Recogni Inc., a startup developing artificial intelligence hardware for self-driving cars, today exited stealth mode with backing from a Who’s Who of auto industry players.

Toyota AI Ventures, BMW i Ventures and Faurecia SA, one of the world’s largest auto parts suppliers, participated in a $25 million round into the startup prior to the launch. GreatPoint Ventures led the investment. Recogni previously raised a seed round of undisclosed size to support early development efforts.

San Jose, California-based Recogni has built a hardware module that enables autonomous vehicles to see their surroundings. It’s based on a homegrown application-specific integrated circuit that turns raw data from a car’s sensors into information the navigation system can understand. AI models running on the module identify nearby vehicles, pedestrians and other objects of interest.

The device collects environmental data from three built-in cameras that each have different color sensitivities. There’s also an Ethernet cable for connecting to an external lidar or radar sensor, which the onboard chip can use to supplement the camera footage.

Recogni is looking to take on Nvidia Corp., whose chips power the vision systems in many of the self-driving cars currently being developed by automakers. What sets the startup’s hardware apart is efficiency. 

The custom chip inside the Recogni module can perform more than 1,000 tera operations per second, or a quadrillion calculations, using just five watts of electricity. The startup claims that makes the system upwards of 500 times more power-efficient than competing solutions.

Recogni’s promised energy savings are significant because analyzing sensory data is one of the most taxing operations performed by a self-driving car’s computer. As a result, streaming the task can in theory significantly improve the overall power efficiency of the vehicle.

The Recogni module’s efficiency is the result of several factors. One of them is that it relies on passive cooling, meaning there’s no need for a power-hungry fan, and another contributor is the onboard chip’s close physical proximity to the three included cameras. This reduces the amount of electricity that has to be spent on moving sensory data from the cameras to the processor.

Recogni said that it’s currently in talks with multiple potential buyers in the auto industry. Gilles Backhus, the startup’s head of AI, said at an AI event last year that Recogni is aiming to bring the system to market by the middle of next year for $200. Backhus said the cost is expected to go down as production ramps up over time.

Photo: Unsplash