DeepScale develops AI perception software for Advanced Driver Assistance Systems (ADAS) and autonomous vehicle applications. The company focuses on implementing efficient deep neural networks (DNNs) on automotive-grade processors, with the ability to leverage GPUs, DSPs, accelerators, and the Armv8 cores found on most automotive SoCs. DeepScale’s DNNs use data from various sensors to help vehicles of all automated driving levels better understand the world around them. The company is working to bring more deep learning capability to embedded processors, helping to bridge the gap between accuracy and cost for commercially viable automotive perception systems.

"Arm’s leading position and widespread use across the automotive market allows DeepScale to optimize our software for a variety of leading automotive SoCs. Arm has created an ecosystem of IP that is used across many edge processors. Being able to run DNNs on Arm CPUs enables DeepScale to achieve out-of-the-box performance on a variety of SoCs. This opens the door for leveraging other processor cores like accelerators and GPUs for heterogeneous DNN implementation, bringing more efficient solutions to market more quickly."

Forrest Iandola, CEO, DeepScale


DeepScale’s ADAS software product is called Carver and is currently being delivered to DeepScale’s key customers and partners. Carver offers automotive perception functions for development of differentiated ADAS functions, including AEB, ACC, LKA for NCAP. Carver’s AI software produces high-accuracy insights about the driving scene in real-time on embedded automotive-grade silicon.

DeepScale also collaborates with customers on deploying its efficient AI solutions in autonomous vehicles. Efficient AI on low-power processors is a key requirement to profitably scale autonomous fleets.

For more information, please visit