Global Foundries 7NM cancelled - AMD moves all 7NM production to TSMC

[CAT-THE-FIFTH]

GF again fails:

https://www.anandtech.com/show/13277...nm-development
https://www.extremetech.com/computin...uction-to-tsmc

Hopefully AMD can get fully out of WSA now.

[Corky34]

I can understand why they've chosen to do that but it defiantly sucks for us consumers, EUV has been lauded as a savior for decades but caused as many problems as it solved and ended up costing a fortune, costs that will inevitably be passed onto us.

I get why AMD are pushing for 7nm and I hope the advantages outweigh the disadvantages as if they don't it could really backfire on them, i get the impression everyone is expecting big things from 7nm, i hope they're not setting themselves up for disappointment.

[CAT-THE-FIFTH]

I can understand why they've chosen to do that but it defiantly sucks for us consumers, EUV has been lauded as a savior for decades but caused as many problems as it solved and ended up costing a fortune, costs that will inevitably be passed onto us.

I get why AMD are pushing for 7nm and I hope the advantages outweigh the disadvantages as if they don't it could really backfire on them, i get the impression everyone is expecting big things from 7nm, i hope they're not setting themselves up for disappointment.

It was kind of weird Epyc was touted to launch first at TSMC,so it kind of tells you something was amiss.

The main issue will be capacity methinks.

[Biscuit]

Cancelled? Christ, Global Foundries have done AMD no bloody favours over the years. At least this time AMD seems to have set them selves up with some redundancy.

[Hoonigan]

Are we approaching the point where we can't shrink dies any further? I remember owning a 0.18um CPU about 10/11 years ago, then two years on from that I had a 65nm, then a few years on, a 45nm and now a 14nm.

It's ages since we took any massive leaps.

Are we stuck?

[aidanjt]

Are we stuck?

Kinda, TSMC is on track to get 5nm up to production ~2019-2020 (probably the latter), but we're fast running up against hard physical limitations. Can't really make transistors from fractions of a single silicon atom (~0.2352nm).

[Corky34]

Are we stuck?

I wouldn't say stuck as smaller transistors are totally doable, it's just the cost effectiveness of doing so.

I'm surprised we've made it this far without any major rethinks on how transistors are made, other than FinFET we've largely been using the same techniques and materials for the last few decades, even with FinFET it only really changed a 2D structure into a 3D structure.

Kinda, TSMC is on track to get 5nm up to production ~2019-2020 (probably the latter), but we're fast running up against hard physical limitations. Can't really make transistors from fractions of a single silicon atom (~0.2352nm).

That and just trying to connect them is proving to be a real headache, when the bumps and wires connecting the transistors are almost as large as the transistor themselves it causes all sorts of problems, e.g when a copper wire drops below a certain size its resistance actually starts going up so you have to start looking at alternative.

[watercooled]

I didn't take the news as a big surprise to be honest, they've not been turning a profit for a long time, they're struggling for capacity and capacity expansions are eye-wateringly expensive. Plus with AMD already having multiple 7nm products announced at TSMC (probably at least partly due to capacity limitations at GloFo but I wonder if they've known about this for a while now, at least as a possibility so were mitigating risks)?

I also wonder about what it means for the WSA; I doubt GloFo have much of a leg to stand on WRT demanding wafer orders from AMD if they can't meet requirements for their products going forward. Much like 28nm before it, 14nm isn't going away any time soon but longer-term I wonder how they'll continue to differentiate from the other fabs who've tapped out of the leading-edge race?

WRT capacity at TSMC - losing a major competitor like this likely strengthens their position as far as capacity expansions go, and I do wonder if Samsung will be making more appearances in these market segments?

I do wonder how IBM feel about all this though? IIRC they still have obligations to manufacture some stuff in the USA and it doesn't leave them many options now, unless Intel can be convinced to contact manufacture for a competitor? Then again maybe Samsung having some of their fabs in USA opens other options?

Edit: Given their size and past dealings, I wonder if there's a possibility of Samsung buying GloFo? TSMC are pretty huge in comparison to the others now: https://semiengineering.com/foundry-challenges-in-2018/



Edit2: Samsung are bigger than they look in that revenue table though (which I think only relates to foundry services):
http://www.icinsights.com/news/bulle...-Fab-Capacity/
I think a good deal of that is memory for Samsung though.

[CAT-THE-FIFTH]

GF is still developing SOI type processes on 12NM and 22NM AFAIK.

[badass]

I wonder if there's a possibility of Samsung buying GloFo?

Seems like a sensible move if it's doable.