A team creates a technique for underwater neural networks.

Microprocessors in smartphones, computers, and data centers process information by manipulating electrons through solid semiconductors, but our brains have a different system. To process information, they control the ions in a liquid medium.

 

Researchers have long worked to create “ionics” in an aqueous solution, using inspiration from the brain. Scientists believe that the variety of ionic species with various physical and chemical properties could be utilised for richer and more diverse information processing, even though ions in water move more slowly than electrons in semiconductors.

 

However, ionic computing is still in its infancy. To date, laboratories have only created individual ionic devices like ionic diodes and transistors; nobody has, however, combined many of these devices into a more sophisticated circuit for computing.

 

In partnership with DNA Script, a biotech firm, a group of researchers at Harvard’s John A. Paulson School of Engineering and Applied Sciences (SEAS) constructed an ionic circuit with hundreds of ionic transistors and carried out a fundamental neural network computing operation.

 

Advanced Materials published the study.

The scientists started by utilising a method they recently invented to create a new kind of ionic transistor. The transistor is made out of an aqueous solution of quinone molecules that is connected to two concentric ring electrodes and a bullseye-shaped centre disc electrode. By creating and trapping hydrogen ions, the two ring electrodes electrochemically reduce and regulate the local pH around the centre disc. An ionic current flows from the centre disc into the water as a result of an electrochemical reaction when a voltage is applied to it. By adjusting the local pH, the reaction rate can be sped up or slowed down, increasing or decreasing the ionic current. In other words, the pH regulates (gates) the ionic current flowing through the disc in the aqueous solution, acting as the ionic equivalent of the transistor.

 

The pH-gated ionic transistor was then designed so that the disc current is a mathematical multiplication of the disc voltage and a “weight” parameter that represents the local pH gating the transistor. To convert the analogue arithmetic multiplication of individual transistors into an analogue matrix multiplication, they arranged these transistors into a 16 16 array, with the array of local pH values acting as the weight matrix found in neural networks.

 

The most frequent calculation in neural networks for artificial intelligence, according to Woo-Bin Jung, a postdoctoral fellow at SEAS and the paper’s first author, is matrix multiplication. Our ionic circuit completely relies on electrochemical machinery to conduct the matrix multiplication in water in an analogue fashion.

 

Donhee Ham, the Gordon McKay Professor of Electrical Engineering and Applied Physics at SEAS and the paper’s senior author, explained that microprocessors “manipulate electrons in a digital manner to execute matrix multiplication.” The electrochemical matrix multiplication in water is beautiful in its own right and has the potential to be energy efficient, even though our ionic circuit cannot be as quick or precise as digital microprocessors.

 

The team is currently working to increase the system’s chemical complexity.

 

The gating and ionic transport in the aqueous ionic transistor have so far only been made possible by 3 to 4 ionic species, such as quinone and hydrogen ions, according to Jung. More different ionic species will be used, and it will be interesting to see how we might use them to enrich the information being analysed.

 

Han Sae Jung, Jun Wang, Henry Hinton, Maxime Fournier, Adrian Horgan, Xavier Godron, and Robert Nicol all contributed to the study’s creation.