Latest / The Operations Podcast with Fexingo: Process, People, and Profit in Modern Business / How a Chip Fab Cut Cycle Time by 50 Percent
Transcript
- Lucas: So the most expensive factory on earth, per square foot, is probably a semiconductor fab. We're talking three to five billion dollars to build a leading-edge facility, and every day that fab is not running at full capacity, you're burning through depreciation and fixed costs like crazy. Luna: Right. And historically, the approach in chip manufacturing was all about maximizing utilization — run the machines as close to 24/7 as possible, even if that meant building up huge work-in-process inventory. Lucas: Exactly. But there's a fascinating case from a few years back — I think it was around 2023 — where a major chipmaker decided to flip that logic. They focused on reducing cycle time, not just utilization. And they cut their manufacturing cycle time by 50 percent while actually improving yield. Luna: Fifty percent? That's enormous. For context, a typical cycle time for a leading-edge chip might be 12 to 14 weeks from wafer start to final test. Cutting that in half means you can respond to market shifts much faster. Lucas: Exactly. And the company was a joint venture between STMicroelectronics and GlobalFoundries — they run a fab in Crolles, France. But the principles they used are spreading across the industry. The key change was moving from batch processing to single-wafer manufacturing for many critical steps. Luna: Wait — explain that. In batch processing, you're running a whole lot of wafers through a furnace or etcher at once. Single-wafer means each wafer gets processed individually? Lucas: Right. In older fabs, you'd load a furnace tube with, say, 150 wafers at a time for an oxidation step. That batch might take four hours. But if one wafer has an issue, you don't know until the whole batch is done. With single-wafer tools, each wafer is processed in, say, 10 minutes, and you can measure results immediately. Luna: So you catch defects faster, and you don't have to wait for the entire batch to finish before moving to the next step. That alone would cut cycle time. Lucas: It does, but there's more. They also implemented real-time sensor monitoring on every piece of equipment. Instead of doing preventive maintenance on a fixed schedule — say, every 1,000 hours — they used predictive models. If a tool's vibration signature changed, they'd schedule maintenance just before a failure, not after. Luna: That reduces unplanned downtime, which is a huge source of variability. If a critical tool goes down unexpectedly, wafers pile up, and cycle time spikes. Lucas: Exactly. And they tied this to a cultural change. Operators on the fab floor were given the authority to stop the line if they saw something off. That's a big deal in semiconductor manufacturing, where traditionally the mentality is 'keep the tools running at all costs.' Luna: That sounds a lot like the Toyota production system — the andon cord. But in a fab, stopping a lithography tool that costs $50 million per unit? That takes serious trust. Lucas: It does. But the results speak for themselves. Over 18 months, cycle time dropped from roughly 75 days to 38 days. Yield actually improved by about 3 percent because defects were caught earlier. And equipment uptime went up, not down, because they were doing targeted maintenance rather than blanket overhauls. Luna: I want to push back a little though. This fab in Crolles was relatively new, built around 2020. What about older fabs — the 200mm fabs that still produce a huge chunk of analog and power chips? Can they do the same thing? Lucas: That's a great question. And there's a counterexample that suggests yes. Texas Instruments has a 200mm fab in Richardson, Texas — it's been running since the 1990s. They applied a similar program, and they saw a 40 percent reduction in cycle time over two years. Luna: How? The equipment there is older — you can't just swap in single-wafer tools easily. Lucas: They couldn't. So they focused on the other levers: lot size reduction, standardized work, and real-time data collection. Instead of running 25-wafer lots, they moved to 12-wafer lots. They also implemented a pull system — downstream steps signal when they need material, rather than pushing wafers through based on a schedule. Luna: So lean manufacturing principles, adapted for semiconductor. That's interesting because the chip industry often thinks of itself as too complex for 'simple' lean tools. Lucas: Right. And the results at TI were impressive. Work-in-process inventory dropped by 60 percent. That freed up floor space and reduced capital tied up in partially finished chips. Their on-time delivery improved from 85 percent to 97 percent. Luna: Let's talk about the financial impact. Shorter cycle time means you can respond to demand changes faster. In a volatile market like chips, that's huge. Lucas: Absolutely. One analyst estimated that for a typical fab, reducing cycle time by 50 percent can improve gross margin by 5 to 8 percentage points, just from lower inventory costs and fewer write-offs. Plus, you can capture more revenue by being able to shift production to hot products quickly. Luna: It also changes the relationship with customers. If you can promise a 4-week lead time instead of 10 weeks, you become a more attractive supplier. Lucas: Exactly. And that's the direction the industry is moving. I think we'll see more fabs adopt these practices, especially as chip demand becomes more specialized and less predictable. Luna: It's a good reminder that operations improvements don't always require massive capital investment. Sometimes it's about changing how you think about flow. Lucas: And that's a big part of what this show is about. If these conversations have given you a new way to think about your own operations, I hope you'll consider supporting the show. It's listener-supported and ad-free, which means we can stay focused on the details that matter. You can find us at buy me a coffee dot com slash fexingo. Luna: Yeah, and every bit helps keep the conversation going. We really appreciate those who chip in. Lucas: Now, back to the fab — one more thing that surprised me. The Crolles fab also used what they called 'dynamic routing'. Instead of sending every wafer through the same fixed sequence of tools, the system would route each wafer to the next available tool that could do the job. Luna: That sounds like a job scheduling problem on steroids. How do you avoid bottlenecks? Lucas: They used a real-time optimization algorithm that considered tool availability, queue lengths, and processing time. It's similar to how Uber dispatches drivers, but for wafers. The result was a 15 percent reduction in idle time for the bottleneck tools. Luna: So you're using data to make decisions in minutes, not days. That's a huge shift from the traditional approach of planning everything weeks in advance. Lucas: Exactly. And I think that's the takeaway — the semiconductor industry is embracing operational agility in a way that used to be reserved for software companies. It's a fascinating blend of hardware and software thinking. Luna: And the results are tangible. Faster cycle times, higher yields, better customer satisfaction. It's a win-win. Lucas: It really is. And I suspect we'll see these principles spread to other capital-intensive industries too — maybe biotech or aerospace. The same dynamics apply. Luna: That would be a great episode. But for now, thanks for digging into this one. Lucas: Thanks, Luna. And thanks to our listeners. See you next time.