This sounds more like a joint venture than a foundry agreement. Apple provides the capitol, TSMC provides the process IP and the manpower. Apple gets 100% of the product, TSMC gets ??? dollars per wafer. Would it be possible for Apple to make a similar joint venture with Intel, or GF, or IBM, or ??? I don't see why the partner has to be a foundry at all, an IDM might make more sense.
Put 2 + 2 together and it equals Apple in this case. Here's the logic:
First mathematical term: capacity buy
Second mathematical term: capital to buy it
and the only answer is Apple that has both. TSMC and Samsung are arch-rivals in foundry business. Likewise I have to think that Apple sees the conflict of interest in sourcing parts and litigating against the same company, Samsung. Also, knowing Apple's Supply Chain capacity buyouts, if they want to plunk money down to buy it they want an iron clad guarantee they will get what they pay for, no questions asked, and that it will go into a place where there are no channel conflicts.
I think before 2012 is done you will have an announcement about a dedicated Apple fab operated by TSMC.
I wonder if this opens up the risk that the fabs may end up becoming unique in some way that is optimized for the big customer. So let's say TSMC's Qualcomm fab get's adjustments one way and Apples' the other. They it makes it harder for them to adjust to run one companies waffers on anothers fab. Sort of a pandora's box.
They better say they want to work for Qualcomm. I do not think it is a model they can sustain since a single customer will put risks on their bottomline with cyclical changes we notice in the business.
Dedicating one fab to one customer model can lead to retain high volume customers but this model is susceptible to fluctuations in the demand from customer. So, running fab for multiple customer makes more sense to me.
The cost of leading-edge fabs is astronomical and growing. But this is the business that TSMC is in: laying out that capital with the expectation that they will get it back and reap a healthy return on investment when the fab ramps. If TSMC starts building dedicated fabs, presumably with money up front from the customers that they will be dedicated too, doesn't that represent a fundamental change to the foundry business model? What happens if one of the dedicated customers lands in real financial trouble and needs to liquidate assets? Could it sell its stake in a fab operated by TSMC?
One would expect that before TSMC dedicates any facility to a single customer regardless of its location, it would secure long-term contracts with said customer. From the customer's standpoint, it might make sense to contract for a dedicated fab that handles 80% or less of the customer's expected volume, and then purchase any additional needs from the general-availability fab(s). If such a business case was presented, TSMC might consider building a dedicated facility wherever it made most sense for the customer, e.g., close to customer's main board assembly operation(s).
What TSMC is really saying is that risks are getting too much even for TSMC as a few customers becoming so large(Mainly Qualcomm, Apple and could be Nvidia), they can't put all the CAPEX alone. TSMC wants to share the financial risks. I don't know what the CAPEX splits going to be but TSMC will retain the rights to serve many customers in the same fab.
PS- What TSMC really needs is just the right amount of capacity at a given node to be available exactly when it's needed. And that is challenging. But it doesn't need to be a dedicated fab here and a dedicated fab there. Qualcomm acknowledged that 28-nm demand far exceeded its expectations. If that hadn't been the case, chances are they would have negotiated with TSMC to secure adequate capacity up front. TSMC could have built the capacity if it knew it was needed.
Blog Doing Math in FPGAs Tom Burke 23 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...