Going where few imagined AI would go — on a solar-powered chip
A Swiss R&D organization says it has an AI system on a chip that will cut operating costs and electricity consumption and better protect sensitive data, like biometrics.
The hardware and software can operate either on coin cell battery for months or via solar power, according to the Swiss Center for Electronics and Microtechnology, or CSEM.
Some AI subtasks can be jumped over if not needed, cutting power consumption and operating costs. For example, according to CSEM, in a facial recognition role, the system’s first accelerator fields fundamental questions: Are their faces in the image? Are they wearing masks?
If masks are not detected, the chip does not engage the second accelerator, a convolutional neural network that in other systems might fire up automatically and waste power.
One of the researchers also notes the solar SoC’s potential for voice biometrics.
And simplifying the design by putting everything on a single chip, including and e-paper monitor should make manufacturing more streamlined, lowering costs for buyers.
And the chip becomes an edge device, reducing the need to transact with cloud systems. That can be expected to increase system performance and lower the risk of data theft, according to the company.
Designed in modules, the company says, the system can be customized for applications needing real-time signal and image processing.
CSEM is applying miniaturization to other biometric missions. It has worked with Global ID, for instance, to shrink Global ID’s identification-card chip scanners.
Edge AI chipsets are expected to make up a market worth over $50 billion by 2025, with edge facial recognition accounting for $2.3 billion.