New Tesla Self-Driving Car Computer Has Two AI Brains

A Tesla logo on a Model S is photographed inside of a Tesla dealership in New York
FILE PHOTO: A Tesla logo on a Model S is photographed inside of a Tesla dealership in New York, U.S., April 29, 2016. (Photo: REUTERS/Lucas Jackson)

Tesla is one of the most aggressive and highly innovative companies in developing autonomous vehicles and their technological components. On Tuesday, tesla unveiled details on how it was able to fine-tune its Artificial intelligence chip design so a couple of them are smart enough to power the full self-driving abilities of its upcoming vehicles.

The latest in-house chip from the American autonomous vehicle maker is 21 times faster compared to older Nvidia model that Tesla used. Interestingly, each of the car's computer has a couple of AI brains for ultimate safety. Tesla CEO Elon Musk and his team launched the third-generation computing hardware last April.

However, at the Hot Chips Conference recently, Tesla's chip designers showcased how heavy the new Tesla custom AI chips' performance was boosted by heavy optimizations. The company requires better hardware to hit its 2019 full self-driving target. This means that Tesla cars can freely navigate not only in freeways but also on local streets with traffic lights and stop signs.

The processor chip requires major effort to design and the amount of work is shown in the gigantic number of transistors, which is approximately 6 billion for every Tesla chip. The company has powerhouse experts from processors to battery manufacturing to software and charging stations, which is its primary edge over other conventional automakers. The two AI chips are a redundant design for optimum safety.

Additionally, the chips also have redundancy in its power supplies, as well as in its data input feeds. The cameras on the tesla vehicle are on two separate power supplies to protect it against failures. These redundancies according to Tesla are deliberate to ensure that no untoward incident happens to the system in the event that the sensor, power supply, camera, or component fails.

Every Tesla chip creates its own assessment of what the vehicle will do next. The computer then compares these assessments and if both chips agree, the vehicle executes the action. In the event that the chips disagree, the vehicle simply throws the frame of video data and tries one more time. Each Tesla chip performs 36 trillion operations per second and runs at 2 GHz.

This is possible because Tesla made sure that the chips for self-driving vehicles are optimized and dropped anything more general purpose. The Tesla AI chip endures 14 long months to design with Samsung now manufacturing the processor. The chips are now on newer tesla vehicles and owners of older models have the option to upgrade.

© 2019 Business Times All rights reserved. Do not reproduce without permission.
Sign Up for Newsletters and Alerts