Advanced Computing Architectures (ACA) Pilot Project Paving the Way Towards Jülich Research in Neuromorphic Computing
Researchers from Forschungszentrum Jülich, RWTH Aachen University, the University of Manchester, and Heidelberg University came together in mid-September at Burg Obbendorf near Jülich for a final meeting of the project "Advanced Computing Architectures (ACA): towards multi-scale natural-density neuromorphic computing", which ends in November. Funded by the Helmholtz Association, ACA is a pilot project with the long-term goal of building a computer system to enable the simulation of biological learning processes in time-lapse in order to better understand learning and brain development.
During the four-year project period, the foundation was laid for future developments of architectures specifically tailored to neuroscience simulations. Ideally, such neuromorphic accelerator architectures could in the future be coupled to the Jülich modular supercomputing infrastructure. Almost 30 peer-reviewed articles have resulted from the interdisciplinary ACA project. For example, a publication on connectivity concepts in neural network modelling, which can guide the implementation of connection routines in neuromorphic hardware systems, appeared in PLOS Computational Biology (DOI:10.1371/journal.pcbi.1010086), while another publication on a system-on-chip (SoC) based approach for a hybrid software-hardware neuromorphic compute node architecture appeared in Frontiers in Neuroscience (DOI:10.3389/fninf.2022.884033). The researchers also succeeded in simulating a neuronal network consisting of 80,000 neurons connected via 300 million synapses at a previously unattainable speed. For their implementation, they used a prototype of IBM’s INC-3000 FPGA-based "neural" supercomputer (Frontiers in Neuroscience, DOI:10.3389/fnins.2021.728460). These publications will form the basis for future research in this field at Jülich.
from JSC News No. 291, October 2022