Eta Compute Launches with Ultra-Low-Power Machine Learning Platform for Edge Devices
November 05, 2018
Eta Compute has developed a machine-learning SoC that includes autonomous learning. The TENSAI device performs ultra-low power image classification, keyword spotting, and wakeup word detection.
Eta Compute has developed a machine-learning SoC that includes autonomous learning. The TENSAI device performs ultra-low power image classification, keyword spotting, and wakeup word detection. In fact, this permits execution on battery-powered devices.
The TENSAI chip includes the third generation of the company’s delay insensitive logic, which enables products to reliably operate at a very low supply voltage. And by using the company’s own kernel for spiking neural network (SNN) and CNN, operations and power consumption can be lowered even more.
The SoC can perform autonomous learning of speech, images, and other data where classification occurs on the data without labels. The image classification application consumes just 0.4 mJ per picture. And the always-on wakeup word application consumes only 500 uA during classification or 50 uA during silence.
Eta Compute’s custom kernel further optimizes the trained model using a tightly integrated DSP and microcontroller architecture. This solution can support a wide array of applications in audio, video, and signal processing where power is a key attribute.
The device is sampling now and mass production is expected in the first quarter of next year.