Author(s): Raghu Gopal
Tesla cars and trucks are electric-powered, rolling machines that need significant computing power. As these vehicles begin making driving decisions, there's a growing need for intelligent silicon that can learn and adjust using data from dozens of sensors to enable Tesla's Autopilot autonomous-driving feature.
The first version of its Autopilot system was based on chips from Mobileye, a leading supplier of self-driving car technology that's now owned by Intel. After falling out with Mobileye, Tesla built a second-generation Autopilot using solutions from Nvidia. However, rumours surfaced in September 2017 that the car-maker was developing its own artificial intelligence chipsets for its autonomous driving programme.
Last week, at the Neural Information Processing Systems conference in Long Beach, California, Tesla CEO Elon Musk confirmed that the company's vehicles will run its artificial intelligence software for self-driving cars on a chip designed by the automaker itself.
Tesla is apparently looking to reduce its reliance on Nvidia, or any other third-party supplier, for its autonomous car technology. This is consistent with Mr Musk's broader ambition to produce as many components in-house as possible. Although doing so involves significant costs, there are certainly legitimate arguments to support such tight integration. Tesla will be its own best customer for such a component and the chipset's developers should have a better understanding of the company's needs than an external provider. In theory, this should lead to better performance and greater safety on the road.
Tesla is following other high-tech companies including Apple, Google, Huawei, Intel, Microsoft, Nvidia and Qualcomm in creating custom chips to enable deep learning in the cloud or on mobile devices. Apple and Huawei have put them in their smartphones; Google has made its own to power its cloud artificial intelligence services; and custom vision chips are appearing in drones and consumer cameras. To varying extents, these companies have embarked on creating their own artificial intelligence chipsets or adopted a heterogeneous approach to deliver appropriate performance and efficiency.
Earlier in 2017, Mr Musk had predicted that Tesla's self-driving cars would be able to drive 10 times more safely than a human within three years. It's a bold prediction and Tesla, as a high-profile leader in the market for future-generation cars, needs to move quickly to support its reputation. For example, Waymo, part of Alphabet, is already ahead of that timetable in some regards, with fully driverless cars on public roads (see Waymo Hits Milestone with Autonomous Cars). Although the concept of autonomous driving is moving evolving at an impressive pace, expectations need to be tempered. We believe that widespread availability of so-called level 5 of autonomous driving, that is, cars which need absolutely no human intervention and don't even have steering wheels, is still years away.
Tesla's ambition to develop its own artificial intelligence chipsets is reflective of its self-perception and its aim to be as independent as possible. With the move, the company joins an IT trend of developing artificial intelligence hardware. Its computers are on wheels, but this is a high-tech company in every respect. However, Tesla faces the challenging task of delivering the autonomous-driving vision and doing so at scale. The jury is out on whether designing its own chipsets at this stage is a critical point of control and differentiation, or yet another proprietary component that slows Tesla's expansion and makes it vulnerable to mounting competition.