MIT’s new computer chip design lets you clip on parts like LEGOs
Chips are in everything: smartphones, supercomputers, remote-sensing robots. MIT engineers have created an electronics chip design that allows sensors and processors to be swapped out or added to, much like LEGO bricks. This modular, reconfigurable chip could be used to upgrade smartphones, computers and other devices, without generating as much waste. It could also be used for artificial intelligence applications. Their paper describing the tech was published this week in the journal Nature Electronics.
Here is how the chip works. It has alternating layers that allow for processing and sensing. Instead of copper wires, the layers communicate with each other internally using optical signals. These two features allow elements from different layers to be interchangeable with each other.
“As the internet of things based upon sensor networks enters the era, the demand for multifunctioning, edge-computing devices with multiple functions will increase dramatically,” Jeehwan Kim (associate professor of mechanical engineering at MIT) stated in a press statement . “Our proposed hardware architecture will allow for high versatility in edge computing in future.” (Edge computing is electronics that can process data independently of connecting to a central server.
To test how the chip performs on simple tasks, the team made a prototype with image sensors, LEDs, and a processor containing “artificial brain synapses“—components made of silicon, silver, and copper that mimic how the brain transmits information (the team also calls these memristors). Instead of transmitting binary information (as 0 and 1), the strength the memristors’ electrical current depends upon the strength the incoming current. It can therefore have a range based on the strength of the signals. It also remembers which value corresponds to what strength of the signal, so calculations remain constant. These neurons could be connected to form an array that can process and classify signals directly on-chip.
Researchers trained a version the stacked chip to recognize letters M, I and T. (For MIT. The chip was equipped with photodetectors to receive the visual signal. It then passed it down onto other layers that encoded the image in a sequence LED pixels and classified the signal according to the strength of the incoming light. Researchers used laser light to illuminate different letters on the chip. It was able to recognize which letter was being given. However, it did better with brighter and clearer images. The researchers later added a “denoising processor” to the chip that allowed it to understand blurry images better.
The team envisions that this modular capability allows them to add features such as image recognition to smartphones cameras or health monitoring sensors for electronic skins.
” We can make a general chip platform and each layer could sell separately like a game,” Jeehwan Korea said. “We could make different types neural networks, such as for image or voice recognition and let the customer pick what they want and add to an existing chips like a LEGO .”