A petri dish with a shiny, golden disc inside is the first thing that catches the eye on the lab bench. Postdoc researcher Saurabh Soni points to the golden disc surface, where a couple of dozen irregular pieces of glass are attached. ‘This round silicon wafer is coated with a very thin layer of gold, with these glass pieces glued on top,’ Soni explains. ‘By removing such a piece, the gold sticks to the glass surface and is removed from the silicon.’
In a special, oxygen- and water-free glove cabinet, he carefully removes one of the glass pieces using tweezers. Then he submerges the gold-plated piece in a solution containing special molecules. ‘These so-called HATNA molecules now attach as a very thin layer to the gold, they are the information carriers in our new system,’ he explains ‘Now they are attached to the gold, we can test the flow and behavior of electrical currents through the HATNA molecules and compare these to ion flows that occur in the brain.’
Globally, as much as five to ten percent of all energy consumed is used for data transfer, storage and analysis, and this amount is steadily increasing. To process vast amounts of data, recognizing patterns in internet surfing, moving a self-driving car or to run ‘smart’ household appliances consume more and more energy.
In contrast to the energy-consuming data centers, the brain has an extremely energy-efficient way of processing and transferring information. Using very short-lasting electrical pulses, information is moved from one brain cell to the next. Due to this efficient system, the brain only uses a fraction of the energy that a supercomputer needs. According to Nijhuis, the reason is that computers function by a continuous flow of electrons, that transforms the energy carriers, the bits of the computer, into ‘zero’s’ and ‘ones’. In contrast, the brain only uses very brief electrical pulses, that result in ion flows. In addition, only the pathways needed for the particular information flow are used.
‘Brain easily outperforms any computer that exists today’
‘Despite the really low energy consumption, the brain easily outperforms any computer that exists today, especially when it comes to complex tasks, like filtering information and recognizing images and speech,’ Nijhuis says. ‘When somebody calls your name in a full, noisy stadium, you will easily pick it out among all other noise. The same is true for recognizing, for example, animals in a picture.’
Research to find a system that can be controlled by electrical pulses, and transfer information, similar to the processes in the brain, would be a breakthrough in saving energy. But to design such a system, detailed knowledge how the brain processes and transfers information is essential.
Brain cells are connected and communicate to each other through so-called synapses. Via these synapses, the information can pass from one brain cell to the other. ‘Information transfer consist of two steps and starts with a sudden and quick wave of electrical discharge over the synapse’s membrane of the sending cell,’ Nijhuis explains. ‘This wave is driven by ion flows, not electrons. This is a so-called action potential.’ This results in the relatively slow second step: the release of a messenger molecule, the neurotransmitter, from the sending cell, that diffuses through the synaptic cleft to the connecting brain cell, where it binds to receptors. This may trigger another action potential transferring the signal further to the next brain cell.
Nijhuis and his team managed to copy this brain system surprisingly close, including a kind of quick electrical pulse, combined with a slower step, to mimic the time-dependent switching of synapses.
The basis of their brain-inspired system is their specially built HATNA molecule, that serves as the synaptic cleft and information carrier. The team placed the molecule between two electrodes, mimicking two brain cells. ‘Similar to brain cells, one of the electrodes sent out small electrical pulses, that resulted in the quick transfer of electrons to the HATNA molecule, making it negatively charged,’ Nijhuis explains. ‘To keep the charge neutral, and remain stable, protons, H+ ions, coming from water that is present in the air, diffuse slowly and attach to the HATNA molecule, similar to the diffusion of neurotransmitters.’
The essence of the finding is that in two steps the molecule has changed into a new version that also carries different energy levels and thus different information. The same process can be repeated several times, and each time, the system creates new molecules, representing new information, that is dependent on the number and timing of previous voltage pulses. Nijhuis: ‘This is some sort of memory or even learning, since the number of new molecules formed depend on the ones that were previously generated.’
The team successfully established the proof of principle of a new way to transfer and process information in the lab, by creating self-learning molecules that are very energy-efficient in processing information.
The scientists are now focusing on new and real-life applications, and designing new systems to use it. They see the future of these findings in many applications containing artificial intelligence, like in smart sensors, robots composed of compliant materials (soft robotics), and completely new technologies requiring some sort of learning or memory. For example, the technology might prove invaluable for use in sensors in self-driving cars, where a lot of information has to be processed. With the current technologies, this would consume way too much battery capacity, but using the brain-inspired information processing, such technologies are much more feasible. Also, since these molecules are biologically compatible, they can be applied in electronic implants.
‘Smart molecules will open doors to improve current technologies’
Nijhuis is realistic where the method will fit into a modern world: ‘Smart molecules that process information in a brain-like way will have their place next to existing technologies. We will not replace current computer technologies. But the team will also continue its search for other ‘smart’ molecules that could be applied in new materials and maybe have a better performance. One thing is sure: thanks to their minimal energy use, smart molecules will open doors to improve current technologies and make completely new technologies possible.’