Paul van Gerven
17 November

Never knowing what new applications Moore’s law will enable next, ASML decided to start expecting the unexpected.

After walking an audience of battle-hardened analysts through an even rosier growth projection than last year’s, ASML CEO Peter Wennink used his closing remarks to try and take away any lingering incredulity among his listeners. “I remember when this company had three billion in sales and we said that we could grow to five, we were laughed at. When we were at five, we thought that we could grow to ten billion. And we were laughed at,” he said at ASML’s Investor Day on 11 November. This little ritual was repeated a couple of times more, all the way to an expected 21 billion euros in sales this year.

So, Wennink suggested, perhaps this time investors could be a little more open-minded when ASML says it could double or even triple revenue by 2030. “Every time it turns out that the industry grows faster than we think.”

What we’re dealing with, Wennink explained, is “a structural underappreciation of the value of Moore’s law.” When semiconductor technology keeps gaining performance without becoming proportionately more expensive, you can expect the unexpected: new killer applications will emerge, but you won’t know what they are until they arrive. Or as Wennink put it: Moore’s law is about reducing cost per function. Functional gains unlock additional value. As long as the additional value exceeds the cost increases needed to reduce cost per function, chipmakers and their customers will keep creating new applications.

Investors can be confident this dynamic will continue for another decade at least. “The innovation pipeline is filled to the brim,” said one of the sheets in Wennink’s presentation, showing Imec’s transistor roadmap all the way to 2036. Yes, keeping Moore’s law alive is going to be ever more costly and ever more complex, and yes, shrinking chip structures won’t play as prominent a role as it once did, but it’s going to happen and it’s going to be worthwhile.

Hyper-NA

This reassuring message may have had something to do with CTO Martin van den Brink recently hinting at the end of the lithography roadmap, which would turn ASML into a cost-driven enterprise that’s not nearly as attractive to investors as it is now. In an interview with Bits&Chips, Van den Brink confirmed ASML was looking into a successor of high-NA EUV lithography but expressed skepticism about the economic viability of these hyper-NA systems. “For years, I’ve been suspecting that high-NA will be the last NA, and this belief hasn’t changed,” he said.

Admitting he was advised by his fellow executives to “tone it down a lot,” Van den Brink took a more nuanced view at the Investor Day’s Q&A session. “I always believe in technology, so I will believe we’ll get there,” he said about hyper-NA becoming a reality. But given the cost and complexity limitations that are piling on, the next-gen lithography scanner won’t come “whatever it takes. We have to constrain ourselves and look into cost-effective solutions that are competitive to whatever alternatives the customer has in terms of processing.” Such alternatives include double or multiple patterning using existing tools.

End of the world

Whatever happens with hyper-NA, the megagrowth projected by ASML for the remainder of the decade will be propelled by its existing offerings, supplemented with high-NA scanners from 2025 onward. As it turns out, perhaps unsurprisingly given Moore’s law’s unpredictable outcome in terms of commercial application, gauging that growth hasn’t been straightforward. Another important reason to organize an investor event only a year after the last one, therefore, was to fill in investors about growth markets that ASML had missed or assessed too conservatively.

ASML semiconductor forecast
In ASML’s revised forecast, the semiconductor market will reach almost 1.1 trillion dollars in 2030. Source: ASML

In the advanced-node arena, growth is faster than anticipated because of a better outlook for the server market, accelerated uptake of AR/VR technology and faster node migrations in smartphone and consumer applications. Sensors, power ICs, display and audio drivers – they’re increasingly moving to the advanced-ish nodes such as 28nm or even beyond.

Growth in the mature (28+nm) markets has yielded the biggest surprises for ASML, however. Wennink said he expected some more in the future, indicating it may not be possible to get a completely clear picture of market development. This is particularly true for what ASML calls “industrial electronics,” which covers basically everything that doesn’t qualify for any of the six other market categories (smartphone, PC, data center and so on) it employs.

One such mature-node application area that wasn’t as prominently on ASML’s radar as it deserves to be is the energy transition driving the demand for semiconductors. Power generation with green technologies such as wind turbines and solar panels requires thousands of euros of chips per megawatt. And that’s not even considering the smart grid technology that’s likely to take off sooner or later.

On the power consumption end, the energy transition is also a major driver of semiconductor growth: electrification of mobility. Electric vehicles have twice the semiconductor content on board compared to internal combustion engine cars, and that’s not even considering the rise of advanced driver assistance systems. These insights aren’t new, but ASML found that even more wafer manufacturing capacity will be needed to satisfy automotive OEMs and their suppliers.

On top of these market-driven adjustments, finally, a technological development is causing an upward correction: perhaps counterintuitively, die sizes in some applications are getting bigger. This is the result of chipmakers’ increased focus on energy efficiency. To increase energy efficiency without compromising performance, chips need to get bigger.

Pipe dream

In line with previous estimates, ASML confirmed it will increase annual capacity to 600 DUV (2.5x more than in 2020) and 90 EUV (3x) systems by 2025 or 2026, while ramping high-NA EUV to 20 systems in 2027-2028. Taking also into account productivity improvements, ASML will be able to deliver 3x more capacity in DUV and 5x more capacity in (0.33 NA) EUV compared to the start of the decade.

And if these projections prove to be one big pipe dream? In that case, ASML would be stuck with about 200 million euros per year in depreciation for expanding manufacturing capacity and a similar amount for investments in the supply chain, Wennink said. For a company making a profit of almost 6 billion euros last year, that wouldn’t be the end of the world.