Peter Bobbert from the group M2N and the CCER is co-author of the publication A deep-learning approach to realizing functionality in nanoelectronic devices in Nature Nanotechnology, Oct. 2020.
In this publication the complex electronic behavior of nanodevices consisting of dopant atoms in silicon contacted by electrodes is modeled with a Deep Neural Network (DNN). These devices can be used for energy-efficient reconfigurable logic. A desired logic functionality with specific settings of the electrode voltages is searched for within the DNN model, after which the functionality is realized by applying the found voltage settings to the physical device. This approach to realizing functionality is orders of magnitude faster than a direct search for functionality in the physical device. The approach can in principle be applied to finding functionality in any complex physical system. The work was carried out at the Center for Brain-Inspired Nano Systems (BRAINS) at the University of Twente, to which Peter Bobbert is attached as part-time professor.