Maxim MAX78000 MCU, iZip VWW combination
Adding human motion detection capabilities to low-power IoT systems


Maxim Integrated Korea announced on the 20th that it has applied its neural network MCU, 'MAX78000', to the 'Visual Wake Words (VWW)' model of Aizip, an artificial intelligence (AI) specialist company for Internet of Things (IoT) applications.

Low-power networks extend the operating time of battery-powered IoT systems for building energy management and human detection, such as smart security cameras. The MAX78000 low-power neural network acceleration-based MCU improves the operating time of battery-powered edge AI applications by executing AI inference using less than 1/100th the energy of existing software solutions.

The VWW network with mixed precision technology is a product of the Aizip Intelligent Vision Deep Neural Network (AIV DNN) series for image and video applications. It was developed using Aizip's own design automation tool and has achieved over 85% human recognition accuracy.

The combination of a low-power MCU system-on-chip (SoC) and a high-efficiency AI model enables human recognition with 0.7 mJ of energy per inference, 100 times lower than existing software and IoT human recognition solutions. It also enables 13 million inferences on a single AA/LR6 battery. Extreme model compression enables accurate smart vision with a low-cost AI-accelerated MCU with limited memory and an affordable image sensor.

“iZip has leveraged their per-layer quantization capabilities to reduce storage weight and implement a compact, energy-efficient model for human recognition,” said Robert Muchsel, senior research engineer at Maxim Integrated, who designed the MAX78000 MCU. “We look forward to working with iZip on other projects in the future.”