Water-Based Chips Could be Breakthrough for Neural Networking, AI
Scientists have followed nature’s unique design in developing ion microprocessors that can prove particularly energy efficient compared to traditional semiconductor-based processors.
as posted in advanced materialsis a team of researchers at the John A. Paulson School of Engineering and Applied Sciences (SEAS) at Harvard University working with biotechnology start-ups. DNA scriptHave Developed an ion circuit consisting of hundreds of ion transistorsI was even able to run the core processes of neural net computing in it.
The design of the ion processor aims to take what we’ve learned from biological processing systems (especially the brain) and create a processor that uses electrochemistry instead of electricity to perform computations. Studies paint ion processors as a technology that is likely to be developed and expanded in the future. Its energy efficient design can be of value in certain deployment scenarios.
The technology is still in its early stages, as researchers have just implemented the first circuit boards containing hundreds of actual ion transistors. So far, only single ion transistors have been shown. Therefore, this work aims to create real processors by operating hundreds, thousands, or even millions of ion transistors in tandem. open the way
The researcher’s ionic circuit was created by increasing the number of single ion transistors that can work together. Their ion transistor design consists of an aqueous solution of quinone molecules, connected to two concentric ring electrodes (blue and red) and his third (yellow) central disk electrode, essentially a bullseye-like transistor design form the
By applying a voltage to the transistor, two concentric electrodes can locally adjust the pH of the water by increasing or decreasing the amount of hydrogen ions present in the water. A feat of electrochemistry, this change allows the transistor’s ionic current to be used as an on and off switch. Gate, with the transistors we are used to hearing. This gating of ionic currents by changing the pH of the transistor unlocks the ability of the transistor to process binary information.
The researchers further adjusted the design of the microprocessor, placing these analog transistors (which can represent 0 or 1) in a 16 x 16 matrix grid array. This allows ion processors to perform matrix multiplication tasks, bringing them closer to the functionality of neural networks and increasing their value in artificial intelligence processing scenarios that may require very specific performance/power balance requirements. rice field.
“Matrix multiplication is the most common computation in artificial intelligence neural networks,” said Woo-Bin Jung, postdoctoral researcher at SEAS and lead author of the paper. “Our ionic circuit performs matrix multiplication in water in a completely electrochemical-mechanical analog manner.”
The trade-off is that they are slow, but compared to light, they are all slow. I would like to continue the development of the ion processor by making it possible. Researchers look forward to programming additional functions into these systems.
This, coupled with the increased number of available transistors, should provide performance benefits while opening up the Ionic processor to a wider variety of tasks and real-world specific or general-purpose computing.
“So far, we have only used three to four ionic species, such as hydrogen and quinone ions, to enable gating and ion transport in aqueous ion transistors,” says Jung. “It will be very interesting to take more diverse ion species and see how they can be exploited to enrich the information content processed.”