Tesla will start creating its own computer chips, optimized for machine learning, to power what Tesla founder and CEO Elon Musk calls the “world’s most advanced computer specifically for autonomous operation.”
The automaker intends to swap their chip, known as Hardware 3, into existing Tesla models to give them a performance boost and advance the cars’ self-driving capabilities. The new chip would deliver much better performance than the NVIDIA hardware the automaker has been using since 2016.
What sets Tesla apart from other self-driving car developers is its focus on developing its own hardware for performing artificial-intelligence computations that self-driving cars need—rather than continuing to rely on third-party hardware manufacturers.
Developing its own chips allows Tesla to capitalize on the benefit of “knowing what our neural networks look like, and what they’ll look like in the future,” said Pete Bannon, director of the Hardware 3 project. Bannon was tapped in 2016 to mastermind Tesla’s in-house AI chips after leading the development of cutting-edge chips at Apple.
Speaking during the company’s recent earnings call, Musk added that, “you have to do these calculations in the circuit itself, not in some sort of emulation mode, which is how a GPU or CPU would operate. You want to do a massive amount of [calculations] with the memory right there.”
By creating a custom chip, the company will be able to build hardware optimized for its needs rather than relying on off-the-shelf solutions that won’t deliver the performance the vehicles need.
And, if the hardware is found lacking, the automaker doesn’t have to wait for a third party to fix it. This is a problem Tesla seems to be tacitly acknowledging with this decision, as the NVIDIA-produced Hardware 2 chip was criticized for not being powerful enough to perform the self-driving capabilities promised by the company.
The company appears confident that its home-grown hardware will be up to the task and intends to start installing it in vehicles next year. “The chips are up and working … [and] all have been driven in the field,” Bannon said. “They support the current networks running today in the car at full frame rates with a lot of idle cycles to spare.”
The decision to create its own chips could be a big win for Tesla, allowing the automaker to build exactly the hardware it needs while retaining full control over its development. "Nobody was doing a bottoms-up design from scratch, which is what we elected to do," Bannon said. Tesla claims that creating their own chips will cost about the same as current hardware.
Learn more about the use of machine learning in self-driving cars at Autonomous Vehicles Armed with Machine Learning Algorithms.