Qualcomm considering getting into the Fab business

[Phynaz]

They aren't happy with the supplies of their 28nm SOCs.

http://www.eetimes.com/electronics-news/4376352/Qualcomm-wafer-fab-not-ruled-out

 

[Martimus]

Sounds like empty threats to try and light a fire under their supplier. They are far more likely to pay TSMC more to get a bigger percentage of the wafers produced. I can't imagine how they would be able to build a cutting edge fab, unless they just bought GF and used theirs.

 

[Idontcare]

Sounds like empty threats to try and light a fire under their supplier. They are far more likely to pay TSMC more to get a bigger percentage of the wafers produced. I can't imagine how they would be able to build a cutting edge fab, unless they just bought GF and used theirs.

On the contiumm of options that Qualcomm has access to when it comes to spending money to secure a higher supply volume, buying a fab and investing in all that goes into maintaining a fab is way way way out there on the tail of the curve.

Long before they elect to take such a drastic move are options just as Martimus highlighted. They probably pay ~$6k per 28nm wafer from TSMC. If they wanted more of TSMC's limited 28nm capacity they need simply outbid the other companies that TSMC is selling 28nm wafers to (and not now, contracts are locked, they needed to do it back when 28nm negotiations were still under negotiation).

You can bet that TSMC would take wafers away from Nvidia and AMD if Qualcomm enticed them with a $12k/wfr contract.

Further, yet another option that would be exercised before the decision to buy a fab would be the so-called "co-op" contract which is when a fabless supplier essentially partners with a foundry to more or less do a JV (joint venture) in spirit but not in name. Qualcomm would contract TSMC for a wafer contract that specified a guaranteed volume of wafers at a specific price (say 100k 28nm wfrs/month @ $6k/wfr) in exchange for a lump-sum fixed payment in advance to cover some of the expenses of the capacity buildout that would be required for the wafer volumes to be feasible (say a $2B payment 2yrs in advance).

All cheaper than buying your own fab, and not unfamiliar to the industry.

I put the odds of Qualcomm buying a fab at about the same place as I put the odds of Intel becoming fabless.

 

[ShintaiDK]

I put the odds of Qualcomm buying a fab at about the same place as I put the odds of Intel becoming fabless.

Agree. The ROI for Qualcomm would ruin them. They cant charge 100-200$ for something that cost 5-10$ to make.

 

[hans007]

i could see them coopting a new fab. sandisk and toshiba basically do this

 

[soccerballtux]

why the heck is nobody using Global Foundries? Don't they have some serious untapped production potential? Perhaps the power consumptions at their 32nm suck (acceptable to good for desktop/laptop chips, not for mobile)?

 

[ShintaiDK]

why the heck is nobody using Global Foundries? Don't they have some serious untapped production potential? Perhaps the power consumptions at their 28nm suck (acceptable to good for desktop/laptop chips, not for mobile)?

I dont think Global Foundries have shipped a single 28nm chip yet.

 

[lol123]

No wonder that Google has partnered with Intel through Motorola over the future of the Android ecosystem, and made such an effort to optimize Android for x86. They might have seen this coming.

Also, this is just so damn funny. The ARM party is over before it even started.

 

[soccerballtux]

I dont think Global Foundries have shipped a single 28nm chip yet.

Sorry, you're right.

32nm is close enough?

 

[Idontcare]

why the heck is nobody using Global Foundries? Don't they have some serious untapped production potential? Perhaps the power consumptions at their 32nm suck (acceptable to good for desktop/laptop chips, not for mobile)?

Sorry, you're right.

32nm is close enough?

32nm is SOI, not bulk-Si, so cost/wafer would be a factor. Plus being 32nm design rules and not 28nm design rules, the chips would be bigger making them more expensive, more power hungry, and slower than their 28nm competitor's chips.

As for "why no one is using GloFo 28nm", even if Qualcomm switched plans last year to use GloFo 28nm when they realized 28nm at TSMC would be so capacity constrained, and Glofo had volume 28nm available today, Qualcomm would still be at least 1yr away from having retail-ready chips shipping in volume (absolute best timeline possible, assuming zero hiccups).

These things have to be planned and coordinated years in advance. That's why its best for Qualcomm to figure out what it can do to bolster their supplier's ability to supply them chips rather than rattling their saber with PR soundbites that accomplish nothing outside of making more friction in the negotiation room.

 

[cbn]

Further, yet another option that would be exercised before the decision to buy a fab would be the so-called "co-op" contract which is when a fabless supplier essentially partners with a foundry to more or less do a JV (joint venture) in spirit but not in name. Qualcomm would contract TSMC for a wafer contract that specified a guaranteed volume of wafers at a specific price (say 100k 28nm wfrs/month @ $6k/wfr) in exchange for a lump-sum fixed payment in advance to cover some of the expenses of the capacity buildout that would be required for the wafer volumes to be feasible (say a $2B payment 2yrs in advance).

All cheaper than buying your own fab, and not unfamiliar to the industry.

Great info. Thank you.

 

[ViRGE]

I put the odds of Qualcomm buying a fab at about the same place as I put the odds of Intel becoming fabless.

Indeed. Any time someone says they're going to get into fabs, ask this one simple question: do they have 15 billion dollars in the bank to freely spend and likely lose? It's a very short list.

 

[Homeles]

I dont think Global Foundries have shipped a single 28nm chip yet.

Not for consumer consumption, no. I'm sure they've sent out 28nm chips for R&D purposes though.

 

[dealcorn]

The 28 nm chip supply issue may ease as Atom ramps at 14 nm. Perhaps earlier if 22 nm Atom is spot on.

 

[Arachnotronic]

Indeed. Any time someone says they're going to get into fabs, ask this one simple question: do they have 15 billion dollars in the bank to freely spend and likely lose? It's a very short list.

Weeelll....Qualcomm does have...exactly....$15B in the bank.

 

[ShintaiDK]

Weeelll....Qualcomm does have...exactly....$15B in the bank.

So they can build 2 fabs for around 8-10B. Then wait 3-4 years till complete. Then start R&D teams...I assume you already see the issue by now.

 

[Arachnotronic]

So they can build 2 fabs for around 8-10B. Then wait 3-4 years till complete. Then start R&D teams...I assume you already see the issue by now.

Haha yeah, I was just pointing out that they had this exact amount!

 

[Ferzerp]

As much as I hate Apple, they're on the short list of companies who could take the risk, but why do that when you can bully suppliers in to doing it much cheaper for you (Why build a fab to make ARM chips basically, because they can't make their x86 ones)?


Though I suppose if they wanted to go AMD, they could build a fab, and make a deal with AMD.... (why?)

 

[blckgrffn]

As much as I hate Apple, they're on the short list of companies who could take the risk, but why do that when you can bully suppliers in to doing it much cheaper for you (Why build a fab to make ARM chips basically, because they can't make their x86 ones)?


Though I suppose if they wanted to go AMD, they could build a fab, and make a deal with AMD.... (why?)

Its kind of funny (sad?) when you can do better by having suppliers gouge themselves to get your volume cheaper and more efficiently than you can vertically integrate.

Apple is a marvel of a company, and I respect their ability to print cash even if I don't buy their stuff or completely agree with their ethics to do so - and think their market capitalization versus companies like Exon is borderline bogus.

Apple is in the position of being able to do what IDC spoke of, clearly. Do they take advantage of that? Or do they just lord their massive volume purchasing power about in a way that this is unnecessary?

 

[ShintaiDK]

The problem is the ROI. Right now basicly only TSMC, Intel and perhaps Samsung can make money on 14nm for example. And below that it might only be Intel. And Intel can only do it due to the massive margins.

Apple could potentially, but why? They would have to sacrifice huge margins just to run the fabs, meaning less profit.

 

[Ferzerp]

Apple is a marvel of a company, and I respect their ability to print cash even if I don't buy their stuff or completely agree with their ethics to do so - and think their market capitalization versus companies like Exon is borderline bogus.

Oh, there is no doubt that they are massively overvalued. That bubble will eventually collapse.

 

[tommo123]

The problem is the ROI. Right now basicly only TSMC, Intel and perhaps Samsung can make money on 14nm for example. And below that it might only be Intel. And Intel can only do it due to the massive margins.

Apple could potentially, but why? They would have to sacrifice huge margins just to run the fabs, meaning less profit.

might work for mobile devices re apple but not desktop. no way intel would let apple fab their chips. unless apple moves all products over to arm.

i'm assuming that there's patent issues involved as well which intel could use to prevent another major player joining?

p.s i though samsung had already said they were investing in new fabs?

p.p.s could we see TSMC/GF or samsung (or all 3) joining forces at some point in the fab game? if not and intel rules the roost that would mean intels cell phones would be far more power efficient/faster and overall better than the competition no? assuming good designs and a healthy advantage in process tech

 

[lol123]

Apple will never fab their own chips. Anyone who understands the slightest bit about Apple's business model (maximal profits with a minimum of capital expenditures, and minimal risk in the manufacturing chain) can figure out why.

I can imagine that Qualcomm would, since they are already making ARM architecture designs that set themselves apart from the rest of the ARM manufacturers (and doing a good job at that) and positioning themselves as a direct competitor to Intel. If they can have any success with it when even IBM, TSMC and GlobalFoundries are faltering is another story.

p.p.s could we see TSMC/GF or samsung (or all 3) joining forces at some point in the fab game? if not and intel rules the roost that would mean intels cell phones would be far more power efficient/faster and overall better than the competition no? assuming good designs and a healthy advantage in process tech

Logically, that is what should happen, just like I think that logically we should have seen a move toward deeper partnership between IBM and AMD with Intel's increasing dominance of the server, desktop and laptop CPU markets. But so far there haven't been any indications of that happening. It might be that IBM simply feel that hardware market shares aren't worth pursuing and are happy to just keep their customers on IBM software and middleware stacks even if that is increasingly on Intel's silicon and platforms. Maybe other, smaller players are similarly feeling that they can't compete no matter what they do.

It should be added that the belief that it's fine to stay with older processes as long as you have an "elegant" architecture (which seems to be the PR message of the day adopted by some of the ARM players) is a highly precarious approach. It was part of the reason why the Itanium, which was always one or two process nodes behind enterprise server x86 CPUs, failed so catastrophically to gain any market share, despite what many would call an efficiency advantage over x86.

 

[Ferzerp]

Apple will never fab their own chips. Anyone who understands the slightest bit about Apple's business model (maximal profits with a minimum of capital expenditures, and minimal risk in the manufacturing chain) can figure out why.

No one claimed they would. We just said they were one of few who could.

 

[Ferzerp]

Logically, that is what should happen, just like I think that logically we should have seen a move toward deeper partnership between IBM and AMD with Intel's increasing dominance of the server, desktop and laptop CPU markets. But so far there haven't been any indications of that happening

That would be a stupidest possible move for IBM. Any large seller of x86 servers would have to be an idiot to tie themselves to AMD. IBM has no personal computer business. Their only x86 product is in the server space. Where did you get the idea that IBM partnering with AMD would make any sense at all? Wow.

They could then either be tied to AMD, and sell only AMD servers, letting them compete for the tiny server space (we're talking 5-8%?) that AMD is purchased in, or they could be tied to AMD, and still sell Intel servers. Which of those make sense? To me, neither.