The foundry dollar goes further now:
What's the difference between a foundry and a full-service ASIC supplier? Put another way, what is it that Taiwan Semiconductor Manufacturing Co. (TSMC) can't do for its customers?
An ASIC supplier provides libraries, offers design services, either does the place-and-route itself or checks the customer's work, "signs off" on the final design, and assumes some liability if it doesn't work. A foundry relies on the customer for the design, offers a "hot" process, but assumes no liability for a faulty design.
And how will any of that change once IP cores come into more widespread use? Justin Wang, a planning manager at TSMC (Hsinchu, Taiwan) said "the age of IP is coming but we aren't really sure ourselves what it will mean to TSMC, though we are discussing it a lot internally."
TSMC had $1.432 billion in 1996 revenue, and net income was an astounding $705 million. However, with an intra-island foundry price war between TSMC and UMC, and Japanese and American companies bidding for business, TSMC's earnings are expected to drop by 52.9 percent for 1997, on a 10 percent drop in revenues.
Will portable IP put the dazzle back into TSMC's business? Already, IP developers are asking TSMC to test and verify their cores for use in TSMC's process. TSMC may contribute to a Web site that offers information on core vendors. But most of it is in the planning stages: Wang said TSMC is studying how to develop a formal program to verify IP in its processes.
John Y. Chen, a senior manager, said most of TSMC's efforts are being spent to bring up its 0.25-micron process by late 1997, and at its Camas, Wash., fab in early 1998. And 0.18-micron design rules are expected by the turn of the century. Early in the next century, about 50 12-inch megafabs will feed much of the electronics industry, and two-thirds of them will be producing either MPUs or DRAMs. The others will be foundries, using 12-inch slices at 0.18-micron and better designs rules, Chen predicted.
Increasingly, TSMC is licensing building blocks: primitives, and megacells such as the DSP Group's Oak and Pine cores. Customers need cores that have universal appeal, including MPEG decoders and MCUs, as well as PCI and USB cores, which have been proven to work in a TSMC process.
"Most of the time we don't even know until later that a customer has used an outside IP. But timing is always the issue with an unknown portable IP. If we run into a problem with a design that has three cores, how do we know who to contact? And creating a synthesis script is always very painful," said Yea-fu Tsao, director of TSMC's ASIC division.
None of this says that TSMC will greatly increase its full-service ASIC business, which now accounts for about 10 percent of its total sales. To change its basic foundry model would risk alienating its system-house customers.
But to keep an edge over new entrants, TSMC must broaden its IP offerings and learn how to work with customers who need help.
If TSMC can provide a steady bridge between the IP vendors and the designers, that would make TSMC's second decade as exciting as its 1987-1997 period most certainly was.
Happy birthday, TSMC. Now, get back to work! |
