AI inferencing, Software Ecosystems and Trends in Industrial Automation

Engineering & simulationArm Ethos U854 times performance improvement, 20% more power efficient

The Ethos U85 accelerator enables transformer networks for video-based processing, allowing industrial applications like fault and defect detection on production lines to detect events more efficiently than previous convolutional network approaches.

“one of the early applications is enhanced video systems. So we're seeing it being deployed in industrial for applications like fault detection or defect detection on production lines…”

Engineering & simulationArm Virtual Hardware Platform10 X improvement in performance

Arm uses its virtual hardware platform to allow developers to recompile code for new silicon without waiting for physical chips, demonstrating a 10x performance improvement for transformer networks on the new Ethos U85.

“with our arm virtual hardware platform, we're able to see people, you know, recompiling the code they've already generated for the current generation U 55s and seeing in, you know, in the demo here, we're showing a 10 X improvement in performance…”

Engineering & simulationSilicon Labs XG26

Silicon Labs designed the XG26 chip with efficient machine learning capabilities to enable inference tasks, such as wake word detection, to run on battery-powered devices like coin cell batteries.

“our goal is to put inference on a battery, you know, that is not an easy thing to do, still, even today, and we're working our way, you know, down that path, but how would you do wake word detection on a coin cell battery?”

Engineering & simulationAnalog Devices (ADI) neural networks

ADI developed low-power neural networks that allow AI to run at the edge on battery power, enabling vision systems for robotics and asset health monitoring in retrofit factory environments.

“We have some of the lowest power, neural networks, in processing, which means you can do AI now at the edge in a battery power system. Okay. And we're showing some of that out there again in terms of vision systems for robotics.”