Development of neuromorphic systems requires competences at many different levels of organization. The consortium of this project covers most of these competences, ranging from basic neuroscience, system modeling, material science and circuit design, to system integration, system housing and community integration. It enables a close interaction between neuromorphic system development, neuroscience and applications. The close proximity between many of the involved partners, a feature which is clearly unique world-wide, strengthens the interaction between the work packages, fosters mutual education in formal seminars and personal meetings, and, ultimately, leads to more efficient and faster development cycles (integrative loops). Moreover, the consortium includes leading national and international experts in neuromorphic systems (BrainScaleS, SpiNNaker) with whom a trusted relationship has been build-up over the last decade in a sequence of projects funded by the European Union and other sources. The present project is not in competition to the neuromorphic approaches developed so far, but combines the competence to solve common fundamental problems. At the same time, it targets a new neuromorphic system for the specific niche of supporting brain research with an appropriate balance between flexibility and performance.
Researchers at INM-6 develop multi-scale models of the brain, combining data-driven development of brain theory with the bottom-up approach of directly simulated structured networks, and the top down approach, mapping functional models of higher brain function to spiking dynamics. In ACA, INM-6 delivers the neuroscience background for the project as well as the test cases which guide the devolopment of the new neuromorphic architectures.
The SimLab Neuroscience is an interdisciplinary team of scientists and engineers with complementary backgrounds and skills, dedicated to supporting neuroscientists in using high-performance computing and data resources for their research. Expertise in both neuroscience and HPC is based on in-house research and development, as well as collaborative joint projects with the Institute of Neuroscience and Medicine (INM), Jülich Research Centre, and other national and international partners. In ACA, the SimLab Neuroscience contributes with the development of a modeling language for spiking neuron and synapse models, and investigates novel neuromorphic system architecture approaches, such as hybrid soft- and hardware designs that combine a traditional von Neumann architecture with hardware acceleration units.
Researchers in this group work on exploring and developing HPC architectures and technologies for scientific applications, including applications from neuroscience. This work is done in close collaboration with relevant industrial partners through joint exascale labs. In the context of this project we work on performance analysis and modelling as well as on the integration of neuromorphic systems into existing HPC infrastructures.
The Institut für Elektronische Materialien (PGI-7) forms a union with the Institut für Werkstoffe der Elektrotechnik II (IWE II) at the RWTH Aachen, with common goals and complementary equipment. We focus on the physics and chemistry of electronic oxides, which are promising for potential memory, logic, energy conversion and sensor functions. In particular in the area of Neuromorphic Computing. Within ACA we will provide the integration technology and the basic knowledge about working and failure mechanisms of memristive devices required for their use in energy-efficient, bio-inspired, non-von-Neumann computing approaches. We will perform multi-scaling modelling of the dynamics of memristive devices and will provide compact models that can be employed for circuit design.
The mission of PGI-10 is to conduct research on concepts for a next-generation massively parallel and scalable neuroscience simulation platform, well suited for high-speed simulation of ultra-large network models. In ACA, PGI-10 contributes concepts for dedicated high-performance ODE-solvers well suited for hardware implementation as well as dedicated low-latency communication concepts. Particular architectures will be prototyped and evaluated on the Xcape neural supercomputer.
The Central Institute of Engineering, Electronics and Analytics – Electronic Systems (ZEA-2) is a scientific-technical institute of Jülich Research Centre GmbH and carries out R&D projects in cooperation with other institutes and external partners. The main focus is the development of electronic and information technology system solutions for complex challenges in sensor and detector technology, signal and data processing and measurement systems for the areas information technology, renewable energies and bio economy. The focus of ZEA-2 inside the ACA is the development of system models and implementation approaches for next generation NC system solutions combining scalable highest intra- and inter chip connectivity with minimized communication latency based on very large scale integration (VLSI). These should incorporate the best of existing solutions and advanced approaches inspired e.g., from modern communication systems.
The chair of Integrated Digital Systems (IDS) focusses on efficient hardware realizations suitable for high-throughput processing of modern-day algorithms. Within this project IDS leads the effort of developing functional models of the massively parallel hardware platform in systemC. This also includes circuit level design space exploration of architectural building blocks in nano-scale CMOS technologies considering emerging devices.
The Electronic Vision(s) Group at the Kirchhoff-Institute for Physics at Heidelberg University, Germany, has developed the accelerated analog neuromorphic computing system BrainScaleS, providing a configurable, physical emulation of spiking neural networks, running at 10000x real time. In ACA, the BrainScaleS team will research the resource efficiency of the BrainScaleS system for different benchhmark neural networks created in Jülich, starting with the cortical microcolumn model.
The SpiNNaker team in the Advanced Processor Technologies group at the University of Manchester, UK, have developed the worlds’s largest neuromorphic computing platform - SpiNNaker - comprising a million ARM processor cores and capable of simulating spiking neural networks of the scale of a mouse brain in biological real time. In ACA, the SpiNNaker team will improve the efficiency of the existing SpiNNaker implementation of the cortical microcolumn model developed in collaboration with Jülich and scale it up to encompass a larger area of cortical circuitry.