Some time ago, Shang-Yi Chiang, a veteran in the semiconductor industry, accepted an invitation from the California-based Computer History Museum to tell the sotry about his services at Taiwan's TSMC and China's SMIC, and the transcripts of his interview were recently released. Many people who have made comments in the media on his revelations have probably never had a conversation with Chang, shared a meal with him, or attended a TSMC technical forum or other professional events.
I've done all of these things over the years. I have data from our researchers, and I have many friends who have worked at TSMC and SMIC, so I'm going to share my experience with you and try to provide a different perspective so that you can understand more about TSMC.
Chiang asserted that Intel has not lost by a wide margin. I think that is closer to the truth. Chiang also said that TSMC is not that great that that it is just doing a lot of things right. In fact, this involves issues such as business models, corporate cultures and market positioning. We can find out some truths about the business operation in light of Chiang's comments; we shouldn't take his remarks as gossips that we can forget about later.
When Chiang revelations about TSMC's breakthroughs in 180nm and 130nm node technologies actually show us how TSMC managed to overtake competitors. He said that IBM was leading in the development of these two nodes and invited several major companies to form an alliance to jointly develop new technologies. But TSMC was left out.
During the development of those two nodes, the process technology was transitioning from aluminum to copper, and Chiang was TSMC's R&D VP.
From 1998 to 2000, DIGITIMES hosted SEMICON Taiwan conferences and invited Jim Morgan and Dan Maydan, chairman and president of Applied Materials, respectively, and David NK Wang as keynote speakers.
At that time, Applied Materials Taiwan collaborated with DIGITIMES to produce several feature stories concerning copper processes. TSMC just made the right judgment on the design architecture and achieved a thin-margin lead. It was around those generations that TSMC began to expand its R&D team significantly, not only to close the gap with UMC, but also to become a truly world-class company.
After 2015, TSMC spent more than 30% of its revenue on capital expenditures, and after entering the EUV generation in 2019, the proportion of capital expenditures to revenue began to exceed 40%, and now it is even investing 50%, which is really keeping competitors at bay.
When the industry approached the 10nm and 7nm generations, Intel defined the specifications, and was even a generation ahead of TSMC. We can tell from the technical specifications announced by the two companies. In terms of the number of transistors per square millimeter, Intel's 10nm specification is similar to TSMC's and Samsung Electronics' 7nm specifications, while Intel's 7nm specification is even more advanced than TSMC's and Samsung's 5nm.
But the key is what Chiang describes: TSMC is not particularly great, but it has managed to do it well for each generation of specification. Intel may have been standing tall but it lacks top customers. Intel can now hardly swallow its pride and will continue to pursue technological perfection. But it is missing the point.
In the foundry sector, customer structure is key to building an ecosystem and recovering your investment. If you don't have money, you can't do anything.