By Michael Irving
June 20, 2022
Facebook
Twitter
Flipboard
LinkedIn
![](https://image.dost-dongnai.gov.vn/webtiengviet/modular-optical-computer-chip.PNG)
MIT engineers have developed a new modular computer chip
that uses flashes of light to communicate between components Photo credit: MIT
VIEW 1 IMAGES
The electronics market has gotten to a point where many
consumers will turn over a smartphone every year for a newer, shinier, slightly
better model. Upgrading individual parts isn’t really an option for many
devices, requiring the whole thing to be replaced. That’s not the most
environmentally responsible attitude to electronics.
Modularity could go a long way, allowing users to swap in
new or improved functionality, like bigger batteries or upgraded cameras. So
for the new study, the MIT team has now demonstrated this approach within a
single computer chip.
The team’s modular chip is made up of layered components
like artificial intelligence, processors and sensors, which can be stacked or
swapped in to build a chip to perform specific functions as needed, or upgrade
it as new technology becomes available.
“You can add as many computing layers and sensors as you
want, such as for light, pressure, and even smell,” says Jihoon Kang, an author
of the study. “We call this a LEGO-like reconfigurable AI chip because it has
unlimited expandability depending on the combination of layers.”
But perhaps the most impressive thing is how the layers of
this chip interact with each other. Modular electronics face a problem in
getting new and old components to communicate with each other in a fast and
simple way. The MIT chip, however, uses flashes of light to convey information
between each layer.
The team fitted each layered component of the chip with
LEDs and photodetectors that line up with those of the next component. When one
part needs to communicate with another, it flashes its LED pixels in a certain
pattern that encodes the data, which the photodetectors of the receiving layer
can interpret.
To demonstrate this design, the team created a chip
measuring 4 mm2, made up of three computing layers. Each layer contained an
image sensor, an optical communications system, and an artificial synapse array
which was designed to recognize a specific letter – M, I or T.
To test it out, the researchers exposed the chip to
pixelated images of random letters, then measured the strength of the
electrical current that each array produced in response. The stronger the
current, the better the array recognizes the letter.
Using this process, the team found that the chip was able
to classify images of letters it was trained on very well if the images were
clear, but less so when blurry. To demonstrate the modularity of the chip, the
engineers then slotted in a “denoising” processor that could handle the blurry
images better, and sure enough the chip’s letter recognition improved.
“We showed stackability, replaceability, and the ability to
insert a new function into the chip,” said Min-Kyu Song, an author of the
study.
The team plans to apply the technique to “edge computing
devices,” which are small, specialized sensors for the Internet of Things.
The research was published in the journal Nature
Electronics.
Source: MIT