Although Qualcomm, as fifth largest chip company, has about $6 billion of cash on hand, I don't think Jacob's reported sentiment means that Qualcomm is going to leave the ranks of the fabless any time soon. There are many ways to finesse ownership issues and almost all of them will be about partnership and finding common ground.
So for me this disclosure is more about the likes of Globalfoundries, Samsung and Intel, versus Taiwan Semiconductor Manufacturing Co. Ltd. and playing one of against another than it is about the creation of the Qualcomm Semiconductor Manufacturing Company.
Such is the long lead-time and high expense of bring up wafer fabs – typically between $5 billion and $10 billion at today's leading-edge – it is almost impossible to think Qualcomm would start from scratch on chip manufacturing.
An alternative would be for Qualcomm to buy all or a part-share in an existing wafer fab that can produce its processors. Some chip manufacturers are struggling for profitability and might be willing to cut a deal, although most of the chip companies able to manufacture at 28-nm and 20-nm are in a more profitable situation. Nonetheless at the leading-edge capital expenditure is ferocious and Qualcomm has been piling up the cash and has interest in changing a situation that has left it short of chips and disappointing customers.
Another more likely possibility is that Qualcomm would be asked to put a significant amount of cash down to reserve a proportion of the output of an existing leading-edge wafer fab's output. Such part equity, part pre-payment deals have existed in the past although the circumstances that have driven their creation have not always been long-lived. And Jacob's himself has said that the 28-nm chip supply situation is improving and Qualcomm would hope to be meeting customers' requirements by the end of 2012.
So if Jacob's musings are about partnership who is he thinking about partnering with?
I'm not sure it is about planning earlier. As a fabless company Qualcomm is relying on its suppliers to come up with the goods.
But it may need to spend a little of its cash pile to give TSMC and others a little boost or risk losing design slots to rivals using Samsung or Intel.
I would agree with you, Peter, that there is virtually no change of Qualcomm opening up its own fab. The idea of Qualcomm striking a deal with UMC is an interesting one. As the article mentions, such a deal might be just what UMC needs to get closer to TSMC on the leading edge of process technology.
It seems to me that what Qualcomm really needs to do is strike a better deal with TSMC, including ponying up more money to secure the wafers it needs. Certainly, any such deal would be cheaper and more realistic than the idea of establishing a Qualcomm fab. Qualcomm execs admitted a few months ago that the 28-nm capacity issues had a lot to do with the fact that 28-nm chips were in greater in demand, earlier on, than they expected.
If you do the math right, it does not even make sense for a company like Qualcomm to own a fab. They always had a option to purchase additional capacity ahead of time with take or pay option in their supply agreement if they planned it right. The problem is that they never execised its option until it's too late. To me it's a Qualcomm's indecision and management issue and not able to plan their demand picture correctly.
Peter, thanks for enjoyable speculation. What would you think of yet another scenario in which QC funds some of the competent Japanese Fabs now in dire straits ? e,g. Elpida, convert their DRAM Fab to 28 nm Processor ( a la Samsung ) or even Renesas. When it comes to Fab discipline the Japanese are still tops, expect that after much Fab technology diffused out from the US to lower cost Taiwan and So. Korea and they got most of the business from the Smart Phone / Tablet boom, the Japanese could no longer keep up.
@chipmonk- an interesting idea. I found it quite interesting that when all the speculation was swirling a few months ago about a mega merger involving Japanese SoC makers, Globalfoundries was somehow involved taking over Renesas fabs. What if Qualcomm partnered with GloFo in order to secure some of that capacity for its dedicated use?
Qualcomm can plan better, and partner better, but there's no guarantee that the same thing wont happen again at the next node. And it makes no sense for them to build or buy a fab. Qualcomm needs to accept the fact that they will be about a node behind the IDMs for the next few years and try to innovate in other ways (more than Moore).
Yes, for the most part. That's just the technology side of the equation though. Then there's the manufacturing side of the equation. Taken together some other IDMs probably have an inherent advantage in cost/performance/time to market even if they are at the same node or even a node behind TSMC. As semiconductor technology gets even more complex the IDM advantage will likely grow. As others have noted, the free ride for fabless companies is over.
I would n't quite put Samsung in the same League as INTC yet. FinFETs represent a Physics barrier that even with a crash program funded by $10 billion+ R&D budget would take about 3 years to cross ( put into HVM ). Perhaps till then it should not be just "IDMs" but IDM_1 and IDM_0.5
Taking multiple approach rather than relying on one company could benefit Qualcomm. Rather than spending money and resources on buying a stake in fab, Qualcomm can buy advance nodes chips from Samsung and other chips from GF. Both should be desperate to expand their business due to Apple looking for another chip supplier than Samsung and GF looking to expand to Abu Dhabi.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.