Multiple heterogeneous, large neural networks, such as YOLOv3, M2Det v can run on Ergo simultaneously with high accuracy, high performance, and ultra-low power. For example, Ergo can run YOLOv3 at up to 246 frames per second (batch size = 1) at 30 fps while using about 20 mW of power .
Ergo supports a wide variety of currently popular styles of neural network for video, audio, and other sensor processing, including CNNs (multiple sizes of convolutions and residual edges), RNNs, LSTMs, and others, multiple solutions for solving real-world problems. The chip doesn't need any external RAM and comes in a small, 7x7 mm package.
The company will support t he Ergo inference chip with reference boards and standard imaging and audio inferencing applications which can be tuned by customers . The Ergo chip and reference board are being sampled to leading customers and will be mass produced in Q2, 2020.
Perceive was incubated by and is a majority-owned subsidiary of Xperi Corporation, a leader in audio, imaging and semiconductor technology.