The future of neural network computing may be more challenging than we anticipated.
A team of physicists has succeeded in developing an ionic circuit – a processor that relies on the motions of charged atoms and molecules in an aqueous solution, rather than electrons in a solid semiconductor.
Since this is closer to the way the brain transmits information, they say, their devices could be the next step forward in brain-like computing.
“Ion circuits in aqueous solutions seek to use ions as charge carriers for signal processing,” write the team Led by physicist Woo-Bin Jung of Harvard John A. Paulson School of Engineering and Applied Sciences (SEAS) in a new research paper.
“Here, we report an aqueous ion circuit…This demonstration of a functional ion circuit capable of analog computing is a step toward more complex aqueous ions.”
A large part of the transmission of signals in the brain is movement called charged particles ions through a liquid medium. Although it is difficult to reproduce the brain’s amazing processing power, scientists thought a system similar to computing could be used: pushing ions through an aqueous solution.
This would be slower than traditional silicon-based computing, but it could have some interesting advantages.
For example, ions can be created from a wide variety of molecules, each with different properties that can be exploited in different ways.
But first, scientists need to prove that it can work.
This is what Young and his colleagues have been working on. The first step was to design a functional ion transistor, a device that switches or boosts a signal. Their latest advance involved combining hundreds of those transistors to work together as an ionic circuit.
The transistor consists of a “pulse” arrangement of electrodes, with a small disc-shaped electrode in the center and two concentric toroidal electrodes around it. This interferes with the aqueous solution of quinone molecules.
A voltage applied to the central disk generates a stream of hydrogen ions in the quinone solution. Meanwhile, the two annular electrodes adjust the pH of the gate solution, increasing or decreasing the ionic current.
This transistor physically multiplies a “weight” parameter set by the toroidal pair gates by the disc voltage, producing an answer as the ionic current.
However, neural networks rely heavily on a mathematical process called matrix multiplicationwhich includes multiple multiplication operations.
So, the team designed a 16×16 matrix of their transistors, each capable of arithmetic multiplication, to produce an ionic circuit that can perform matrix multiplication.
Matrix multiplication is the most common computation in neural networks for Artificial intelligence” Jung says. “Our ion circuitry strikes the matrix in water in an analog fashion that relies entirely on electrochemical machinery.”
There are, of course, significant limitations to the technology. The 16 currents could not be resolved separately, which meant that the process had to be done sequentially rather than synchronously, which slowed down an already relatively slow technology.
However, its success is a step towards more complex ionic computing: only by seeing the problem can we find the solutions.
The next step will be to introduce a wider range of molecules into the system to see if that allows the circuit to process more complex information.
“So far, we have only used 3 to 4 ion species, such as hydrogen and quinone ions, to enable ion transport and gate in an aqueous ion transistor,” Jung says.
“It would be very interesting to employ more diverse ionic species and see how we can exploit them to enrich the contents of the information to be processed.”
The team notes that the ultimate goal is not to compete with electronics or replace them with ions, but to complement them, perhaps in the form of hybrid technology with the capabilities of both.
The search was published in advanced materials.