Researchers at MIT have designed a modular chip that uses flashes of light to convey information between its components. One of the design goals of the chip is to enable people to swap in new or improved functionality instead of replacing the whole chip, in essence paving the way for perpetually upgradeable devices. “The general direction of reusing hardware is a blessed one,” Dr. Eyal Cohen, CEO and co-founder of CogniFiber, told Lifewire over email. “We genuinely hope that such a chip will be usable and scalable.”
Light Years Ahead
The MIT researchers have put their plan into action by designing a chip for basic image-recognition tasks, currently trained specifically to recognize three letters: M, I, and T. They’ve published the details of the chip in the Nature Electronics journal. In the paper, the researchers note that their modular chip is made up of several components, like artificial intelligence, sensors, and processors. These are spread across different layers and can be stacked or swapped in as required to assemble the chip. The researchers argue the design enables them to reconfigure a chip for specific functions or upgrade to a newer, improved component as and when it becomes available. While this chip isn’t the first to use a modular design, it’s unique for its use of LEDs as the means of communication between the layers. Used together with photodetectors, the researchers note that instead of conventional wiring, their chip uses flashes of light to convey information between the components. The lack of wiring is what enables the chip to be reconfigured, as the different layers can be easily rearranged. For instance, the researchers note in the paper that the first version of the chip correctly classified each letter when the source image was clear but had trouble distinguishing between the letters I and T in certain blurry images. To correct this, the researchers simply swapped out the chip’s processing layer for a better denoising processor, which improved its ability to read blurry images. “You can add as many computing layers and sensors as you want, such as for light, pressure, and even smell,” Jihoon Kang, one of the researchers, told MIT news. “We call this a LEGO-like reconfigurable AI chip because it has unlimited expandability depending on the combination of layers.”
Reducing E-waste
Although the researchers have only demonstrated the reconfigurable approach within a single computer chip, they argue the approach could be scaled, allowing people to swap in new or improved functionality, like bigger batteries or upgraded cameras, which could also help reduce e-waste. “We can add layers to a cellphone’s camera so it could recognize more complex images, or make these into healthcare monitors that can be embedded in wearable electronic skin,” Chanyeo Choi, another researcher, told MIT news. Before they can be commercialized, however, the chip design will need to address two key issues, suggested Dr. Cohen, whose Cognifiber is building glass-based chips to bring server-grade processing power to smart devices. For starters, the researchers will have to look at the interface quality, particularly over fast transmission and across multiple wavelengths. The second issue that needs to be further analyzed is the robustness of the design, especially when the chips are used for a long duration. Do they need tight temperature control? Are they sensitive to vibrations? These are just two of the many questions that’ll need to be explored further, explained Dr. Cohen. In the paper, the researchers note that they’re eager to apply the design to smart devices and edge computing hardware, including sensors and processing skills inside a self-sufficient device. “As we enter the era of the internet of things based on sensor networks, demand for multifunctioning edge-computing devices will expand dramatically,” Jeehwan Kim, another researcher and MIT’s associate professor of mechanical engineering, told MIT News. “Our proposed hardware architecture will provide high versatility of edge computing in the future.”
