Latest / The Operations Podcast with Fexingo: Process, People, and Profit in Modern Business / How TSMC Became the World's Indispensable Chip Factory
Transcript
- Lucas: Luna, I want to talk about a factory that most people have never set foot in, but whose output touches nearly every piece of electronics you used today — from your phone to your car to the server this podcast is probably hosted on. Luna: You're talking about TSMC, the Taiwan Semiconductor Manufacturing Company. I've been meaning to dig into their operations for a while. Lucas: Exactly. TSMC is the world's largest dedicated independent semiconductor foundry. They make chips for Apple, NVIDIA, AMD, Qualcomm — basically everyone except Intel and Samsung, who do their own manufacturing. And here's the number that blows my mind: TSMC produces something like 90 percent of the world's most advanced logic chips — those are the ones with transistor geometries at 7 nanometers or smaller. Luna: Ninety percent. That's almost monopolistic. And it's all made in a handful of fabs on one small island — Taiwan. Lucas: Right. So the operational question is: how did TSMC get to this position? And more specifically, how do they achieve such consistently high yields on the most complex manufacturing process humans have ever devised? I want to focus on yield rate today, because it's the single most important metric in semiconductor manufacturing. Luna: Yield rate — that's the percentage of chips on a wafer that come out functional, correct? Lucas: Exactly. A modern chip like Apple's M-series processor — which is used in MacBooks and iPads — has tens of billions of transistors. The wafer is a disk of silicon, typically 300 millimeters in diameter. You lay down dozens or hundreds of identical chips on that wafer, run it through hundreds of processing steps over several months, and at the end you test each chip. If even one transistor out of billions is defective, that chip is trash or gets downgraded. So yield rate directly determines cost and profitability. Luna: And TSMC's yield on their leading-edge nodes — like 3 nanometer — is reportedly over 90 percent. That's extraordinary. Lucas: It is. In the early 2000s, a good yield on a new process node was 60 to 70 percent. Competitors like GlobalFoundries and UMC have often struggled to get above 80 percent even on mature nodes. TSMC's ability to hit 90 percent plus on the bleeding edge gives them a massive cost advantage, which in turn lets them invest even more in R&D and equipment. It's a virtuous cycle that's very hard to break. Luna: So what's their secret? Is it the equipment — like those EUV lithography machines from ASML that cost over 150 million dollars each? Lucas: Equipment is part of it, but it's not the differentiator. Every leading foundry buys the same ASML machines. The difference is process discipline — the way TSMC manages the thousands of individual steps in chip fabrication. They have a doctrine called 'copy exactly.' The idea is that once a process is proven to work in a pilot line, every subsequent fab must replicate it exactly — same machine settings, same chemicals, same temperature ramps, same timing down to the second. Luna: That sounds simple in theory, but incredibly hard in practice. Fabs are built in different locations, with different water quality, different ambient humidity — even the vibration from nearby traffic can affect lithography. Lucas: Right. So TSMC goes to extraordinary lengths to control those variables. When they built Fab 12 in Hsinchu and later Fab 15 in Taichung, they essentially cloned the environment. They treat each fab like a clean room inside a clean room. Operators are trained to follow procedures with almost no deviation. There's a famous story from the early days: Morris Chang, TSMC's founder, once said that if you want to change the length of a coffee break on the factory floor, you need to run a qualification test first. Luna: That level of rigor is almost military. But it clearly works. Lucas: It does. And there's another layer: TSMC runs what's called a 'shuttle' service. They offer multiproject wafers where dozens of different chip designs from different customers share a single wafer. This lets smaller companies prototype without paying for a full mask set — which can cost millions. But it adds huge complexity to the production scheduling. You have to manage hundreds of different designs, each with different process requirements, flowing through the same line simultaneously. Luna: So the operational challenge isn't just making one chip well — it's making thousands of different chips well, on the same equipment, with the same high yield. Lucas: Exactly. And that's where their information systems come in. TSMC has developed internal software that tracks every wafer through the fab, predicting yield issues before they happen. They use statistical process control at every step. If a parameter drifts even slightly, the system flags it and engineers intervene. They've been refining this data feedback loop for over three decades. Luna: Three decades of data. That's a moat that competitors can't replicate overnight. Even if someone built a state of the art fab tomorrow, they wouldn't have the accumulated knowledge. Lucas: Precisely. And this brings up a geopolitical point. Because TSMC's manufacturing is concentrated in Taiwan — which sits in a region of high geopolitical tension — there's enormous concern about supply chain resilience. Governments around the world are pouring billions into trying to bring advanced chip manufacturing home. The US CHIPS Act, for example, is funding TSMC's fabs in Arizona, plus Intel's expansion, plus Samsung's Texas facility. Luna: But even with those investments, replicating TSMC's operational culture is a huge challenge. I've read that TSMC's Arizona fab has faced delays and labor issues — partly because they're trying to impose the same 'copy exactly' standards in a different labor market with different work habits. Lucas: Right. TSMC sent hundreds of experienced Taiwanese engineers to Arizona to train local workers, but there have been cultural clashes. The pace, the attention to detail, the willingness to work weekends — it's different. And yield rates at the Arizona fab, when it eventually ramps, will likely be lower than in Taiwan for years. That's not a knock on American workers; it's just that operational excellence at this level takes a long time to build. Luna: It makes you wonder whether the 'made in America' chips will ever truly be competitive on cost and performance. Or if the natural monopoly of TSMC in Taiwan will persist. Lucas: That's the billion-dollar question. And it's not just about cost — it's about trust. Apple, NVIDIA, AMD all rely on TSMC for their most critical components. They've built their product roadmaps around TSMC's process cadence. Switching foundries would be a multi-year, multi billion dollar ordeal. So TSMC's operational reliability is directly tied to the stability of the entire tech industry. Luna: Speaking of trust and reliability, that's actually something we think a lot about on this show. We deliberately keep these episodes free of ads — no sponsor breaks, no commercial interruptions. And that's only possible because some listeners choose to support us directly. Lucas: Yeah, we've never wanted to compromise the conversation with a mid-roll ad for a mattress company or a meal kit. So if you find these episodes useful — if something we discussed today moved your work forward or gave you a new way to think about operations — you can support that choice at buy me a coffee dot com slash fexingo. That's buy me a coffee dot com slash fexingo. Luna: And we genuinely mean it — no pressure. Just a low-key way to keep the show exactly as it is. Now, back to TSMC — one more angle I want to explore before we wrap. Lucas: Sure. What's on your mind? Luna: I'm curious about the human side. TSMC employs over 70,000 people. Half of them are engineers. How does a company maintain such a rigid process culture across such a large workforce without stifling innovation? Lucas: Great question. TSMC balances this through what I'd call 'structured innovation.' They have a clear hierarchy: process integration engineers define the baseline recipe, and then process engineers in each fab optimize within tightly controlled bounds. If someone wants to change a step, they propose it, run trials on a small batch, and if the yield improves, the change is adopted globally. Luna: So innovation is channeled through the system rather than happening spontaneously on the floor. Lucas: Exactly. And there's a strong culture of continuous improvement — kaizen, borrowed from Toyota. Workers at all levels are encouraged to suggest small tweaks. But every suggestion goes through validation. This prevents the chaos you might see in a less disciplined environment. Luna: It's almost like TSMC treats its entire operation as a single, massive machine that needs to be tuned millimeter by millimeter. Lucas: That's the perfect metaphor. And it's why, despite all the talk of chip nationalism and reshoring, TSMC's operational advantage is incredibly durable. The combination of equipment, process discipline, accumulated data, and human culture creates a moat that won't be crossed quickly. Luna: So what's the one takeaway you want listeners to remember from this episode? Lucas: That yield rate isn't just a technical metric — it's the output of a whole operating system that has been refined over 35 years. And that system is so tightly coupled to its specific environment — Taiwan's ecosystem, its workforce, its supplier relationships — that replicating it elsewhere is one of the hardest problems in modern industrial policy. Luna: Hard, but not impossible. And certainly worth trying, given the stakes. Lucas: Agreed. Next episode, we'll look at a company that took a completely different approach to operational excellence — one built on decentralization rather than central control. Luna: Looking forward to it. Until then.