This is the third part of a series answering the most frequently asked questions I get from Wall Street. Please read the previous two articles on Intel’s Manufacturing Lead and Intel’s SoC Challenge before flaming me in the comment section. First let’s look at why there is a foundry business and go from there.
The big semiconductor companies are the ones who started the foundry business in the 1980’s by renting out excess capacity to help pay for manufacturing costs. Unfortunately for them this unleashed a new era of semiconductor innovation and hundreds of fabless companies to compete with.
It all began when Dr. Morris Chang worked on a project for TI where manufacturing was done by IBM. This was one of the early semiconductor foundry relationships. Morris also noticed that top TI engineers were leaving and forming their own semiconductor companies. Unfortunately the capital requirement of semiconductor manufacturing was a gating factor. The cost back then was $5-10M to start a semiconductor company without manufacturing and $50-100M to start one with manufacturing.
The first two fabless companies, Chips and Technologies (C&T) and Xilinx, are profiled in the book “Fabless: The Transformation of the Semiconductor Industry” but here is the short version:
In the case of Xilinx, employees of a semiconductor company made a proposal internally for a programmable device. That proposal was rejected so the employees resigned and Xilinx was founded. Originally Xilinx’s business plan included building a fab to attract funding but in reality they were committed to the fabless model. C&T took the same approach with the “build a fab head fake” while negotiating fab space with Japanese semiconductor companies. Today Xilinx is the FPGA leader and it wasn’t long before C&T had more chips on a PC motherboard than Intel which lead to Intel acquiring them for $400M in 1997. All thanks to the fabless business model.
Unfortunately the “rent extra fab space from semiconductor companies” business model had one serious flaw. You were at the mercy of a company who did not really have your best interests at heart. Sometimes they competed with you, other times your business was not on the top of their priority list, and your IP was always at risk.
Today the Pure-play semiconductor foundry business model put in place by TSMC and replicated by UMC, SMIC, GlobalFoundries, and others is the backbone of the fabless semiconductor ecosystem. The ability to second, third, and even fourth source your design is a big part of that backbone. Unfortunately the fumbling of 28nm by everyone except TSMC has opened the door again to the big semiconductor companies who have excess fab space. For Intel this seems to be a tar pit if 14nm is any indication (Intel has no big volume customers at 14nm).
Of course I said the same thing about Samsung when they started in the foundry business 5+ years ago. Recent events however have changed my mind. Their ability to work with the fabless ecosystem and build relationships with other manufacturers (GlobalFoundries and STMicroelectronics) speaks volumes. Their aggressive pricing will also enable more design starts and overall industry growth. Samsung will be a major force in the foundry business starting at 14nm, absolutely.
Given that, the question you really have to ask yourself is: Can Intel survive a foundry turf war between TSMC and Samsung?