ESAB

Overview

The goal of the "Event-based Supercomputer for AI and Beyond" (ESAB) project is to expand the HPC infrastructure at Forschungszentrum Jülich by adding a novel, energy-efficient, event-based neuromorphic computing module for real-time artificial intelligence (AI) workloads, large-scale neuroscience simulations, and other application areas such as quantum computer emulation, and to make this resource available to the wider scientific community. On 30 April 2026, the ESAB project officially kicked off at the JSC. The project is funded by the European Regional Development Fund (EFRE) and the State of North Rhine-Westphalia (NRW) under the 'EFRE/JTF-Programm NRW 2021–2027' program and will run from 2026 to 2028.

Neuromorphic Computing

Neuromorphic computing is an innovative approach that mimics the brain’s network of neurons and synapses. It utilizes Spiking Neural Networks (SNNs), asynchronous, event-based communication, and compute-in-memory, creating a brain-like computing paradigm that promises energy efficiency and novelty.

The SpiNNaker2 System at JSC

SpiNNaker (short for spiking neural network architecture) is a massively parallel, many-core supercomputer designed to simulate the neural networks of the human brain in real time. A two-rack SpiNNaker2 system, called JUBIS (JUelich Brain-Inspired Supercomputer), was installed at the JSC in April 2026. The system comprises 90 SpiNNaker2 48-chip boards with 4320 SpiNNker2 chips and 656,640 ARM cores with neuromorphic accelerators. The SpiNNaker2 architecture is designed to be highly scalable, versatile, and adaptable to various applications.

Project Objectives and Scope

The ESAB project is located at the Jülich Supercomputing Centre (JSC), and we collaborate with various institutes at the Forschungszentrum Jülich (IAS-6, IET-1, INM-1, INM-9, PGI-4, and PGI-15) and external partners (TU Dresden, University of Sussex). The main research and development objectives of the ESAB project include:

1. System benchmarking
2. Application prototyping
3. Real-time, large-scale simulations of biologically realistic spiking neural networks in neuroscience
4. Data-driven research on the microstructural organization of the brain
5. Rethinking algorithms and developing novel methods to make AI and non-AI workloads accessible to NC technology, for example, simulations in the field of biomolecular dynamics
6. Integration of NC into the HPC landscape
7. Closed workflows that combine HPC, Quantum Computing and NC

The main objective of the EASB project is to establish a novel, brain-inspired computing infrastructure accessible to a broad scientific community, providing a technical basis for research into energy-efficient NC and AI.

History and Background

The ESAB project builds on a two-decade history of research and development in neuromorphic computing.

SpiNNaker was initiated in 2005 at the University of Manchester under the leadership of Professor Steve Furber, with the aim to support the simulation of biologically realistic neural networks in real time and that way contribute to neuroscience research. The SpiNNaker architecture was designed for massively parallel compute using low-power ARM mobile phone processors. In 2018, a one-million-core machine was installed at Manchester University, which is part of the EBRAINS infrastructure today.

Strong impulses were given by the "Advanced Computing Architectures" (ACA) project, which ran from November 2018 to October 2022. Funded by the Helmholtz Association, ACA was 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. More than 50 peer-reviewed articles have resulted from the interdisciplinary project with researchers from Forschungszentrum Jülich, RWTH Aachen University, the University of Manchester, and Heidelberg University. The project's key results include a feasibility study, the development of brain connectivity concepts, test cases, and a benchmarking framework.

Key Contributions and Impact

The integration of the SpiNNaker2 technology into the Jülich High-Performance Computing infrastructure will:

• Provide a cutting-edge computing resource for AI workloads, large-scale, real-time neuroscience simulations and other applications.
• Enable the research of energy-efficient, event-based computing with potentially orders of magnitude better performance than conventional architectures can deliver.
• Support inter-disciplinary collaborations and knowledge-sharing within the scientific community.
• Facilitate innovation and drive the research on neuromorphic computing forward.

Software Development

In order to promote system usability and user acceptance, it is essential to provide interfaces, tools, and frameworks for different application scenarios, as well as convenient, user-friendly access to the system. Therefore, a key aspect of the ESAB project is the development of software. The main objectives of the project include:

• The provision of a robust, modular software environment that supports a wide range of application scenarios.
• The integration into the existing High-Performance Computing landscape and tools.
• The development and implementation of a usage concept.

Role and Responsibilities at the JSC

The ESAB project is located at the JSC and was initiated and is led by the Simulation and Data Lab Neuroscience (SDLN). The main tasks of the JSC include:

• Hosting and maintaining the SpiNNaker2 system within the supercomputing ecosystem at JSC.
• Provide training and technical assistance.
• Developing and maintaining software tools for use with the SpiNNaker2 technology.
• Fostering collaborations in the highly interdisciplinary field of neuromorphic computing.

Long-term Goal

The integration of the SpiNNaker2 computing module into JSC's Modular Supercomputing Architecture (MSA) represents a long-term goal. By establishing an energy-efficient computing resource, the project aims not only to accelerate and enable groundbreaking research across AI, neuroscience, and beyond but also to pave the way for future advancements in these fields and drive the research on neuromorphic computing forward.