have released new information on their respective plans for the next few years. TSMC has announced its intention to double its 2011 R&D capital expenditure to $700 million, while it simultaneously spends $7.8B over the next year in order to increase its manufacturing capacity by approximately 20 percent. This is presumedly over and above what the company has spent thus far on constructing
its new "gigafab" foundry, Fab 15. TSMC began work on Fab 15 last summer, but the plant isn't scheduled to come online until 2012; TSMC is most likely building out capacity at an already established plant.
TSMC's production growth over the past few years. Information provided by DigiTimes.
TSMC's expansion plans are driven by the rapid growth of the cell phone / mobile device market. A record 47 percent of the foundry's revenue in 2010 was related to mobile sales; computer-related component sales revenue actually fell by six percent over the same period.TSMC's aggressive ramp plans are meant to make certain the company can meet estimated sales targets at the end of the year. "We stand to benefit from some of these mobile devices," CEO Morris Chang said. "The center of gravity of the computer market is shifting. Maybe we should say it's getting closer to us."
GlobalFoundries, meanwhile, is discussing its own expansion and scaling plans. In addition to Fab 8
, which is scheduled to come online in 2012, GlobalFoundries is investing $5.4B in expanding its own foundry facilities this year. Doug Grose recently spoke about the company's roadmap from 28nm production to the 20nm node.
According to Grose's presentation, GlobalFoundries will receive design/tapeout data from its first 28nm customer in the April/June timeframe. GlobalFoundries has recently finished incorporating Chartered Semiconductor Manufacturing's own factories into its own manufacturing portfolio; the company has also won Renesas Electronics Corporation as a recent partner and is discussing producing processors for Toshiba under a consignment basis.
GF will offer three different 28nm processes to interested customers: 28nm SLP, HP, and HPP. SLP (Super Low Power) is intended for mobile devices and very low-power processors, HP (High Production) is meant for chips targeted at midrange applications or that can exist within larger power envelopes—ARM's more aggressive
Cortex-A15 chips might be built on an HP process rather than an SLP. Finally, there's HPP, which stands for...Super High Speed according to Tech-On. No, we don't get it either.
Surprisingly, GlobalFoundries has also announced that it's skipping the 22nm entirely and jumping straight to 20nm gate-last production. We discussed GlobalFoundries' decision to switch to gate-last over gate-first earlier this month; the company's CEO again affirmed that this was a customer-driven decision, saying "We considered that our customers will require the gate last method for 20nm technologies."
It's not clear what this means for AMD and that company's future foundry plans. AMD hasn't said much about the shape of its process roadmap below 28nm; the company could simultaneously jump for 20nm and switch from SOI to bulk silicon, or GF may be planning a separate 22/20nm SOI process for its largest customer. If AMD does move to GF's next-generation 20nm (bulk or SOI) it could close its manufacturing gap with Intel, assuming AMD would be in a position to deploy 20nm parts while Intel was still shipping Ivy Bridge's 22nm successor, Haswell.
GlobalFoundries plans to begin deploying Extreme Ultraviolet Lithography (EUV) as earlier as the latter half of 2012, with volume production beginning in 2014/2015. In the meantime, GF will pair double-patterning and immersion lithography at 20nm with EUV reserved for an eventual 16nm process.
Over the next year, GF will expand production at both its 300mm and 200mm facilities, with 3D stacking technology coming online later this year or early in 2012.