Untether AI and GM to develop next-generation autonomous vehicle perception systems; “at-memory computation”
29 April 2022
Untether AI, a developer of at-memory computation for artificial intelligence (AI) inference acceleration, announced a collaboration with General Motors, supported by the Government of Ontario through the Ontario Vehicle Innovation Network (OVIN), to accelerate the development of next-generation artificial intelligence perception systems for autonomous vehicles.
More than 90% of the power consumption in AI workloads is from the movement of data, according to Untether AI. Untether AI moves the compute element to where the data is stored (“at memory”), reducing the power consumption for data transfer by 6 times. This is the fundamental innovation that allows Untether AI to provide unprecedented compute density, untethered to traditional approaches.
At the heart of the unique at-memory compute architecture is a memory bank: 385KBs of SRAM with a 2D array of 512 processing elements. With 511 banks per chip, each device offers 200MB of memory, enough to run many networks in a single chip.
Perception systems are at the heart of autonomous vehicles (AVs). Under the agreement, Untether AI and GM will work together to demonstrate an autonomous vehicle perception system based on Untether AI’s unique at-memory computation technology. This project is supported with $1 million from the OVIN R&D Partnership Fund which leverages $2.09 million of industry contribution.
AV perception systems use AI, in the form of neural networks, to understand the environment around the vehicle, which is critical to the functionality of autonomous vehicles. These systems are highly complex since they must make sense of many external inputs, such as multiple channels of video cameras, radar, ultrasound, and lidar data.
The project will incorporate Untether AI’s next-generation at-memory compute technology to accelerate neural networks for AV perception systems. GM will contribute guidance and insight on AV perception system requirements, as well as providing access to the neural network technology that is utilized in the company’s current AV perception system.
The project’s goal is to help increase autonomous vehicle performance by providing faster and more accurate vision response, reducing the power draw of various AV systems to increase battery life, and even helping to enable the downsizing of vehicle batteries. Additionally, this solution will aim to save on overall autonomous vehicle control system cost.
Operating as the global venture capital arm of General Motors, GM Ventures strategically invests in startup companies that share GM’s vision of a world of zero emissions, zero crashes and zero congestion, and are helping to accelerate the company’s evolution into a leading transportation technology enterprise. Having served as a strategic investor in Untether AI since 2018, this pilot project represents an exciting next step in the exploration of potential applications for the technology and demonstrates how establishing a connection with GM’s network of resources can lead to opportunities for tangible innovation.
Founded in Toronto in 2018, Untether AI is funded by CPP Investments, Intel Capital, Radical Ventures, and Tracker Capital.