During the 1980’s, I worked briefly in the petroleum industry developing new additives for gasoline and diesel fuels. The work for me was new and exciting, but my colleagues constantly bemoaned the fact that we were in a “sunset industry” – the high times once enjoyed by the research community supporting petroleum refining were over, and my coworkers made sure I knew it. I left the petroleum sector a few years later in 1986, making the jump into that decade’s shining star for industrial expansion: semiconductor manufacturing. Cutting-edge technology, rapid growth, prestige, what else did one need? Now, thirty years later, I’m beginning to wonder as Yogi Berra once did, is it “deja vu all over again?”
Consider the following companies consumed during mergers that took place in the past two years (courtesy of 24/7 Wall Street ):
- Freescale Semiconductor
- Integrated Silicon Solution
- Broadcom (name changed from Avago back to Broadcom post-merger)
- EZchip Semiconductor
- Hutchinson Technology
- Mattson Technology
- OmniVision Technologies
- Fairchild Semiconductor
24/7 Wall Street goes on to list four additional companies likely on the acquisition target list: IDT, InvenSense, MaxLinear, and Power Integrations. It goes far beyond mergers, though.
Another way to look at the changes within the semiconductor manufacturing industry is to look at fab closures. IC Insights published in 2014 a pair of graphs detailing closures between 2010 and 2014:
As you can see, 83 fabs were closed after the “Great Recession” of 2008. It is clear that the older technologies of 200 mm wafers and below took a greater hit than the current state-of-the-art 300 mm fabs. It is equally clear that there is a regional effect within East Asia (with the notable exception of Japan), where only a tiny fraction of the total fabs closed. Are these mergers and fab closures signaling a significant change in the character of the semiconductor manufacturing industry – is it following the petroleum industry into the sunset? And if so, how far away is that sunset? Or is this merely a natural maturation of the industry?
For the past three decades, the semiconductor industry has been considered an “evergreen” industry, one driven by an odd paradox: the prospect of massive profits for the successful players paired with an endless drive to lower the cost of manufacturing. This sector is well known for Moore’s Law: “the number of transistors per square inch on integrated circuits has doubled every year since their invention.” Moore’s Law has shown no signs of slowing, as continual innovations have improved chip design, and constantly evolving manufacturing techniques and materials have provided the world’s consumers with an array of electronic devices that consistently outperformed the previous year’s devices. This year’s cell phones have ever more functions than last year’s models, and accomplish them faster; your televisions are thinner, have sharper images, and are increasingly woven into your digital life; most amazingly of all, mainframe computers have shrunk to table-top PCs, laptops, and tablet computers. You can do more with the device on your wrist or in your pocket than what was once accomplished with a 1950’s mainframe that filled a room.
But (and there is always a “but”) these improvements came with a cost to the manufacturer: heavy and near-constant investment in manufacturing technology. New fabs that once cost in the millions now routinely cost over $2 billion. This alone has led to some of the changes depicted above in the fab closure by wafer size graph. It is cheaper on a per-chip basis to manufacture on a 300 mm (and in the foreseeable future, 450 mm) wafer than the older sizes of 200 mm and below. This is one of the biggest changes driving both fab closures and M&A activity. A 200 mm company trying to compete against a 300 mm competitor knows that the proverbial clock is ticking for their fab, and that 200 mm fab is ultimately destined for eventual closure. And of course, the regional effects of fixed costs derived from insurance, interest expense, property taxes, utilities expenses and depreciation of assets, to say nothing of variable costs such as materials and labor are the true arbiters of whether manufacturing stays in Europe, America and Japan versus Korea, Taiwan or China.
Another change that has played an important role is the technology used to create a new chip. This is a categorical effect that has had very different outcomes across the industry as a whole. For example, if the chip in question is a logic chip such as a CPU or ASIC, or a memory chip such as Flash or DRAM, then the improved chip performance that comes from advanced manufacturing techniques will drive out of business entirely any lesser-performing chip as well as the companies dependent upon making them. On the other hand, less-demanding applications such as those found in the growing field of the Internet of Things (IoT) might well manage comfortable with many older manufacturing technologies. And of course, there is a “but” here, too – profit margins for advanced CPUs are far higher than those for the much simpler IoT devices. Thus, even though it may be cheaper to make an IoT chip, the economics of making them may simply be insufficient to keep the manufacturer in question in business alone. The other “but” for manufacturers of advanced chips is the massive investment in new manufacturing technologies required. Thus, a potential manufacturer must choose between a high margin product with extremely high investment costs or low margin products with low investment costs but also low ROI. We can see from the rapid loss of fabs and companies via M&A activity (or outright closure) that fewer and fewer companies are willing to make either choice, and as such are leaving the industry.
So, is semiconductor manufacturing a sunset industry or not? The answer, like the technology, is complicated. For some regions like Europe, without dramatic changes the answer may well be a qualified yes. For North America and Japan, it is unquestionably not yet a sunset industry, but perhaps well into their metaphorical day, not yet evening but nearing it; that sunset for most of Japan’s and America’s fabs probably lies within the next 10 years. There are many stalwarts in both countries (the likes of Toshiba and Intel but also many niche manufacturers) that will continue on well into the 21st century. Meanwhile in other parts of Asia, it is still noon with a busy afternoon still in their future. Lastly, it is well worth noting, these comments are meant only for semiconductor manufacturing. There is a wholly separate set of manufacturing industries such as solar, LED, and MEMs that are at different points in their growth trajectories than semiconductors. Their sunsets, for one, are far into the future – for now.