neuromorphic computing
neuromorphic computing

Detecting waves with single atoms for neuromorphic computing

With the increase of the energy consumption and data production at the same time, it is required to find new ways to store and process data faster, at higher density and with less energy. Unconventional techniques such as neuromorphic (brain-inspired) hardware could perform processing tasks faster and more energy-efficiently than existing computing systems. Realizing this type of hardware requires innovative use of special material properties, for example using the fluctuating dynamics of individual atoms on a surface. Researchers at Radboud University's Institute for Molecules and Materials (IMM) discovered that a single atom can detect the frequency of these waves and reflect this information in its behavior. This breakthrough could lead to advancing the development of brain-like and efficient computers.

The basis for the studies is a single cobalt or iron atom on the surface of black phosphorus. Research from 2018 demonstrated that by applying voltage to the atom, it can be induced to "fire", shuttling between a value of 0 and 1 randomly, resembling the function of a neuron. When multiple atoms are assembled together, they function as an artificial neural network similar to those used in artificial intelligence:  a so-called Boltzmann machine.

Individual atoms detecting waveforms

In this project, the goal was to study how the “firing” of individual atoms responds to more complex input signals, namely sinusoidal waves. The team, led by Professor Alex Khajetoorians, tested how individual atoms respond to different input frequencies and used a scanning tunneling microscope to see how these atoms switch between different states when they receive these wave-like inputs. While increasing the frequency of the waveform, they found that iron atoms react differently to low frequencies compared to high frequencies. In other words, the iron atom’s behavior is sensitive to the input frequency. “We have discovered how to manipulate the favorability of the atom’s “0” and “1” state by adjusting the frequency of the oscillating voltage”, IMM researcher Werner van Weerdenburg says. At the same time, the dynamics of the atom synchronizes to the input signal. 

Experimentalists and theorists working together 

Collaboration with the machine learning group led by Professor Bert Kappen at the Donders Institute facilitated understanding the unique responses of Fe and Co atoms to dynamical waveforms. Their  models successfully captured the synchronization effect and distinct frequency responses of iron and cobalt atoms, as observed in the experiment. This alignment between experimental findings and theoretical predictions elucidates the differences in behavior between the two types of atoms. It also enables prediction of their frequency response based on their firing rate reaction to applied voltages. Insights gained from this study reveal methods to modify the atomic system's frequency response, including amplification, attenuation, or complete deactivation. It stresses the versatility of implementing neuromorphic functionalities using individual atoms.

Brain-inspired computing

The gained fundamental knowledge could potentially lead to create new types of brain-inspired hardware, which could be more efficient and adaptable than current technologies. The last year has seen a large increase in the availability and use of AI software, such as ChatGPT. While these tools are very helpful, they also use a lot of energy, especially compared to the energy consumption of our brains. New hardware with neuromorphic architectures, such as the atomic Boltzmann machine, could drastically reduce the energy consumption of AI technology. The frequency response found in this study means that an individual atom can classify input signals with low and high frequencies. “In the future, we can apply waveform input signals to larger ensembles of atoms. This way, we can explore the functionality of the atomic Boltzmann machine and potentially perform tasks such as voice recognition”, Van Weerdenburg concludes.

TOC figure
Literature reference

Stochastic Syncing in Sinusoidally Driven Atomic Orbital Memory
Werner M. J. van Weerdenburg, Hermann Osterhage, Ruben Christianen, Kira Junghans, Eduardo Domínguez, Hilbert J. Kappen, and Alexander Ako Khajetoorians
ACS Nano (2024)
Stochastic Syncing in Sinusoidally Driven Atomic Orbital Memory | ACS Nano

Contact information

Werner van Weerdenburg, w.vanweerdenburg [at] ru.nl (w[dot]vanweerdenburg[at]ru[dot]nl)
Hermann Osterhage, h.osterhage [at] science.ru.nl (h[dot]osterhage[at]science[dot]ru[dot]nl)

Contact
W.M.J. van Weerdenburg (Werner) , Dr H.U. Osterhage (Hermann) , Prof. A.A. Khajetoorians (Alex) , Prof. H.J. Kappen (Bert)
About person
W.M.J. van Weerdenburg (Werner) , Dr H.U. Osterhage (Hermann) , Prof. A.A. Khajetoorians (Alex) , Prof. H.J. Kappen (Bert)
Theme
Sustainability, Innovation, Molecules and materials, Laws of nature