Modern microprocessors, the heart of modern smartphones and computers, process information by controlling the flow of electrons moving through solid semiconductor materials. However, living brains, which are biological computers, work on other principles, in which information is processed by moving ions in a liquid medium that fills the volume of neurons and synapses.

Various groups of scientists have been trying to develop principles of ion computation in a liquid medium for some time now. Of course, ions are much more massive and move much slower than electrons in semiconductors. However, solutions may contain ions of a large number of various chemical elements with different physical and chemical properties; this would allow more complex information processing to be implemented in a single traditional time.

The field of ion computing is currently in its infancy. In several laboratories, primitive devices, ionic analogues of diodes and transistors have already been created, but until recently no one has managed to combine many such devices in one functioning circuit.

But recently, researchers from Harvard University’s School of Engineering and Applied Sciences (SEAS) and biotechnology company DNA Script have created the first ionic circuit that contains hundreds of ionic transistors and functions as a primitive neural network.

The basis of the new device was an ion transistor with a very unusual structure. It consists of two electrodes connected to concentric ring electrodes and a central disc-shaped electrode. All this is placed in an aqueous solution with quinone molecules. Two annular electrodes that trap hydrogen ions in their traps regulate the pH level on either side of the disc electrode. If we do not go into the details of the electrochemistry of the processes that take place, we can say that the difference between the pH levels determines the amount of ionic current passing through the device, making it an ionic analogue of an electronic transistor.

Ion transistors were built in a 16 by 16 matrix, and their connection made it possible to apply the analog arithmetic matrix multiplication and matrix multiplication functions of weight coefficients used in the construction of traditional neural networks.

“Matrix multiplication is the most common type of computation in neural networks, the basis of artificial intelligence,” the scientists write. they cannot yet provide the speed and accuracy of digital microprocessors, but they are performed without any program involvement and require incomparably less energy.

In the future, the scientists plan to significantly increase the “chemical complexity” of the ion computing system. “Until now, we have only used 3-4 types of ions, including hydrogen ions and quinone ions, to control and transfer charges inside the ion transistor,” the scientists write. run and perform more complex parallel processing”. Source