The Founding of ASML Part 2: Oil to Electric
The new company starts to fix its problems
If chips make the world go round, ASML may be the closest the multi-trillion-dollar global tech industry has to a linchpin. The Economist
Most of the transistors that have ever been built on this planet have been built with ASML scanners and Zeiss optics. Carl Zeiss
This is Part 2 of the story of the founding of ASML. You can find Part 1 here, and I’d recommend reading it before this post if you haven’t already. In this short series we are tracing the unlikely origins of a company that today has a market cap of over $260 billion1 (more than Intel and AMD combined) and is now central to the production of modern leading-edge semiconductors.
Just a reminder that if you’re interested in becoming a paid subscriber then you can still get a massively discounted annual subscription. Help support us and get access to the supplemental post on this article with more discussion and lots of relevant links.
When we left the story in Part 1, it was March 1984 and ASML had just been formed as a joint venture between the Philips and ASM International (ASMI). Philips was a huge electronics conglomerate which was starting to struggle financially. ASMI was a much smaller but fast-growing supplier of semiconductor manufacturing equipment.
Before we pick up the story of ASML again, it’s worth setting the context. 1984-6 saw the introduction (from announcement to mass production) of the Intel 80386 microprocessor, first made using a 1.5 µm2 technology node and including 275,000 transistors on an 104 mm squared die.3 Later versions of the Intel 386 would shrink to use a 1 µm technology node.
The ability of firms like Intel to keep pace with Moore’s Law shrink ‘features’ on their chips to use technology nodes at 0.7 µm and below by the early 1990s would crucially depend on their lithography tools, often built by external suppliers such as GCA David Mann in the U.S. and Nikon in Japan, leading firms in the market for these tools in the mid 1980s.
But each ‘shrink’ in the size of the ‘features’ on the chip created an opportunity for new lithography technology and for new suppliers. At the same time, those firms that were still using tools developed internally saw that further development of these tools would be too expensive for them to support. If they wanted to continue to develop the technology they would have to find external customers, and that meant spinning off into a separate company. For Philips, that led to the joint venture with ASM International and the creation of the ASML joint venture between Philips and ASMI.
Draining the Oil
Philips contributed its technology in the shape of its stepper4, a machine that could repeatedly project the image of an integrated circuit onto a silicon wafer. ASMI would contribute cash and its marketing expertise to help sell the new machines to its existing customers around the world.
There were major problems though. The Philips stepper design had a major flaw. It used an oil-based hydraulics system to move the wafer around. Although the system was sealed, one oil leak could play havoc with the super-clean environment of a semiconductor fab. Customers saw this as a massive risk. Martin van den Brink, one of ASML’s first employees, would recall fixing leaks, just before demonstrations of the steppers to potential customers:
After all the clean-up was done and the customers came in, it was smelling like an old car shop. The oil was still evaporating so it was a very difficult time.
ASML’s management eventually came to the conclusion that the oil-based stepper, known externally as the PAS 2000, was impossible to sell.
We can see the PAS 2000 in action with (I think) it’s hydraulics in this early film from ASML.
If ASML was to become a successful business, then it would need to replace the oil-based hydraulics with a system that used electric motors. Fortunately, Philips research laboratory, known as NatLab, already had such a system in development. However, the rights to this electrical system were not included in the contract that created the new company.
The other major problem was that the amount of cash that ASML had available to it to develop its next generation of systems was tiny. ASMI had provided $2m to the new joint venture, but Philips contributed seventeen unsellable oil-based steppers and very little cash. ASML would soon need a lot more money.
For some of the new ASML team, now working from huts outside Philips offices in Eindhoven, the situation looked hopeless.
Seeing the Opportunity
If all these problems weren’t enough, then ASML’s management soon realised that they had another one. The company’s products relied on sophisticated optics with multiple high-precision lenses focusing the image of the chip down onto the silicon wafer.
For many years, Philips had relied on and built a close relationship with the Paris-based firm CERCO. Now that relationship was starting to look like a liability. So far, the lenses provided by CERCO had generally been good enough, but as the size of features on the wafers shrank, better optics would soon be needed. ASML’s key competitor GCA David Mann was using optics produced by Carl Zeiss based in Oberkochen in Central Germany5, and the higher quality of these Zeiss lenses was a major advantage for them.
But one key issue for the company was resolved by the time it launched in March 1984: finding a chief executive. Gjalt Smit combined a deep scientific background, with a master’s in magnetohydrodynamics6 followed by a doctorate in astronomical plasma physics, with varied business experience. He’d joined Philips after completing his academic work, spending time working in Italy, before moving to work for telecommunications conglomerate ITT back in the Netherlands.
He was approached by headhunters at the end of 1983 to check his interest in leading a new joint venture between Philips and ASM. He soon hit it off with Arthur del Prado, Chief Executive of ASMI, and despite some misgivings about his former employer, accepted the role.
When he joined the company, Smit was impressed by the quality of the technology and by the enthusiasm of key staff. But he was dismayed by the progress that has been made in getting that technology to market.
He attended the SEMICON West conference in San Matteo in 1983 and received overwhelmingly negative feedback from potential customers. Competition from GCA David Mann and Nikon was strong and customers were not interested in what ASML had to offer.
But the needs of the fabs were changing. The industry was about to enter the VLSI era, and that would require lithography that could create images at less than 0.7 µm resolution. It was clear to Smit that that he could deliver this. There was a real opportunity for his company if they could get in ahead of the competition.
One key issue is soon resolved. Philips agreed to sell the electronics motor system that will replace the hydraulics in the PAS 2000. Even better, it was sold to ASML for the modest fee of $930,000.
Smit developed a timetable for the successor to the PAS 2000. It would be called the PAS 2500. The first working prototypes were planned to be delivered by 1 January 1986, and it needed to be ready for mass production six months after that.
Whilst the ASML team were working to develop their technology, the semiconductor industry was heading into recession. GCA, previously the market leader in steppers, was also starting to suffer strong competition from Japanese firms and especially from Nikon.
The opportunity was clear, but ASML needed to deliver with the PAS 2500.
New Lenses
If the PAS 2500 was to be competitive, then ASML needed competitive optics.
Over the previous years, ASML’s management had several discussions with Carl Zeiss, who were already producing lenses for ASML’s competitor GCA. The relationship with GCA was starting to get more difficult though, and diversifying their customer base would be beneficial.
On one final visit to CERCO in Paris, the French firm asked for cash to support development of the next generation of lenses. It was the final straw for the ASML management team and the decision was made to switch to Zeiss. The PAS 2000 was the last machine to use CERCO optics and the PAS 2500 would use Zeiss lenses, the start of an close relationship bet ween the two firms that would last until the present day.
A New Home
The company also needed to find a permanent home away from its temporary location in huts outside Philips offices in Eindhoven. The new home needed to be purpose built for the special nature of ASML’s business. It had to have clean rooms for the development, assembly, and testing of ASML’s machines. It even had to have special foundations to reduce the impact of vibrations from the external environment.
But where to locate this new facility? The town of Veldhoven, just West of Eindhoven, was chosen, but the choice soon faced a legal challenge from other towns, upset at losing the high-quality jobs that ASML would have provided. Veldhoven has great transport links as it’s close to two motorways and near to Eindhoven airport, but the town’s planners stretched zoning requirements in giving ASML permission to build.
Bulldozers started work at the new Veldhoven location in October 1984, even as the challenge was waiting to he heard at the Dutch Supreme Court. Fortunately for ASML, the legal challenge failed, and the company went ahead and built its new headquarters as planned.7
PAS 2500 Falls Behind
But if building the headquarters was going to plan, then development of the PAS 2500 wasn’t.
When planning the development of the PAS 2500 the ASML team realised that they had to make a clear break with the approaches that they had used at Philips in the past. They had taken almost a decade to go from the first Silicon Repeater in 1973 to the first delivery of the PAS 2000. Now they would need to accelerate development of the new model.
The answer is to do things in parallel. They split the machine into subsections that could be developed separately, and set out to build five prototype machines at the same time. It was an expensive approach, but one that potentially took several years off the time needed to develop the machines.
By the middle of 1985 with development of the PAS 2500 was well underway, but management started to realise that it wouldn’t be possible to deliver according to plan at the start of 1986. Over the following months, the project fell further and further behind.
A huge amount was resting on the PAS 2500 and customers, particularly Elcoma, were expecting delivery on time. Elcoma’s management made it clear that if they didn’t receive ASML’s steppers on schedule then they would switch their purchases to Nikon instead.
Selling the PAS 2000
ASML also needed to do something with the unsellable PAS 2000s. With the electric table technology now available, it might be possible to salvage these by removing the hydraulics and retrofitting electric motors. With the new machine free of oil risks (and still called the PAS 2000) Smit and his team felt they can go on the offensive.
In time for the SEMICON West 1985 conference, they mount a bold advertising campaign focusing on their key advantage when compared to the competition: speed. Headlines included, “74 wafers per hour” and “ASM Lithography presents some wafer stepper numbers GCA and Nikon would like to ignore” along with “Count on us to keep you in the chips”.
At the start of 1986 ASML’s US sales team sold an electric PAS 2000 to Monolithic Memories Inc (MMI)8 a small memory manufacturer.
PAS 2500 Delivered
At the start of 1986 Smit was briefed on the state of PAS 2500 development. The start of January deadline had been missed, but he hadn’t been aware that the project was struggling.
So Smit gambled. The project was failing with its existing management, so he reshuffled. Senior members of the team were moved out, and more junior engineers were brought in.
And the gamble worked. The new team pulled together, working much longer hours to get the project back on track.
The deadline was pushed back by a few months, but the team were now confident that they can deliver to Elcoma and to the SEMICON conference in the Spring.
The first PAS 2500 left Veldhoven on 7 May 1986 to travel to the SEMICON West show in San Mateo, California. There it managed to expose five hundred wafers on its first day on public display.
The second PAS 2500 was then delivered to Elcoma, where it would be used on the MegaChip project, which was Europe’s attempt to try to compete with the US and Japan in chip production, and to build one megabit memory chips. At Elcoma ASML engineers continued to make refinements and work to make the machine truly ready for production.
And that first PAS 2500 delivered sub-micron resolution. Quoting a later ASML history.
Using g-line light (from a pressurized mercury lamp) with a wavelength of 436 nm, the first machine in the 2500 series delivered a working line-width resolution of 0.9 microns.
Before long, orders started to arrive from external customers. First off the mark was Cypress Technologies, a recent Silicon Valley startup. They ordered two PAS 2500s to be delivered in August 1986. Other firms soon follow suit.
The End of the Beginning
By the middle of 1986 ASML finally had competitive products and was starting to acquire real paying customers from outside the Philips group.
From a small base of shipments in 1984 and 1985, sales in 1986 and 1987 started to grow dramatically.
There was still a long way to go before it takes the shape that the company has today. The joint venture ownership still had to evolve into a more sustainable shape, but ASML’s long journey as a viable and successful independent company had begun.
Supplemental Post
For more commentary and background on this post, including loads of links on ASML and photolithography as well as sources for this article then, please see the supplemental post here which is exclusive to paid subscribers.
In the supplemental post we also explore the similarities between ASML’s origins and those of another key firm in the the modern semiconductor industry.
This post would not have been possible without Rene Raaijmakers comprehensive and detailed history of ASML’s early years. If you want to know more about ASML’s origins, then it’s the definitive source.
Market cap of $264 billion on 20/1/23 according to Google Finance.
µm is a micrometre also known as micron which is one millionth of a metre.
For comparison the Apple M1 System on Chip of 2019 squeezes 16 billion transistors onto a die of 119 mm squared - that’s an increase of over 58,000 times (source Semianalysis).
It’s important to draw a distinction between ‘steppers’ than exposed the whole of the die and ‘stepper-scanners’ that replaced them which only exposed part of a die at a time. Stepper-scanners are now commonly known as steppers. Its confusing!
Carl Zeiss relocated there after World War 2 as its original home in Jena was in the Soviet Zone and later became part of DDR, commonly known as East Germany.
The study of the magnetic properties and behaviour of electrically conducting fluids
Today ASML’s World Headquarters remains in Veldhoven and it employs over 19,000 in the town.
Monolithic Memories was founded by former Fairchild engineer Ze'ev Drori, who would later go on to become CEO of Tesla Motors. MMI merged with AMD in 1987.