networks.

Background and motivation

The critical biological ingredients and underlying principles of brain functions are still widely unknown. Research on neural network dynamics and function is to a large extent based on simulations on traditional von Neumann computer architectures, in which memory and computation are separated. The resultant performance limitation of such systems is known as the von Neumann bottleneck. Even with the advanced highly parallel petascale supercomputers available today, simulations of neural networks are orders of magnitude slower than biological real time, hindering the study of slow biological processes such as learning and development. Accelerated processing at a lower cost while maintaining flexibility, accuracy, and reproducibility of simulation results is the current challenge that a neuromorphic computing system has to cope with to serve as a research platform for neuroscience. This trade-off between flexibility and efficiency requires novel system architecture approaches beyond the von Neumann model.

Our approach

As part of the neuroscience-driven Advanced Computing Architectures (ACA) project, we are exploring novel neuromorphic system architecture approaches, such as hybrid soft- and hardware designs that combine a traditional von Neumann architecture with hardware acceleration units. Those units transform existing simulation codes and algorithms into highly parallel operational digital designs. This could be a computation primitive or special function, a neuron model or even an entire neural network. The possibility of utilizing the state-of-the-art in reconfigurable hardware, e.g. FPGAs (Field Programmable Gate Arrays) or SoC (System on Chip) devices, facilitates access to design space explorations and prototyping to identify opportunities for hardware acceleration. 

Our collaboration partners

This project is being conducted in collaboration with PGI-10.

Acknowledgments

ACA is funded by the Helmholtz Association’s Initiative and Networking Fund under Grant Agreement SO-092.