Processors continue to improve as manufacturers find new ways to upgrade them. One of the latest upgrades comes in the form of a 32-bit bendable microprocessor that can run machine learning models, even when bent to the extreme. The real kicker about these bendable processors, though, is the fact they cost less than a dollar to make.
The new microprocessor is based on the RISC-V open standard but is known as Flex-RV, thanks to its highly adaptable design. It’s also made of an entirely different type of material called indium gallium zinc oxide (IGZO), which is used in place of the more traditional silicon found in processors.
While the bendable microprocessor can run machine learning models, it won’t deliver the most fantastic performance seen from a processor of its nature. But that’s because it only has 12,600 logic gates and a maximum clock speed of 60 kHz (roughly 0.00006 GHz for those more familiar with PC processors).
Despite those somewhat middling performance specs, the chip is able to be successfully integrated into hardware as a low-power machine learning accelerator. But the manufacturer behind the chip—Pragmatic—never intended it to train AI models like GPT-4. Instead, the chip was designed to power disposable medical devices on the edge of the frontier of new medical-based gadgets like improved health wearables, soft robotics, and even brain-computer interfaces.
This kind of bendable microprocessor could indeed prove very efficient for powering small medical devices, like the 3D-printed medical devices scientists have been testing printing directly in the human body. With such low power requirements, it could also work perfectly alongside the world’s first biomedical processor, which was made using human brain tissue.
The bendable microprocessor also maintained its accuracy, even when bent to a 5mm radius curve—making it ideal for usage in small devices that might require steep bending to fit the processor within them. The researchers responsible for the new processor’s design shared complete details about it in a new paper published in Nature last month.
