Maxim, as an edge device-based AI feature reference
MAXREFDES178# Camera Cube Design Announcement
MAX78000, 100x AI inference power compared to similar MCUs


AI applications are typically executed on high-cost processors with high power consumption, which has limited their implementation on edge devices. On the 21st, Maxim Integrated Korea announced 'MAXREFDES178#', a camera cube reference design that presents a method for implementing AI applications on battery-operated edge devices in confined spaces.

The MAXREFDES178# reference design enables auditory and visual functions in ultra-low-power IoT devices. It includes a 'MAX78000' low-power MCU with a neural network accelerator for acoustic and visual inference, a 'MAX32666' ultra-low-power Bluetooth MCU, and two 'MAX9867' audio codecs. Additionally, it is provided in an ultra-small form factor of 41×44×39mm.

The MAX78000 AI accelerator reduces power consumption for AI inference functions in auditory and visual applications by up to 1,000 times compared to existing embedded solutions. AI inference running on the MAXREFDES178# is more than 100 times faster than embedded MCUs with significantly improved latency. In addition, it is up to 50% smaller than the closest GPU-based processor and does not require memory or complex power supplies, enabling low-cost AI inference.

"Going forward, opportunities in the AI field lie in delivering machine learning insights on edge devices," said Alan Descoins, CTO of TryoLabs. "Maxim's Camera Cube reference design is an AI solution that has achieved innovation in power, latency, and size, expanding the AI potential of battery-operated devices."

“Machine learning enables more autonomous decision-making, as well as allowing machines to see and hear like humans,” said Kris Ardis, Senior Director of the Micro, Security & Software business unit at Maxim Integrated. “Before the launch of the MAX78000, implementing AI on edge devices was impossible, but with the MAXREFDES178#, powerful AI inference has become possible even at the edge, including on ultra-small devices sensitive to energy efficiency.”