Overview of the Foundry Model
A dedicated alliance with Apple that gives the company first access to 20nm production and/or a dedicated fab could fundamentally redefine the foundry-customer relationship and have serious impact on both TSMC's other customers and its competitors.
Here's what Chang actually said. The question came from Dan Heyler, an analyst with Bank of America.
Q: (Heyler) Did you think you need to dedicate some times to product specific areas that are very high volume?
A: (Chang) Actually yes. I think that’s almost a natural outcome the way market is trending. I think that they are going to be larger customers, and now it makes complete sense to dedicate a whole fab to just one customer and hold that – to hold fabs in fact to just one customer. Now remember, we made our mark in serving many customers. In fact, that’s a really part of our secret source of success. The ability to serve many customers to their satisfaction and we’ll still will retain that capability, but there are customers that are getting bigger and bigger. So it makes sense that we dedicate a whole fab or even more than a whole fab to just one customer
The easiest way to explain why Chang's remarks are so important is to recap the evolution of what we call the foundry model. From the 1960s to the late 1980s, the overwhelming majority of semiconductor design firms owned their own production facilties. This changed over time as manufacturing techniques became more standardized and the industry as a whole synchronized with what came to be known as Moore's Law. The rising cost of node deployment and the rapid, capital-intensive factory upgrades that were required led IDMs (Integrated Device Manufactures) to seek ways to outsource manufacturing and design. Thus was born the foundry model. Small-to-medium companies that couldn't afford the high capital expense of keeping up with Intel or IBM would be able to pay a "pure-play" foundry to handle their manufacturing. By aggregating the business of many manufacturers, the foundry would be able to afford to keep up with the few remaining vertically integrated IDMs. TSMC, founded in 1987, was the first pure-play foundry.
The foundry model worked well for twenty years because technology ramps were regular, power consumption and die size scaled consistently, technology roadmaps were unambiguous, and higher production costs were offset by increased yield per wafer. For TSMC, that began to change at 40nm. The 28nm ramp has been smoother than 40nm's legendary problems, but it still took the foundry nearly a year to ramp 28nm volume production, with Chang predicting that the foundry would "fully meet" demand for 28nm products in Q1 2013. All current indications are that the 20nm ramp will be at least as difficult; Chang told investors he expects pilot 20nm production in 2013 to be "very small" with 2014 a "ramp year" for 20nm SoC production.
The high cost and long ramp time of new nodes have led companies like Nvidia to call for increased cooperation and risk-sharing on new production. On the one hand, TSMC appears to be acknowledging that its customers are concerned that the old foundry model is no longer able to properly address scaling challenges. A direct partnership with Apple in which the latter gains early, guaranteed access to 20nm production in exchange for a substantial amount of up-front financing, however, probably isn't what anyone else had in mind.