How did TSMC get so good?
There is no simple answer, but we think there are a few factors that really stand out...By Jonathan Goldberg 47 comments
The big picture: By now, we are all familiar with the fact that TSMC is, by far, the most capable semiconductor manufacturer in the world, with all the entails for the industry and geopolitics. And as this reality sets in, many people have been asking us how did they get so good?
There are many good histories of this process, the most recent of which is Chris Miller's Chip Wars (which is very good), that cover the narrative of TSMC's rise, but we think the question really goes to deeper fundamentals. What capabilities does TSMC have that others lack? And can anyone else build those? Not surprisingly, there is no simple answer to this question, but we think there are a few factors that really stand out.
Guest author Jonathan Goldberg is the founder of D2D Advisory, a multi-functional consulting firm. Jonathan has developed growth strategies and alliances for companies in the mobile, networking, gaming, and software industries.
It is important to understand that much of TSMC's current advantage rests in intangibles. Everyone else can buy the same equipment. And their success is not the result of one genius. China's SMIC has poached hundreds of TSMC staff over the years, to no avail. Nor is the answer capital -- building an advanced fab is incredibly expensive, but many countries have sunk vast fortunes into futile attempts.
TSMC's advantage rests in years of learning and process development
TSMC's advantage rests in years of learning and process development. They know how to prepare chip designs for production, build them, and most importantly fix the inevitable bugs that appear. Most importantly, they are able to capture and retain all this learning, so they can do all of this at immense scale, repeatedly.
To be fair, TSMC has gotten an immense boost from the government of Taiwan. The government provided direct subsidies allowing TSMC to purchase the best machines available. This was incredibly important in the early days of the company, but remained important for long afterwards. As we will see, a big part of this story is the longevity and time scale of TSMC's growth.
A second important factor has been the suppression of the New Taiwan Dollar. The NTD has traded at roughly a 30% discount to the US dollar since the 1990's. Everyone knows this, but the US government has never acted on this currency manipulation, largely as an effort to support a regional ally -- check out Brad Setser's work for more on this.
Some will argue that this is not important – TSMC prices all sales in dollars, and its biggest expense, its equipment, is all denominated in US dollars as well. But that overlooks human capital. The bulk of TSMC's employees are based in Taiwan, and paid in NT, so the company is afforded a massive discount on their salaries.
All those intangibles we mentioned above are captured in the company's human capital, which the company has paid at a discount, which itself has compounded over 20 years. It is hard to overstate the importance of anything that compounds for that long.
TSMC produces more silicon wafers than anyone else, by a considerable margin.
Finally, and perhaps most importantly, TSMC produces more silicon wafers than anyone else, by a considerable margin. Practice leads to improvement and TSMC gets more practice than anyone else. It may seem counter-intuitive, but semis manufacturing is as much art as science.
Everyone else can access the same machines as TSMC, but when your manufacturing process is premised on constantly pushing the envelope of what is possible, there is no manual to explain what to do. TSMC's volumes over the decades have allowed them to build all the processes and knowledge bases needed to keep moving forward faster than anyone else. Again, this factor is self-reinforcing.
TSMC gets to advanced nodes before everyone else and thus gets better yields faster, affording them the ability to advance again, and again. This seems abstract, but consider that every new chip design is going to encounter bugs when it is first manufactured. The TSMC engineers have likely seen those bugs already and can address them quickly.
All of this provides TSMC not only with important advantages in capabilities, but also in terms of economics. There are plenty of places in the world with the ability to produce chips on leading edge processes, and give those academics enough time they can probably yield a perfect wafer, but TSMC can get there faster and in massive volume.
All of this is to say, that it is unlikely that anyone can truly catch up with TSMC anytime soon. It took them decades to build up these processes. And this helps partially explain why TSMC is now willing to build advanced fabs outside Taiwan. They know that their fabs in Arizona and Japan and Germany and wherever are going to remain dependent on the company's deep expertise back in Taiwan. The US may get a leading edge fab, but that does not mean the US can really do what TSMC does.