Reduction in energy consumption and latency
Efficiently implement various battery-powered AI
Maxim Integrated Korea launched the 'MAX78000', a low-power microcontroller equipped with a neural network accelerator, on the 8th.

The MAX78000 rapidly executes complex artificial intelligence (AI) inference and makes complex decisions on battery-powered IoT embedded devices using less than one-hundredth of the low-power energy of software solutions. It also extends the runtime of AI-based battery applications and makes previously impossible battery-powered AI use cases a reality. Consequently, latency can be reduced 100-fold for only a fraction of the cost of FGPA or GPU solutions.

Special hardware designed to minimize energy and latency in the MAX78000 integrates a dedicated neural network accelerator to enable seeing and hearing complex patterns with low-power, local AI processing running in real time. This hardware increases efficiency by ensuring minimal interference regardless of the type of microcontroller, so energy and time are consumed only for the mathematical operations that implement the CNN, and ultra-low power ARM Cortex-M4 cores or lower power RISC-V cores are used to secure external data to the CNN engine.

Maxim provides tools for AI development. By including ready-to-use demonstration programs for large-scale vocabulary keyword detection and facial recognition, and documenting all sources, engineers can train on the MAX78000 in TensorFlow or PyTorch tools.

Kris Ardis, Senior Director of Micro, Security, and Software at Maxim, stated, “By cutting the wires for AI at the edge, battery-powered IoT devices can do a variety of things beyond simple keyword detection,” adding, “We expect that innovative technologies that have changed the landscape of general power, latency, and cost will usher in a new era of applications.”