Researchers are working on a water-based microprocessor. This microprocessor may one day be used as a more versatile alternative to current wafer architectures for applications ranging from AI to DNA synthesis and perhaps beyond.

The chip in question is still in the prototype stage, so a processor with built-in water cooling is not yet expected, but the mechanics are truly exciting. The chip uses a technology called ionics. Ionics involves manipulating different ionic species in a liquid, as opposed to the standard electrons that pass through semiconductors today. [The technology may be slower than current semiconductor architectures, but scientists believe that "the diversity of ionic species with different physical and chemical properties will enable richer and more diverse information processing." [This is according to a post from Harvard University's John A. Paulson School of Engineering and Applied Sciences (SEAS) (opens in new tab). In collaboration with the startup company DNA Script (opens in new tab), SEAS researchers recently published how they combined ion diodes and transistors to develop an ion circuit. [While a simple 16 x 16 array is a far cry from the 100 billion transistors found in Intel's 12th generation wafers or the 114 billion transistors power of Apple's M1 Ultra, the researchers have successfully implemented the core process of neural net computing.

They used this array to "extend the analog arithmetic multiplication of individual transistors to analog matrix multiplication." Yeah, I was pretty lost here too. Thankfully, Woo-Bin Jung, a postdoctoral fellow at SEAS, explained that "matrix multiplication is the most common computation in neural networks for artificial intelligence." He says that their "ionic circuits perform matrix multiplication underwater in an analog way, based entirely on electrochemical machinery. [The transistors operate by manipulating local PH values, allowing the researchers to control the chip's weight matrix similarly to a neural network.

In addition to potential applications such as DNA synthesis (open in new tab) and acceleration of other neural network processes, Donhee Ham, professor of electrical engineering and applied physics at SEAS, says that "electrochemical matrix multiplication in water is attractive in its own right and has the potential to become energy efficient It has the potential to be energy efficient," he says.

So far, only "three or four ionic species, such as hydrogen and quinone ions," have been tested, Jung notes, but as researchers move on to test other ionic species, information processing will become richer and more diverse.

One day, neural networks may run from water-based ion chips. The energy efficiency will be much higher, but the operating speed will be much slower.