Foundry Model Busted: Will Apple Fab?

The problems at 28nm, replicating the problems at 40nm, go to show that the semiconductor industry needs a new model.

It seems only Intel and Samsung, out of all the non-DRAM IDMs, made the right call on manufacturing.

Everyone else is at the mercy of foundries which may deliver processes in a timely fashion, or may not. But sure as eggs are eggs everyone else has little control over the timing of their process transitions.

Mike Bryant tells me Intel first started to market 32nm product in February 2010.

Assuming 28nm is the same process as Intel’s 32nm, that means the rest of the industry is still struggling to get onto the same technology with the time-lag growing for every month that 28nm is unobtainable from a foundry.

After 40nm and 28nm, will the problems get easier at 20nm? Is the Pope a Catholic?

So much for all that bollox – largely concocted by the financial community – about advanced digital CMOS becoming a widely-available commodity which didn’t add value!

Well Intel didn’t fall for the bollox and now enjoys this huge process lead. “32nm was easy for them as they moved to hi-k a node earlier,” says Bryant.

The next generation – 20nm planar at the foundries/22nm finfet at Intel – may extend that lead – depending on when Intel moves into for-the-market production on it. July seems to be the current expected date.

So it seems AMD sold its fabs at the very moment process engineering became so difficult that it is now a key competitive differentiator.

While it seems that Qualcomm has no hope of going up against Intel unless it starts fabbing its own chips. It has held such discussions internally for many years.

Moreover, with $100 billion cash, several years of process R&D behind it, strained relations with Samsung and, now, impeded access to 28nm, will Apple bite the bullet and start to fab? It seems a logical step in its increasing verticalisation.

Apple could go it alone, but Qualcomm would probably have to enter a consortium of companies. The obvious partners would, of course, be Xilinx, Altera, Nvidia and, of course, AMD.

But aren’t AMD contracted to GloFo?

No, not actually, they paid GloFo $420 million to get out of their contractual relationships last week.

Well, I wonder why.


Comments

30 comments

  1. There’s a ready market for obsolete fab equipment – it’s called the European semiconductor industry, aka More than Moore.

  2. Well Andy Rem, at TSMC 28nm accounted for 2% of total wafer sales in Q4 2011, 40/45nm and 65nm represented 27% and 30%, respectively, and 65nm and below represented 59% of total wafer sales. The final SICAS report stated Q4b 2011 wafer starts at: over 0.7µm: 70k wpw; 0.7-0.4: 77k; 0.4-0.2: 103k; 0.1-0.12: 114k; 0.12-0.08 812k; 0.08-0,06 75k; under 0.06 1097k. Is trailing edge foundry a good business? I haven’t seem many people making at it.

  3. Excuse my ignorance, but what happens to all the old fab equipment when this new stuff is installed? I trust it gets used somewhere, and not just scrapped off, only to be turned into nasty, cheap, gas-fired patio heaters and other such junk? I doubt that 95% of applications actually “need” cutting edge processes/processors (including just about every TwitFace box out there), so I sincerely hope that it is all happily churning out perfectly useful chips somewhere in the world? If not, perhaps we should start setting up a kind of hi-tech scrapheap challenge series of fabs in European countries to try and bring some manufacturing back here?

  4. move the slider control here to appreciate just how far we have come… http://tinyurl.com/mac6zv

  5. You’re absoutely right of course Mike, the timings make no sense at all. Without the tools, the chip-makers can make nothing – and the tool-makers are telling a very different story to the chip-makers.

  6. It’s always been too expensive Ian, and no one has come up with the reason why it’s not too expensive now. But if, as Reuters reports today, ST-E will announce a re-vamp in 2 weeks prior to a sale of ST-E, this type of technology sell could be intended to help boost the price ST-E will achieve.

  7. Yes it’s good [Anonymous] but see what Mike says about Lamouche only talking about what FD-SOI COULD do IF it is used – rather than saying it WILL be used. And there’s the practical problem of getting a working, yielding, volume production 28nm FD-SOI process – no one seems to know the answer to that one. Furthermore Reuters reports today that ST-E will, within two weeks, announce a re-vamp prior to putting the company up for sale – in which case all this conversation could be irrelevant.

  8. Reading this info on FD SOI I find it even more strange that AMD decides to abandon it at this point.

  9. Hi David,
    ” process engineering became so difficult that it is now a key competitive differentiator.”
    I am the anonymous comenting the stericsson choice for FDSOI
    isn’t it a clear message for them to rush for a differentiator ?
    Did you notice what they can achieve in terms of performance and process simplification ?
    So give them credit for this courageous project…

  10. So what will TSMC make these pilot wafers with – a hammer and chisel ?
    You need to separate TSMC market-speak from reality. Intel and Samsung are fast consigning TSMC to be a Tier 2 player and they have to make these comments to look like that are still on a par with the two leaders. But one look at their technologies shows they have lost ground.

  11. Well I think TSMC have already said they’ll be ahead of Intel on 450mm, Me, TSMC say pilot line in 2013/4 volume production 2015/6 Intel say first production 2016/7 but it all depends on the availability of tools and the No.1 tool-maker ASML says 450mm production will start in 2018. So you pay your money and take your choice.

  12. I’d doubt that anyone’s going to catch Intel at going to 450mm first. Although D1X isn’t going to be 450mm at the beginning (maybe not until past 16nm), it’s got the ability to whenever it becomes cost effective. Unless someone’s secretly building a 450mm factory right now…I don’t see how they could catch up.

  13. Its called Common Platform always aimed to become the world’s largest foundry operation.

  14. Yes i wondered if you were kidding Ian but then thought maybe you know something i don’t.

  15. No, I don’t think they do. My remark was a little tongue in cheek, but at the same time it wouldn’t surprise me if the thought had crossed their mind.

  16. I didn’t think Foxconn had any experience in fab, ian.

  17. Either that or they persuade FoxConn to do it for them.

  18. There’s a sort of a precedent, Ian, The Taiwan government put up the money to found TSMC and build its first fabs while Philips supplied the technology. The China government might see it as a good example to follow with Apple.

  19. Well it could and it might, outsider, but Apple has been moving towards vertical integration for a while and fabbing its own stuff seems more logical.

  20. Why wouldn’t Apple end up with Intel rather than create their own fabs. Intel will have dual smartphone chip strategy (their own chips vs. apple chips)

  21. I remember a problem in a Chinese fab once. The implanters kept shutting down at night with alarms going off because an arsenic leak (I think) had been detected. The guys checked the tools again and again before realising that the arsenic or whatever it was was leaking into the fab from the air outside! Local companies took the opportunity of darkness to vent all kinds of crap into the atmosphere. Lovely.
    That was maybe 10 years ago. I don’t know if things have changed, but if not there are probably significant savings in costs there.

  22. The Yellow Peril won’t seem so bad after some Apple branding and showmanship.

  23. Absolutely [Anonymous] being competitive is what this is all about. As Intel expands into the arena of the fabless players, the fabless players absolutely need to have competitive processes. But where to get them if the foundries can’t hack it unless they do it themselves? Yes Samsung does have a collaboration with Glofo, but both of them are in with the IBM Common Platform Alliance which went for this gate first approach to 28nm which is turning out to be very flaky.

  24. Starting is one thing, but being competitive is another.
    Doesn’t Samsung have a partnership with GloFo and is also having 28nm yield issues?

  25. Apple or the Chinese, Lefty? Personally I don’t fancy being led round a football pitch then shot in the back of the head for insulting the president. But maybe Tim Cook won’t insist on that.

  26. Interesting though David. They already have developed capabilities by clustering companies of a feather together.
    I for one welcome out new masters.

  27. Well, Lefty, Apple could buy SMIC out of pretty cash. But think how much dough the China government would put up to secure the world’s first 450mm fab in partnership with Apple. Half of the $10 billion? Maybe even two thirds.

  28. Maybe apple will invest in a china fab?

  29. Exactly, anon786, but you have to start somewhere, and 450mm might be a good place to start.

  30. Well, it takes 3-4 years to build a fab at full speed, thats not including the billions it would take to build, buy machinery, new expert workers (Engineers etc) and R&D.
    Its not easy to start from scratch, no matter how much money one has. If money was not a problem, it would still take 3 years min.. then what about workers, r&d etc.

Leave a Reply

Your email address will not be published. Required fields are marked *

*