Machine learning algorithms are inspired by the brain and used in pattern recognition. The brain is more efficient than computers in performing cognitive tasks. Emulating the function of synapses and neurons in the brain can lead to more efficient and environmentally friendly electronic devices. However, current machine learning algorithms still rely on conventional computing hardware.
Mimicking human brain's behavior
Unlike traditional computer chips, which rely on separate components for memory and processing, the neurAM project envisions a material that combines the functionalities of both the hard drive and the processor. The "neurAM" project aims to create materials with inherent properties resembling the neuronal structure of the brain: namely materials with plasticity that are self-adaptive. To find such materials, the researchers study the interactions between atoms in two-dimensional arrays. They search for configurations of interacting atoms that can be used to perform specific computational tasks, i.e. recognize patterns in input stimuli, perform logical tasks and have memory. The material assembly and characterization are done using a scanning tunneling microscope. This state-of-the-art technique allows for observing and analyzing atoms and electronic properties of materials, which provide essential information on their behavior. “With this grant, we will bridge boundaries between materials science and machine learning to create arrangements of single atoms that interact like neurons and synapses in the brain and thus understand the physics needed for smarter materials.”
Energy efficient
The unique and innovative approach, inspired by the remarkable efficiency of the human brain, has the potential to vastly improve energy efficiency in electronic devices while enabling advanced cognitive capabilities. The key challenge for the neurAM project lies in identifying a material capable of autonomously exhibiting these brain-like properties without the need for additional circuitry. “Our novel approach to designing and studying materials is to assemble single atoms on a surface with sub-nanometer precision and study the emerging behaviour of this ensemble as we add more atoms to it. This gives us full control in the design process and thus the possibility to study the basic physics needed to create matter with brain-like functionality”, researcher Osterhage explains.
Research team
Dr. Hermann Osterhage, dr. Susanne Baumann (Institute for Functional Matter and Quantum Technologies of the University of Stuttgart) and dr. Eduardo Domínguez Vázquez (Donders Institute) perfectly combined the expertise in the different research fields of single atoms and their orbital, magnetic dynamics, and machine learning. “We combine our expertise to study the interactions of atoms and describe these interactions mathematically with machine learning models. We want to learn how to tweak the interactions, such that the atoms behave like a computing algorithm.”
Volkswagen foundation
The Volkswagen foundation is dedicated to the support of the humanities and social sciences as well as science and technology in higher education and research. It funds research projects in path-breaking areas and helps academic institutions for the improvement of the structural conditions for their work. The Volkswagen Stiftung is the largest private foundation supporting research in Germany. The main goal of the call is to stimulate research on neuromorphic computing, for developing new materials that inherently mimic functions found in biology, or that are useful for neuromorphic microchips and devices.
We warmly congratulate the research team on the grant!
