Many technical systems rely heavily on automatic pattern recognition, as for example autonomous driving. Currently, this is done by software that runs on traditional computer systems. Yet this solution is energy-hungry and cannot be scaled down as needed. Neuromorphic chips are about to change that, as they will be able to recognize patterns on their own, not unlike our brains, using only a fraction of the energy required by conventional systems. Katrin and Helmut Schultheiss from the Helmholtz-Zentrum Dresden-Rossendorf (HZDR) have found an innovative approach to this new technology. In partnership with an international research group, they have launched the EU-funded project NIMFEIA to develop a prototype for industrial production.
Many technical systems rely heavily on automatic pattern recognition, as for example autonomous driving. Currently, this is done by software that runs on traditional computer systems. Yet this solution is energy-hungry and cannot be scaled down as needed. Neuromorphic chips are about to change that, as they will be able to recognize patterns on their own, not unlike our brains, using only a fraction of the energy required by conventional systems. Katrin and Helmut Schultheiss from the Helmholtz-Zentrum Dresden-Rossendorf (HZDR) have found an innovative approach to this new technology. In partnership with an international research group, they have launched the EU-funded project NIMFEIA to develop a prototype for industrial production.
"The Von Neumann-architecture of traditional computers is not cut out to accomplish what a human brain can do,” explains Dr. Helmut Schultheiss, head of the Emmy Noether group "Magnonics” at the HZDR Institute of Ion Beam Physics and Materials Research. "That’s why we have research fields such as neuromorphic, unconventional, or brain-inspired computing. In these fields, scientists are developing technologies based on physical effects that can extract different information from data streams, just like our brains do.”
One way to achieve this is to recreate a brain using conventional computer technology. However, this approach is very resource-intensive, because our brain has about 100 billion switching points, the neurons, which are interconnected with countless lines, the synapses. A structure like that can hardly be reproduced on a silicon chip.
"That’s why we developed a completely new approach,” explains Dr. Katrin Schultheiss, who launched the project Nonlinear Magnons for Reservoir Computing in Reciprocal Space, aka NIMFEIA. She and her husband have been researching this topic since 2015. "We use nonlinear processes to generate magnetic waves in micrometer-sized magnetic disks. This is how we replicate the switching points in the brain.”
She and her research group discovered that information can be processed extremely efficiently within the disk via the interaction of different magnetic waves. "We have already conducted successful lab experiments to demonstrate that this setup is able to recognize patterns,” Helmut Schultheiss summarizes. "The novelty for us is that our research is geared more heavily towards industrial application, because we now want to prove that our idea is industrially viable and that we can develop a prototype on a standard wafer like the ones used in the chip industry.”
The project, which is coordinated by HZDR and funded by the EU to the tune of three million euros, is scheduled for a four-year period. The HZDR scientists work with colleagues from the Université Paris-Saclay, the French National Centre for Scientific Research CNRS, Johannes Gutenberg-University Mainz and Stichting Radboud University. The team also includes two industrial partners, Infineon Technologies Dresden GmbH & Co. KG and GlobalFoundries Dresden Module One LLC & Co. "We are particularly proud to have enlisted both Infineon and GlobalFoundries for our project,” says Katrin Schultheiss.
Katrin and Helmut Schultheiss are certain that their approach offers great potential for applications. Since it is so efficient at recognizing patterns while requiring very little energy, it could be used directly in the sensors of autonomous vehicles to measure distances and speed.
Contact
Dr. Katrin Schultheiss | Dr. Helmut Schultheiss HZDR Institute of Ion Beam Physics and Materials Research Phone: +49 351 260 2919 | 3243 Email: k.schultheiss@hzdr.de | h.schultheiss@hzdr.de