Researchers from ETH Zurich, the University of Zurich and Empa have developed a new material for an electronic component that can be used in a wider range of applications than its predecessors. These components will help create electronic circuits that mimic the human brain and are more efficient at performing machine learning tasks.
Compared to computers, the human brain is incredibly energy efficient. Scientists are therefore inspired by the functioning of the brain and its interconnected neurons to inspire the design of innovative computer technologies. They predict that these brain-inspired computing systems will be more energy efficient than conventional systems, as well as more efficient at performing machine learning tasks.
Much like neurons, which are responsible for both storing and processing data in the brain, scientists want to combine storage and processing in a single type of electronic component, called a memristor. Their hope is that this will contribute to greater efficiency, since moving data between processor and storage, as conventional computers do, is the main reason for high power consumption in machine learning applications.
Researchers from ETH Zurich, the University of Zurich and Empa have now developed an innovative concept for a memristor that can be used in a much wider range of applications than existing memristors. “There are different operating modes for memristors, and it is advantageous to be able to use all these modes depending on the architecture of an artificial neural network,” explains Rohit John, postdoc at ETH. “But previous conventional memristors had to be configured in advance for one of these modes.” The Zurich researchers’ new memristors can now easily switch between two modes of operation during use: a mode in which the signal weakens over time and dies (volatile mode), and a mode in which the signal remains constant (non-volatile mode). fashion).
As in the brain
“These two modes of functioning are also found in the human brain,” explains John. On the one hand, the stimuli at the synapses are transmitted from neuron to neuron by biochemical neurotransmitters. These stimuli start out strong and then gradually weaken. On the other hand, new synaptic connections with other neurons are formed in the brain as we learn. These connections last longer.
John, who is a postdoc in the group led by Prof. Maksym Kovalenko at ETH, was awarded an ETH Fellowship for Outstanding Postdoctoral Researchers in 2020. John conducted this research with Yiğit Demirağ, a PhD student in the group of Prof. Giacomo Indiveri at the Institute of Neuroinformatics of the University of Zurich and ETH Zurich.
Semiconductor known from solar cells
The memristors the researchers developed are made of perovskite halide nanocrystals, a semiconductor material known primarily for its use in photovoltaic cells. “The ‘nerve conduction’ in these novel memristors is mediated by the temporary or permanent chaining of silver ions from an electrode to form a nanofilament penetrating the perovskite structure through which current can flow,” explains Kovalenko.
This process can be regulated to make the filament of silver ions either thin, so that it gradually breaks down into individual silver ions (volatile mode), or thick and permanent (non-volatile mode). This is controlled by the intensity of the current conducted on the memristor: the application of a weak current activates the volatile mode, while a strong current activates the non-volatile mode.
New toolkit for neuroinformaticians
“To our knowledge, this is the first memristor that can be reliably switched between volatile and non-volatile modes on demand,” says Demirağ. This means that in the future, computer chips can be made with memristors that allow both modes. This is a significant advance because it is usually not possible to combine several different types of memristors on a single chip.
As part of the study they published in the journal Nature Communicationscall_made, the researchers tested 25 of these new memristors and performed 20,000 measurements with them. In this way, they were able to simulate a computational problem on a complex network. The problem was to classify a number of different neural spikes into one of four predefined patterns.
Before these memristors can be used in computing technology, they will need to undergo further optimization. However, such components are also important for neuroinformatics research, as Indiveri points out: “These components are closer to real neurons than the previous ones. As a result, they help researchers better test hypotheses in neuroinformatics and hopefully the, to better understand the computational principles of real neural circuits in humans and animals.”
Source of the story:
Material provided by ETH Zürich. Original written by Fabio Bergamin. Note: Content may be edited for style and length.
Comments are closed.