You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
This was mentioned at the TSMC Symposium. All of the automotive CEOs are now C.C. Wei's new best friends:
Stable procurement of semiconductors:
From a medium- to long-term perspective, Honda has been building cooperative relationships with semiconductor manufacturers, including the strategic collaboration with Taiwan Semiconductor Manufacturing Company Limited (TSMC).
Taiwan Semiconductor Manufacturing Co on Wednesday said it will release new software this year to help customers working on advanced computer chips for cars take advantage of its newest technologies more quickly.
Taiwan Semiconductor Manufacturing Co on Wednesday said it will release new software this year to help customers working on advanced computer chips for cars take advantage of its newest technologies more quickly.
I'm wondering if at least a small part of it is, they know 3nm will be a very long term node, being the last of the FinFETs, and they would rather develop for it, than a more transitory node that might not have the same level of commitment from TSMC.
I'm wondering if at least a small part of it is, they know 3nm will be a very long term node, being the last of the FinFETs, and they would rather develop for it, than a more transitory node that might not have the same level of commitment from TSMC.
Good question. I'm guessing many parts will, but given the advantages in terms of density, speed and power use, they may find 28 nm and largely don't adequately address their needs, especially going forward. But, I'd guess many of the less demanding parts will use 28nm or higher.
I know Intel said on the last earnings report their IFS improvement was mainly from car companies, and I'm guessing Intel doesn't have a lot of pre-22nm production going right now. So, at least for some parts, it seems FinFETs address the needs better.
Of course, they have to be forward looking too. Even if 28nm is kind of OK now, will it be a few years from now? And do they want to have to address it down the road.
But, I was thinking kind of the same thing, are automobile devices so sophisticated now they require relatively modern nodes? I guess so, based on this, and also on Intel doing well with automakers.
I think 28nm is still adequate for most automotive MCs for now and for many years in the future. However, noone is building new 28nm fabs so long term supply of 28nm chips for a growing market is likely part of the calculus here. By going to 3nm, they are future proofing themselves from both a technology and long term supply standpoint.
One factor is whether or not an automaker wants to follow Tesla down the internal ADAS development path or not. Maybe Honda does. This Mobileye ASIC looks sophisticated enough it must be on an advanced TSMC process.
The latest additions to our family of Systems-on-Chip, EyeQ®6L and EyeQ®6H bring new levels of performance and efficiency to core and premium ADAS.
www.mobileye.com
Another factor is whether automakers are planning to stay with their usual bunch of distributed microcontrollers (which are old process friendly), or whether they're planning to switch to a central processing strategy with big CPU ASICs and networked signaling. It should be noted that several automakers participate in the Automotive Ethernet IEEE 802 working group, which implies interest in centralized designs. This is a pretty interesting presentation from a few years ago:
One factor is whether or not an automaker wants to follow Tesla down the internal ADAS development path or not. Maybe Honda does. This Mobileye ASIC looks sophisticated enough it must be on an advanced TSMC process.
The latest additions to our family of Systems-on-Chip, EyeQ®6L and EyeQ®6H bring new levels of performance and efficiency to core and premium ADAS.
www.mobileye.com
Another factor is whether automakers are planning to stay with their usual bunch of distributed microcontrollers (which are old process friendly), or whether they're planning to switch to a central processing strategy with big CPU ASICs and networked signaling. It should be noted that several automakers participate in the Automotive Ethernet IEEE 802 working group, which implies interest in centralized designs. This is a pretty interesting presentation from a few years ago:
I had also assumed TSMC was correct, because Mobileye's strategy seems to include a lot of external IP, and that is only emerging on Intel's internal processes. Also, the mostly likely alternative to TSMC, Samsung, is probably not a partner Intel would encourage (the two companies don't seem to have a great working relationship). So even if the link above isn't stating a fact, I think TSMC is likely.
www.automotiveworld.com
/cloudfront-us-east-2.images.arcpublishing.com/reuters/UFNAJ6UWONPBJFD4D7MIVUEB2Q.jpg)
