A breakthrough at the University of Twente is bringing new brain-like computers one step closer. An international group of researchers led by Professor Christian Nijhuis has developed a new type of molecular switch that can learn from previous behaviour. The group published their results today in the scientific journal Nature Materials. Nijhuis: “These molecules learn in the same way our brains do.”
Computers, data centres and other electronics use vast amounts of energy. To meet that energy demand, we are now building enormous wind farms. But according to Christian Nijhuis, we can also turn our attention to making our electronics more efficient. “Our brains are the most efficient computers we have. They use ten thousand times less energy than the most economical electronic computers”, Nijhuis says.
This is because our brains process data in a completely different way. Whereas computers process binary information streams – ones and zeros – our brains are analogue, using time-dependent pulses. “Our brains are able to effortlessly process input from millions of nerve cells, from all our senses. Because unlike traditional electronics, they only use the brain cells and synapses that are actually transmitting pulses”, says Nijhuis. This means that the brain only consumes energy during transmission, allowing it to process a lot of data at once much more efficiently.
Hardware for artificial intelligence
The molecules Nijhuis and his team developed can perform all the Boolean logic gate operations required for deep learning. “Deep learning is a form of machine learning based on artificial neural networks. It’s widely used in the automatic recognition of images and speech, as well as in the search for new medicines. Recently, it has also been used to create art. These are all things that are much more difficult for a computer than they are for our brain”, says Nijhuis. Researchers are making great strides in the field of AI software, but these molecules are now bringing AI hardware closer as well.
To mimic the dynamic behaviour of synapses at the molecular level, the researchers combined fast electron transfer with slow proton coupling limited by diffusion. This resembles the fast pulses and slow uptake of neurotransmitters by the neurons in your brain. The molecules can adjust the intensity and duration of the pulses, demonstrating a form of classical conditioning as they adapt their behaviour based on the stimuli they previously received. In other words, they are able to learn. In the future, molecules like these may also respond to other stimuli, such as light.
Many new applications
This breakthrough will enable the development of a whole new range of customisable and reconfigurable systems. These in turn could lead to new multifunctional and adaptive systems that would significantly simplify artificial neural networks. “This would drastically reduce the energy consumption of our electronics”, according to Nijhuis. Meanwhile, multifunctional molecules that are also light-sensitive or that can detect other molecules could help create new types of neural networks or sensors.
Christian Nijhuis helms the Hybrid Materials for Opto-Electronics research group (Faculty of Science and Technology), which is part of UT’s MESA+ Institute for Nanotechnology. He also serves as principal investigator for the Computing Molecules & (Opto)Electronics research area at the MESA+ Molecules Centre. This research was conducted in collaboration with Damien Thompson, professor of Molecular Modelling and director of the Science Foundation Ireland Research Centre for Pharmaceuticals at the University of Limerick, and Enrique del Barco, Pegasus professor at the University of Central Florida.
The research paper, titled ‘Dynamic molecular switches with hysteretic negative differential conductance emulating synaptic behaviour’, was published in the scientific journal Nature Materials. Nature Materials is one of the top 3 journals in the fields of chemistry, physics and materials science. The paper is available online.