Paul Jacobs, Qualcomm chairman and CEO, said the company is “disappointed that it could not secure enough capacity to meet strong demand 28nm product”. In my opinion it is not capacity issue, but device performance issue because some time ago Mr. Jacobs made a similar statement that it has a strong demand for its Snapdragon S4 processor chips manufactured by TSMC, but TSMC could not deliver. The Snapdragon chips are originally designed to run the full blown Microsoft OS, Windows 8. In order to run Windows 8 Pro, high speed or high performance is required as well as low power. TSMC’s planar bulk 28nm can’t deliver such low power and high performance because when performance is increased, power or leakage current also increases. On the other hand, significantly lower power and higher performance compared with TSMC’s bulk 28nm are achieved by Intel’s Tri-gate 22nm based Ivy Bridge chips, thus ideal to run on Windows 8. Recently, Microsoft announced to adopt Intel’s Ivy Bridge processor chips to run its Windows 8 and may adopt Qualcomm’s Snapdragon chips for Windows RT which is a striped down version of Window 8. Ivy Bridge chips are in high volume manufacturing today. It is reported that Windows 8 based tablets and ultrathin notebooks become available third quarter this year. Intel’s FD (fully depleted) tri-gate technology is published at 2012 VLSI symposium for the first time and is at least three to four years ahead of its rivals. TSMC has no capacity issues for manufacturing the low power and low performance chips for ARM, but not for Qualcomm Snapdragon chips. That is why Mr. Jacobs is so disappointed. Qualcomm and ARM are engaging in vain with multiple foundries, hoping that they could deliver ultralow power and high performance chips. That is the reason why Mr. Jacobs is even contemplating to have its own fab, even though it is highly unrealistic but what he can do if the multiple foundries can’t deliver?
And... which of the Qualcomm products belong to the 28nm process? Which are the Qualcomm's products on a low supply?
I suppose this directly impact some mobile phone manufacturers which depend on their chips. But certainly this has to be only for this and perhaps the upcoming quarter. Unless this isn't enough time to make the new foundries capable of producing their chips. Last I heard this was an enormous job meaning it would take a long time. Around how long would this take?
Even in a supply constrained situation for its high-end products the company is making something like a 25% profit. Not bad considerting some of its customers struggle to get into double digit net profit percentages.
I don't know that Qualcomm has given a specific date on that, but it sounds like it's been longer than we initially thought. They announced in April that they had engaged other foundries, but I suspect it was a while before that. Mollenkopf made a comment on the call to the effect that Qualcomm typically works with multiple foundries at any given time. Seems like the strategy of casting a wider net is going to continue. But Qualcomm execs said, as they have said before, that they need to do a better job of anticipating demand at new technology nodes and planning for it.
Porting the design from one foundry to another is 9-12 months process if everything is going well evenif the process so called compatible to each other. If you face yields and reliability issues in the middle of, it could potentially drag you much longer. So I assumed Qualcomm engaged with the other two foundries at the end of 2011.
How long has it been since Qualcomm engaged with the other 3 foundries? I'd suspect there really won't be too great an impact to Qcualcomm's sourcing issues yet. Perhaps they got pricing concessions or similar out of TSMC as a result, though that won't show up on their books yet, either. Down the road a little (December?) this could work out well for Qualcomm, though now they will have to juggle libraries and flows for 4 fabs not one. Any "unified" libs and flows will be full of compromises.
Blog Doing Math in FPGAs Tom Burke 15 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 ...