Nvidia and Xilinx will increase the proportion of their outsourcing to Taiwan Semiconductor Manufacturing Company (TSMC) and United Microelectronics Corporation (UMC) and decrease their orders to IBM Microelectronics in the fourth quarter
Minor correction. Xilinx manufactures its FPGAs at UMC and IBM, not TSMC. Xilinx did announce it is shifting more 300mm production to UMC.
http://www.eetimes.com/story/OEG20030922S0027 - IBM acknowledges stumble in low-K
^^^ I found this quote particularly hilarious.
Most recently, an online Business Week magazine article quoted Willem Roelandts, the president of Xilinx Inc., as saying that while IBM is strong in technology, it needs to improve its manufacturing skills and move faster. [bold]A Xilinx spokesman later said Roelandts' remarks were taken out of context.[/bold] Xilinx is a high-profile IBM foundry customer.
http://www.eetimes.com/story/OEG20031020S0071 - After manufacturing stumble, Xilinx ready for run
This article isn't directly relevant to the topic on hand (0.13u IBM vs TSMC vs UMC), but it explains Xilinx's motivation behind picking UMC for producing their first 90nm FPGAs. At the time of UMC's (tape-out earlier this year), UMC's 300mm wafer capacity exceeded IBM, and this factor outweighed IBM's superior technology.
All other TSMC customers and TSMC s...te of what Nvidia says regarding TSMC and UMC
One of my coworkers claims that TSMC and UMC are roughly equal in terms of process reliability (wafer-lot to wafer-lot variation.) For many customers (but not all), the choice between TSMC and UMC boils down to political factors and pricing. In the past, UMC persued a 'virtual IDM' business-model (remember UMC's product-line used to include everything from DRAMs, SRAMs, x86-compatible CPUs, to 486 core-logic chipsets and VGA chips!) All things being equal: given a choice between 'T' and 'U', where 'U' is an IDM but 'T' isn't, customers prefer T.
But all things aren't equal. TSMC and UMC have their individual niche strengths. (UMC's customers accounts are heavily biased toward RF/wireless mixed-signal processes.) And UMC discounts their wafer-processing ~20% (compared to the TSMC equivalent)...or maybe it's TSMC that charges a +20% premium compared to UMC (because TSMC knows it can and still book its fabs.)
The experiences of the other manufacturers are irrelevant to nVidia - its their design and how it performs on TSMC or IBM's processes that matter to nVidia.
Exactly. And I might add, with billion-dollar companies like NVidia, ATI, Nintendo, Sony, etc. do with their contracting-arrangements is way out of line of the industry norm. Any of them can make new ways of doing business. Analyzing their behavior in terms of 'the average ASIC company' or 'the average electronics firm' is sheer non-sense. Example: When Microsoft announced they would be handling the manufacturing accounts for the primary chips in the Xbox (CPU, GPU, chipset, etc.), I thoguht to myself "Gee, Microsoft has ZERO previous ASIC experience with a project of this complexity...what a fine time to start with this project!" This operational-shift defies traditional engineering risk-management. If I read those announcements right, Microsoft's next console will be either a spectacular success or a spectacular failure (in terms of mfg cost.)
