Researchers from the Massachusetts Institute of Technology (MIT) have demonstrated a novel device designed to process information by mimicking the neural structure of the brain,media outlet Neowin reported. In essence, MIT’s “brain on the chip” is smaller than a piece of paper, but it holds tens of thousands of silicon-based components called memrhey elements that mimic the synapses of information in the human brain. This “brain on the chip” is part of a large family of neuromorphic devices that draw inspiration from the brain’s synapses to perform complex computational tasks.
Existing memory resistors are designed to work well when voltage stimulates the flow of large numbers of ions from one electrode to another — a large conduction channel. However, these designs lack reliability in thin conduction channels. The team from the Massachusetts Institute of Technology has worked to solve this particular area of work. In order to develop this novel design, they borrowed a key concept from metallurgy, which basically refers to the fact that the alloy has different physical properties than the metal it makes up.
Inspired by this, the researchers combined silver with copper to make a positive electrode for membrane resistance and used silicon to make its negative electrodes. This clever design selection allows ions to be transmitted consistently and reliably along thin conduction channels.
They clamped two electrodes around the amorphous silicon medium. In this way, they patterned a millimeter-squaresilicon chip with tens of thousands of memritis elements.
When the nerve-deformation device is ready, the researchers used it to “remember and reproduce” grayscale images of Captain America shields. To do this, they equate each pixel in the image with the corresponding memri element in the chip, and then modulate the conduction of each memri element, which is as strong as the color in the corresponding pixel.
In this test, neuromorphic chips performed better than chips made of other materials. They also did image processing tasks. Once again, however, the device was able to go beyond the competitive memory resistor design, reliably and effectively sharpening and blurring images from MIT’s Killian Court.
The team’s findings have been published in the journal Nature.