July 27, 2021 Source: TU Dresden TU Dresden, University of Manchester, Racyics GmbH and Globalfoundries (GF) today announced the tapeout of the SpiNNaker2 chip – a new artificial intelligence chip inspired by the human brain. Based on the SpiNNaker infrastructure developed by the University of Manchester, the hybrid AI architecture realized at the TU Dresden, the adaptive body-bias IP platform ABX by Racyics and using the 22FDX solution by Globalfoundries, SpiNNaker2 is a neuromorphic AI Real-time processor with unprecedented efficiency and sub-millisecond latency for event-based systems. SpiNNaker2 is designed to scale up to a cloud system with 70,000 chips under real-time operation, making it a key building block for real-time AI at extremely high data rates. SpiNNaker2 will have a disruptive impact on AI applications such as smart cities, 5G, tactile internet and autonomous driving because no other current hardware meets the simultaneous requirements of low latency, high throughput and energy efficiency. The complete 10 million processor machine SpiNNaker2, called “SpiNNcloud,” is used at the TU Dresden for research purposes. At the same time, the startup “SpiNNcloud Systems GmbH” wants to make SpiNNaker2 commercially available. Artificial intelligence is having an ever-increasing impact on our daily lives. However, current AI hardware and algorithms are only partially inspired by the human brain. Even the best AI hardware currently is a far cry from the 20-watt power consumption, low latency, and unparalleled high-throughput processing power offered by the human brain. SpiNNaker2 is a multiprocessor system for event-based real-time processing of both classic AI and future event-based calculation paradigms inspired by the brain’s holistic, efficiency-optimized and demand-oriented processing. SpiNNaker2 was developed as part of the EU research project “Human Brain Project.” The backbone of the system is a bespoke, lightweight communications infrastructure that scales from edge devices to large server cloud applications without suffering the inherent slowdown typically associated with scaling computations to cloud size. Prof. Steve Furber (University of Manchester): “There is still a lot to learn from biology if we want to exploit the full potential of AI in the future. SpiNNaker2 is designed to bridge the gap between realistic brain models and AI so that both can increasingly benefit from each other.” With the SpiNNaker2 chip, a SpiNNcloud with 70,000 chips and 10 million cores is to be set up at the TU Dresden, which, together with local initiatives such as the CeTI Cluster of Excellence and other national and international partners, is to research applications ranging from autonomous driving to data processing in real time in a smart city, tactile internet applications to biomedical processing. “Our holistic, economy-optimized system design, combined with a unique hybrid AI framework, enables real-time AI on an unprecedented scale. We will use the SpiNNaker2 machine at the TU Dresden to conduct cutting-edge research on novel machine learning, but we are also proud to be able to offer this unique AI commercially through our startup SpiNNcloud Systems,” explains Prof. Christian Mayr, Chair of Highly Parallel VLSI -Systems and Neuromicroelectronics at the Faculty of Electrical Engineering and Information Technology at the TU Dresden. SpiNNaker2 was realized in the 22FDX ecosystem using Racyics GmbH’s Adaptive Body Biasing IP (ABX) platform to leverage the full potential of 22FDX. This IP platform enables SpiNNaker2 to operate at the minimum power point of the processing units at 0.50V with additional support for dynamic voltage and frequency scaling for high performance on demand. SpiNNaker2 thus uses the advanced power-saving properties of 22FDX, Globalfoundries Dresden’s flagship technology. The complete chip design was performed by the TU Dresden team using makeChip, a design environment for 22FDX provided by Racyics. This approach was crucial for the academic group, 22FDX™, GlobalFoundrie’s unique 22nm FD-SOI technology, proves to be the ideal basis for modern AI chips directly inspired by the human brain with its worldwide unique low power consumption and optimized transistor performance. GF’s 22FDX™ technology, manufactured in Dresden, enables highly efficient neuromorphic real-time calculations with a latency of less than one millisecond, as required for AI applications such as smart cities, 5G, tactile internet, autonomous driving. The 22FDX-based SpiNNaker2 chip shows the synergies that are possible when European cutting-edge AI research meets European semiconductor technology: For the first time, the existing gap between AI optimized purely for numerical performance and real brain-inspired AI is closed. “As the world’s leading specialty foundry, Globalfoundries enables world-class teams to achieve commercial breakthroughs. SpiNNaker2 is tangible proof of the outstanding performance achieved by combining the latest advances in brain-inspired artificial intelligence and our high-performance 22FDX technology,” says Dr. Manfred Horstmann, head of Globalfoundries Dresden. The tapeout was financially supported by ERDF and the Free State of Saxony in the ‘SpiNNcloud’ project (application number: 100373652). We thank ARM for IP use. Information for journalists: Globalfoundries Karin Rath’s email: [email protected] TU Dresden Prof. Christian Mayr Email: [email protected] University of Manchester Jordan Kenny Email: [email protected] Racyics GmbH Holger Eisenreich Email: [email protected] SpiNNcloud Systems GmbH Christian Eichhorn Email: [email protected]
