Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Report: Tesla working with Samsung on 5nm FSD chip

This site may earn commission on affiliate links.

diplomat33

Average guy who loves autonomous vehicles
Aug 3, 2017
12,691
18,649
USA
"Tesla has reportedly partnered with Samsung on a new 5nm chip for full self-driving, according to a new report coming from Korea:

According to related industries on the 25th, the Samsung Electronics Foundry Division is currently conducting research and development (R&D) on 5nm-class system semiconductors to be mounted on Tesla autonomous vehicles. The 5nm semiconductor applied with the extreme ultraviolet (EUV) process is a high-tech product that only a small number of companies such as Samsung Electronics and TSMC can produce worldwide."

Tesla partners with Samsung on new 5nm chip for full self-driving, report says - Electrek

Is this a hint to the next version of the FSD computer? FSD 2.0 computer?
 
  • Informative
Reactions: shrineofchance
"Tesla has reportedly partnered with Samsung on a new 5nm chip for full self-driving, according to a new report coming from Korea:

According to related industries on the 25th, the Samsung Electronics Foundry Division is currently conducting research and development (R&D) on 5nm-class system semiconductors to be mounted on Tesla autonomous vehicles. The 5nm semiconductor applied with the extreme ultraviolet (EUV) process is a high-tech product that only a small number of companies such as Samsung Electronics and TSMC can produce worldwide."

Tesla partners with Samsung on new 5nm chip for full self-driving, report says - Electrek

Is this a hint to the next version of the FSD computer? FSD 2.0 computer?
Makes no sense. FSD is like, ready to go. :rolleyes:
 
"Tesla has reportedly partnered with Samsung on a new 5nm chip for full self-driving, according to a new report coming from Korea:

According to related industries on the 25th, the Samsung Electronics Foundry Division is currently conducting research and development (R&D) on 5nm-class system semiconductors to be mounted on Tesla autonomous vehicles. The 5nm semiconductor applied with the extreme ultraviolet (EUV) process is a high-tech product that only a small number of companies such as Samsung Electronics and TSMC can produce worldwide."

Tesla partners with Samsung on new 5nm chip for full self-driving, report says - Electrek

Is this a hint to the next version of the FSD computer? FSD 2.0 computer?
Probably .. thought my guess is less of a seismic shift that HW 3.0 and more a speed bump plus lower power/ heat.
 
  • Like
Reactions: diplomat33
"Tesla has reportedly partnered with Samsung on a new 5nm chip for full self-driving, according to a new report coming from Korea:

According to related industries on the 25th, the Samsung Electronics Foundry Division is currently conducting research and development (R&D) on 5nm-class system semiconductors to be mounted on Tesla autonomous vehicles. The 5nm semiconductor applied with the extreme ultraviolet (EUV) process is a high-tech product that only a small number of companies such as Samsung Electronics and TSMC can produce worldwide."

Tesla partners with Samsung on new 5nm chip for full self-driving, report says - Electrek

Is this a hint to the next version of the FSD computer? FSD 2.0 computer?

I saw another article on this that said it was for the infotainment system, not AP.
 
Can you link to that other article, please?

You could be right. I am just reporting what Elektrek is reporting.

The article was in Korean and a redditor provided a quick translation, quoted and linked below.
“Tldr;translation:

They are teaming up for IVI (infotainment) semiconductors which will become one of the central chips for autonomous vehicles. These include NPU's, processors, DDI, security, etc. They also play an important role in transmitting information from various sensors to the display. As the industry enters 5G and 6G networks, IVI-related chips will partake an increasingly important role. Samsung Foundry previously manufactured Tesla chips using the ArF photolithography process, but the new chips will be manufactured on Samsung's 5nm EUV process.”

https://www.reddit.com/r/teslamotors/comments/l4gk85/tesla_and_samsung_team_up_to_develop_5nm_euv/
 
  • Helpful
Reactions: diplomat33
The article was in Korean and a redditor provided a quick translation, quoted and linked below.
“Tldr;translation:

They are teaming up for IVI (infotainment) semiconductors which will become one of the central chips for autonomous vehicles. These include NPU's, processors, DDI, security, etc. They also play an important role in transmitting information from various sensors to the display. As the industry enters 5G and 6G networks, IVI-related chips will partake an increasingly important role. Samsung Foundry previously manufactured Tesla chips using the ArF photolithography process, but the new chips will be manufactured on Samsung's 5nm EUV process.”

https://www.reddit.com/r/teslamotors/comments/l4gk85/tesla_and_samsung_team_up_to_develop_5nm_euv/

Thanks
 
Why would Tesla need a custom infotainment system chip, instead of using something off the shelf?
When I reread the original reports, it isn’t obvious that it is a custom chip. My thinking is either Tesla is evaluating future mainstream 5nm chips now, or Tesla is considering building HW5 on Samsung fabs. HW4 was built on 7nm at TSMC.
 
Last edited:
I would find it hard to believe the they would have to go to a 5nm process for an infotainment chip, even if it was a relatively high level of integration. You don't really need 5nm until you are designing massive processors, or even a massive neural network processor.

Does the fact that the HW3 FSD chip is sufficient for full self driving mean that development stops there and they never need to enhance it? Hardly! There will always be improvements that can be made, even if just for cost or power reasons (although for pure cost & power reasons, a jump to 5nm seems completely unnecessary). They could potentially be adding more neural net capacity so as to support longer "4D" timeframes or higher resolution cameras or additional sensors or any number of things designed to make FSD even safer.

But another possibility is that these chips will not go into the car at all, but rather be part of the massive Dojo server(s) used for training. They can probably ALWAYS use more capacity in Dojo to use for training, so if you can get the same amount of capacity in one 5nm as you can in 4 to 8 14nm chips you can turn around training runs that much faster or process that much more uploaded scenarios from the fleet. In my opinion, this is most likely where Tesla 5nm silicon is headed.
 
  • Helpful
Reactions: diplomat33
I would find it hard to believe the they would have to go to a 5nm process for an infotainment chip, even if it was a relatively high level of integration. You don't really need 5nm until you are designing massive processors, or even a massive neural network processor.

Does the fact that the HW3 FSD chip is sufficient for full self driving mean that development stops there and they never need to enhance it? Hardly! There will always be improvements that can be made, even if just for cost or power reasons (although for pure cost & power reasons, a jump to 5nm seems completely unnecessary). They could potentially be adding more neural net capacity so as to support longer "4D" timeframes or higher resolution cameras or additional sensors or any number of things designed to make FSD even safer.

But another possibility is that these chips will not go into the car at all, but rather be part of the massive Dojo server(s) used for training. They can probably ALWAYS use more capacity in Dojo to use for training, so if you can get the same amount of capacity in one 5nm as you can in 4 to 8 14nm chips you can turn around training runs that much faster or process that much more uploaded scenarios from the fleet. In my opinion, this is most likely where Tesla 5nm silicon is headed.

Dont forget right now HW3 is running two chips in parallel. At 5nm it's possible they could cram them both onto the same die .. that would save $$ on the chips and even more on the overall mfg cost of the computer.