Supercomputer - inspired by the human brain

The system developed by Prof. Christian Mayr (Chair of Highly Parallel VLSI Systems and Neuromicroelectronics at TUD) is based on the 'SpiNNaker2' chip and currently comprises 35,000 chips and over five million processor cores - an important step in the development of energy-efficient AI systems.

So-called 'neuromorphic computer systems' are based on nature's most powerful computer - the human brain. This opens up completely new perspectives for data center architecture: instead of relying exclusively on improvements to existing technologies, this approach expands the design of computer architectures to include brain-like principles such as distributed memory and event-driven processing. The result: significantly reduced energy consumption combined with high performance and flexibility.

SpiNNaker2 was developed under the leadership of TUD Prof. Mayr as part of the EU flagship project 'Human Brain Project'. "SpiNNaker2 combines high efficiency with real-time processing at latencies below one millisecond," explains Mayr. "Inspired by biological principles such as plasticity and dynamic reconfigurability, the system automatically adapts to complex, changing environments. This combination of biologically inspired architecture and technological innovation opens up new possibilities for AI applications in smart cities, autonomous driving and the tactile internet."

This supercomputer is part of the AI competence center ScaDS.AI Dresden/Leipzig, which closes the gap between the efficient use of mass data, knowledge management and very advanced AI by expanding and bundling regional big data competencies. As one of nine centers of the National High Performance Computing (NHR), the Center for Information Services and High Performance Computing (ZIH) at TUD also offers special HPC resources as well as targeted support and advice. The systems are available to scientists from all over Germany.

"TU Dresden is setting standards with its computing infrastructure - from powerful HPC systems to innovative architectures such as SpiNNaker2, which complement each other perfectly: In addition to classic HPC resources for high-performance simulations and data-driven analyses, SpiNNaker2 offers a unique platform for energy-efficient, real-time capable AI models. This combination creates an exceptional technological basis for further pushing the boundaries of science," explains ZIH and ScaDS.AI Director Prof. Wolfgang E. Nagel.

Prof. Steve Furber, pioneer of the original SpiNNaker architecture at the University of Manchester, underlines the importance of the further development: "With more than five times the number of neurons compared to the first SpiNNaker generation in Manchester, the SpiNNcloud at TU Dresden marks a real breakthrough in neuromorphic computing. For the first time, we can use this energy-efficient, brain-inspired architecture on a truly large scale."

Hector Gonzalez, co-founder and CEO of TUD spin-off SpiNNcloud, explains: "Our vision is to redefine the future of artificial intelligence through brain-inspired systems. Large-scale implementations like this represent an important milestone for AI and enable new ways of deploying extremely efficient state-of-the-art models, especially in today's world where mainstream AI increasingly relies on dynamic, sparse algorithms to cope with global energy shortages." The project benefits from the expertise of several industrial partners: RAFI Group is responsible for the production of the highly complex circuit boards, while Cloud & Heat provides the advanced water cooling for the infrastructure. The adaptive body bias IP platform is provided by technology partner Racyics.

ScaDS.AI and SpiNNcloud

As one of five AI centers in Germany, ScaDS.AI is funded as part of the German government's AI strategy and by the Free State of Saxony. ScaDS.AI combines the expertise of TU Dresden, Leipzig University and twelve partner institutions in an international team with more than 60 senior scientists and over 200 employees. Its research closes the gap between the efficient use of big data, knowledge management and AI and drives innovation.

SpiNNcloud is a deep-tech spin-off of the TUD. The company offers the world's first commercial, brain-inspired HPC platform that redefines traditional computing architectures through energy-efficient parallel processing and real-time capabilities. This technology enables customers to develop energy-efficient and robust AI models and process generic compute workloads.

The realization of the project was made possible by the financial support of the European Regional Development Fund (ERDF) and the Free State of Saxony in the 'SpiNNcloud' project. The software stack used in such a large-scale system was financed by the EIC transition project 'SpiNNode' and developed by SpiNNcloud.