by Bob Haavind, Editorial Director, Solid State Technology
Jan. 31, 2008 – Dr. Morris Chang, legendary founder of Taiwan’s TSMC, the world’s biggest foundry, sees healthy growth ahead in the world economy even if there is a 6-9 month slump in the US, with a steadily expanding role for semiconductors. “Thirty years ago our goal was to make semiconductors ‘pervasive,'” he stated in a talk at the recent International Electron Devices Meeting (IEDM) in Washington, DC. “Now it to make them ubiquitous.”
He pinpointed major shifts in the semiconductor industry landscape, identifying disruptive innovations that led to the new course for the industry. The biggest value now, Chang believes, is in system architecture and circuit level software. In basic ICs the value is in design, in microprocessors it’s in subsystems, and in SoCs (systems-on-chip) it’s in systems.
While Moore’s Law has been a tough taskmaster for the industry, and steady shrinks have expanded total markets for chips, it has also led to what Chang sees as a shift in the business model for the industry. From 1952 until recently, the industry was built around semiconductor companies that served a variety of systems companies, in sectors such as defense, autos, telecom, and consumer electronics. Back in 1985 it took about $200 million to build a new fab — an amount that was not easy to raise. At that time there were less than 50 fabless companies, Chang pointed out, and they could not get reliable fab service because chipmakers that leased them space would take it back when business picked up. Then the foundry concept began to emerge, and by 1995 there were 340 fabless companies bringing in about $6B, though still a fairly small part of the industry. Meanwhile, chipmakers had to continue to invest ever-larger sums to build bigger fabs, and they needed their markets to grow sufficiently to justify new fabs.
By 2000 there were 750 fabless companies bringing in $17B — but by 2006 this had grown to 1400 companies with $50B in revenue, about 20% of the industry total. The foundries had grown to become major manufacturers, with about 30% of their total revenue from making chips for the integrated device manufacturers (IDMs) as well as fabless companies.
In the early days foundries offered standard processes, but to make more advanced chips, they had to start customizing processing to achieve the specifications for individual process designs. This means that foundries must now collaborate with designers in achieving continuous improvement in fab processing. “Now designers and fab people must work in partnership under the same roof,” Chang said. To achieve this, foundries need to put more effort, offering intellectual property (IP) that can make chipmaking more design-friendly.
EDA, SoCs, competitive threats
Following his talk, Chang chatted with reporters, explaining his views on the world economy, changes sweeping the semiconductor industry, and what would be needed for chipmakers to succeed in the future.
He suggested that foundries have become the biggest power users of EDA tools as they work cooperatively with chip designers. While individual process technologies are needed for each fab customer, he said, these are often based on generic technology with some variations in order to be economically realistic.
In SoCs, Chang feels that embedded memory is a key to success, but it is still necessary to have additional memory off-chip. He thinks there is a long way to go in developing successful SoC methods, and he sees 8%-10% growth a year over the next 10 years.
To be successful, he explained, foundries need to provide differentiable services, particularly as there is a shrinking number of IDMs so that more chips must be made by the foundries.
When asked if he saw China as a threat to Taiwan’s dominance in foundry work, Chang replied that he never thought that the Chinese would become big competitors. “Foundry work is a tough business,” Chang said, and he believes that TSMC has an unassailable lead not just because of its IP, but also because it retains a large number of highly skilled people. The entry barriers are so high, he believes, that semiconductor manufacturing, and particularly foundries, have almost become a closed industry.
After getting his BS and MS degrees in mechanical engineering at MIT, Chang earned a PhD in electrical engineering from Stanford. He spent 25 years at Texas Instruments, and was president and COO of General Instruments before founding TSMC in Taiwan in 1987. — B.H.