Computers Inspired by the Brain
Paving the way to a sustainable AI with neuromorphic chipsCopyright: © Forschungszentrum Jülich / Kurt Steinhausen
They could be the key to meeting the energy hunger of artificial intelligence (AI): neuroinspired computer chips modeled on the human brain. Researchers at RWTH Aachen University and Forschungszentrum Jülich are advancing the technology in the NEUROTEC project and the NeuroSys Cluster.
At the Jülich-Aachen Neuromorphic Computing Day, the researchers, together with partners from high-tech and startup companies, gave the around 200 guests from science, industry, and politics an insight into the current state of developments.
“We want to drive structural change in the Rhenish mining area through research and innovation, securing value creation and employment in the region. The NEUROTEC project shows how this can be achieved: By building a strong network of science and industry, we develop the microelectronics of tomorrow and accelerate the transfer to business and industry. With the help of NEUROTEC and the NeuroSys Cluster, the Aachen region is to become one of the leading hubs for AI hardware made in Europe. Neuromorphic chips, such as those being developed in NEUROTEC and NeuroSys, can significantly reduce the resources required for AI in the future. However, successful structural change can only succeed if we also manage to transfer scientific results into useful applications. NEUROTEC and NeuroSys offer optimal conditions for this, forming a new, innovative AI hub in the Rhenish mining area,” said the State Secretary in the Federal Ministry of Education and Research (BMBF), Prof. Sabine Döring, at the start of the networking event.
For years now, data centers have been consuming more and more electricity. The use of AI technologies is considered to be one of the key drivers of this development. ChatGPT and similar AI tools are increasingly exhibiting human-like capabilities. But the computers running these applications work quite differently from the human brain. The difference is also reflected in the energy consumption: Training artificial neural networks on supercomputers consumes a lot of power; it requires almost as much energy as is required to power a small city. The human brain, on the other hand, requires no more energy than a 25-watt light bulb.
“Neuromorphic systems modeled on the human brain make it possible to train and run such AI processes much more efficiently – by several orders of magnitude – than is possible with conventional digital computers,” explains Professor Astrid Lambrecht, Chair of the Board of Forschungszentrum Jülich. This technology represents a key technology.
"Neuromorphic computing will open up numerous possible uses for energy-efficient AI applications in areas such as speech, image, and video processing, but also in medicine. RWTH also “thinks big” in this respect, offering its wealth of experience in the translation of research results. Together with associations such as the Aachen Chamber of Industry and Commerce and the Zukunftsagentur Rheinisches Revier, various scenarios are being considered for the further development of our innovation ecosystem, including the establishment of a factory for computer chips that are ‘made in North Rhine-Westphalia’,” adds RWTH Rector Ulrich Rüdiger.
Memristive Devices as Artificial Synapses
The key elements for the construction of neuromorphic computers are memristive materials and components – RWTH and Forschungszentrum Jülich are world leaders in this technology. Working as artificial synapses in neuronal networks, memristive devices can store and process information in parallel.
In the NEUROTEC project, a novel basic technology for neuromorphic AI is being developed in collaboration with high-tech companies in the region, such as the semiconductor specialist AIXTRON, the measurement technology provider aixACCT Systems GmbH, and the nanotechnology companies AMO GmbH and SURFACE systems+technology GmbH & Co. KG. The joint project entered its second phase in 2021 and will be funded by the BMBF for five years with 36 million euros from structural change funds.
“The first results are already coming to market in the next generation of automotive electronics. This involves energy-saving, fast ReRAM memories from Infineon and TSMC that can store information permanently without requiring continuous power supply. These components were optimized with the help of test and programming procedures from NEUROTEC,” explains NEUROTEC coordinator Prof. Rainer Waser, head of the Chair of Materials in Electrical Engineering II at RWTH and director of the Peter Grünberg Institute for Electronic Materials (PGI-7) at Forschungszentrum Jülich. The main goal of the project – to implement the first neuromorphic demo chips for practical use –is now coming within reach. “We have already extensively experimented with chips based on 180-nm semiconductor technology and have made many small, but significant advances. For example, the question is how reliably these new types of components can be manufactured using CMOS technology, and whether they will continue to function faultlessly after many switching cycles,” says Waser.
The first designs based on the more advanced 28 nm technology went into production in Taiwan at the end of May. After further processing and assembly with memristive elements from Jülich and Aachen, these will soon be tested in the first real-world applications.
A Globally Leading Hub
In the NeuroSys cluster, researchers from RWTH and Forschungszentrum Jülich are working together with regional companies to open up further market opportunities and fields of application. The NeuroSys Future Cluster is part of the Clusters4Future initiative, funded by the German Federal Ministry of Education and Research. The goal is to develop novel hardware for AI applications and establish the region as a global leader in this field.
”Our research spans the entire value chain, including sociological and ethical issues raised by disruptive technologies. This offers opportunities on many technological levels for regional companies and startups seeking to accelerate structural change. We are already seeing the first positive effects on the labor market and on corporate investment. In the long term, semiconductor manufacturing in the region would ideally complement the outstanding, internationally recognized expertise and infrastructure and further increase the attractiveness of the region for the exceptionally well-trained researchers from Forschungszentrum and RWTH,” says NeuroSys coordinator Prof. Max Lemme, holder of the Chair of Electronic Devices at RWTH Aachen University and managing director of AMO GmbH.