The initial processor does the AI ​​in the water


Editor of the website for technological innovation – 05.10.2022

An ion processor, formed by hundreds of transistors in a liquid medium.
[Imagem: Woo-Bin Jung/Harvard SEAS]

initial calculation

Comparing the brain to computers is almost inevitable, but there are fundamental differences between the two “hardware”: processors are made of silicon and other solid-state semiconductors, while the brain processes information by manipulating ions in an aqueous medium.

This inspired Woo-Bin Jung and colleagues from Harvard University in the US to create a single processor in a liquid solution.

Although ions in water move more slowly than electrons in semiconductors, scientists believe that the diversity of ionic species, each with its own physical and chemical properties, may offer new opportunities for richer and more varied information processing.

Several teams had already built ion transistors, but Jung and his colleagues went much further, not only developing a complete ion circuit, composed of hundreds of chained initiation transistors, but also demonstrating that it was born to perform typical artificial intelligence calculations, so-called networks. artificial neural.

“Microprocessors digitally manipulate electrons to perform matrix multiplication,” said Professor Donhee Ham. “While our induction circuit may not be as fast or precise as digital microprocessors, multiplying the electrochemical matrix in the enchanted water itself has the potential to be energy efficient.”

computer and

The processor is essentially a hardware neural network, composed of a dense array of tanks that control the pH at a local level.
[Imagem: Donhee Ham Research Group/Harvard SEAS]

initial neural network

The ion transistor developed by the team consists of an aqueous solution of quinone molecules, which is in contact with two concentric ring-shaped electrodes and a disc-shaped central electrode.

By capturing and releasing hydrogen ions, the two ring electrodes electrochemically lower and adjust the local pH around the central disk. When a voltage is applied to the central disk, an electrochemical reaction is generated that produces an ionic current from the disk to the water.

The reaction rate can be accelerated or slowed down by adjusting the local pH, causing the initial current to increase or decrease. In other words, the pH controls the ionic current of the disk in aqueous solution, creating the ionic counterpart of an electronic transistor—in fact, the entire system already functions as a logic gate, ready for calculations.

Finally, the team made and wired the ionic transistors in such a way that the disk current is the arithmetic product of the disk voltage and a “weight” parameter, which represents the transistor’s local pH. These transistors are arranged in a 16 x 16 matrix, extending the analog arithmetic multiplication of individual transistors into analog matrix multiplication, with a matrix of local pH values ​​serving as a weight matrix like those found in neural networks.

“Matrix multiplication is the most widespread computation in neural networks for artificial intelligence,” Jung said. “Our ion circuit performs matrix multiplication in water in an analogous way, based entirely on electrochemical machinery.”


The team plans to incorporate new types of ions to “enrich” the computation.
[Imagem: Woo-Bin Jung/Harvard SEAS]

Enrich the initial calculation

The team already has plans to advance early computing.

“Until now, we have only used 3 or 4 ionic species, such as hydrogen ions and quinones, to enable ion transport and prevent or enable the signal in the water ion transistor,” Jung said. “It will be very interesting to use multiple different ion species and see how we can use them to enrich the content of the information that needs to be processed.”


Article: Water analog MAC machine
Authors: Woo-Bin Jung, Han Sae Jung, Jun Wang, Henry Hinton, Maxime Fournier, Adrian Horgan, Xavier Godron, Robert Nicol, Donhee Ham
Journal: Advanced materials
DOI: 10.1002/adma.202205096

Follow the tech innovation page on Google News

Other news about:

more topics

Leave a Reply

Your email address will not be published. Required fields are marked *