Earlier this year, Xilinx (San Jose, Calif.) launched its Kria line of adaptive system-on-modules (SOMs) for a host of AI applications, including in factory and healthcare environments. In terms of the former, the SOMs target digital twin, predictive maintenance and defect detection applications.
SOMs, which are small embedded boards about the size of a credit card, enable the abstraction of hardware functionality. As a result, developers can design at the board level rather than the chip level. For hardware designers, SOMs promise to avoid rudimentary design work. SOMs also enable software developers to begin work in parallel with a hardware team.
Smart factory applications related to vision AI are a core focus area for the first product in the Xilinx SOM portfolio, the Kria K26 SOM.
As a result, factory owners deploying SOMs can get smart factory projects up and running faster than in the past, according to Xilinx.
In terms of manufacturing applications of SOMs, smart vision applications in factories are a natural fit. Pharmaceutical manufacturers can benefit from SOMs’ machine vision capabilities, said Chetan Khona, director, industrial, vision, healthcare and sciences at Xilinx. “There’s a lot of commonality in manufacturing, whether you are making pills or food-and-beverage items,” Khona said. While medical facilities have higher regulatory requirements, computer vision holds promise for defect detection.
“You can train an AI model to say, ‘This is what a good medication looks like, or potentially what a bad medication looks like,” Khona said.
Factory operators can also use SOMs for predictive maintenance applications. A SOM could detect a vibration correlated with a bearing failure, for instance. Detecting early signs of bearing failure could give operators tools to limit unexpected downtime.
“We also see the use of digital twins, where you’ve got a digital representation of the physical factory in the cloud,” Khona said. Technologies like SOMs enable a growing amount of processing for digital twins to happen at the network edge.
SOMs also hold some relevance for clinical trials, given their ability to support diagnostics in radiology, in particular.
As the footprint of edge computing grows in clinical environments, so does the potential of edge computing in general and SOMs in particular, said Subh Bhattacharya, lead, healthcare and sciences, ISM at Xilinx. “Managed care services in healthcare are playing a similar role to the process automation services offered by [industrial vendors],” Bhattacharya said. The technology in such environments is “getting mixed into non-standard, but very similar objective environments,” he continued. “Through the layers, from us to our customers to their customers, the challenges are similar, and we are all trying to address similar problems.”