Maxim cube brings AI to the edge

  • July 28, 2021
  • Steve Rogerson

Californian electronics company Maxim Integrated has unveiled a camera cube reference design that demonstrates how artificial intelligence (AI) applications previously limited to high-power machines can be embedded in space-constrained, battery-powered edge devices.

The Max Refdes 178# lets low-power IoT devices implement hearing and vision and showcases the Max 78000 low-power microcontroller with neural network accelerator for audio and video inferences. The system also contains the Max 32666 low-power Bluetooth microcontroller and two Max 9867 audio codecs.

The entire system comes in a compact form factor to show how AI applications such as facial identification and keyword recognition can be embedded in low-power, cost-sensitive applications such as wearables and IoT devices.

“Machine learning promises a lot: that machines can make sense of what they see and hear like humans, as well as make more autonomous decisions,” said Kris Ardis, executive director at Maxim Integrated. “Until the Max 78000, the embedded world was left behind because you couldn’t implement AI at the edge in a power, cost and size constrained manner. Now the Max Refdes 178# demonstrates how meaningful and powerful AI inferences can be run at the edge, on even the smallest and most energy-conscious devices.”

AI applications require intensive computations, usually performed in the cloud or in expensive, power-hungry processors that can only fit in applications with big power budgets such as self-driving cars. But the camera cube demonstrates how AI can live on a low-power budget, enabling applications that are time- and safety-critical to operate on even the smallest of batteries.

The Max 78000’s AI accelerator slashes the power of AI inferences up to 1000 times for vision and hearing applications, as compared with embedded alternatives. The AI inferences also show latency improvements, running more than 100 times faster than on an embedded microcontroller.

The cube measures 41 by 44 by 39mm, showing that AI can be implemented in wearables and other space-constrained IoT applications. The Max 78000 itself is up to 50 per cent smaller than the next-smallest GPU-based processor and does not require other components such as memories or complex power supplies to implement cost-effective AI inferences.

“The next big opportunity in AI is providing machine-learning insights at the edge,” said Alan Descoins, CTO at machine-learning consultancy Tryolabs. “The Max Refdes 178# shows how Maxim Integrated’s AI is a breakthrough in power, latency and size that can unlock the possibilities for AI in battery-powered designs.”