The Founding of ASML - Part 1: The Philips Era
How ASML emerged from the Dutch electronics giant
As the only supplier of equipment needed to manufacture the most advanced semiconductors, ASML has found itself at the centre of a global battle for access to the means to make these chips.
ASML’s most advanced machines use what is known as Extreme Ultraviolet Lithography or EUV to create tiny patterns on silicon wafers with features measured in tens of nanometers. Its customers include the leading semiconductor manufacturers, TSMC, Intel, and Samsung. Each of these firms pays ASML over $100 million for each of its latest EUV machines.
In December 2022, ASML had a market value of over $240 Billion, making it one of Europe’s most valuable companies.
But where did ASML come from? In this multipart post, I’m going to track the story of ASML from the years before its founding to its first major commercial success.
Spoiler alert: it’s a remarkable story. Even if you have limited interest in semiconductor manufacturing in general and lithography in particular, it shows that even the most unpromising start can lead to world-changing success.
It’s also interesting that ASML started in circumstances that are so different from those of most firms in Silicon Valley. There was no venture capital involved, and ASML emerged from a huge industrial conglomerate. It was originally a 50:50 joint venture between two much larger firms. Whilst TSMC can trace some of its founding DNA (for example in the shape of Morris Chang) back to the US, ASML’s roots are firmly European.
Philips
If ASML’s roots are European, those roots emerged from the Dutch soil of Philips Electronics. Headquartered in Eindhoven in the Netherlands, Philips was a true industrial conglomerate. Founded by father and son Gerard and Frederik Philips in 1891, by the 1970s it had interests in lighting, in audiovisual products and in healthcare systems. Philips was a famously inventive company, developing the compact cassette tape in the 1960s, the first video cassette recorders in the 1970s and then laserdisc and the Compact Disc, jointly with Sony, in the 1980s. (For sports aficionados, the leading Dutch soccer team PSV Eindhoven – Philips Sport Vereniging Eindhoven or Philips Sports Association – also emerged from Philips in 1913).
Many of these innovations were nurtured at Natlab (short for Philips Natuurkundig Laboratorium—Philips Physics Laboratory) a major research facility also based in Eindhoven. In 1975 Natlab employed over 2000 people. Natlab has been compared to AT&T’s Bell Labs in the US. Its research wasn’t limited to the development of products for Philips, but also included fundamental research on electronics.
In the 1970s, Philips manufactured semiconductors through its Elcoma (Electronic Components and Materials) subsidiary. In 1975, the company acquired Signetics, founded by employees of Fairchild Semiconductor in 1961. The acquisition of Signetics made Philips the second-largest manufacturer of semiconductors in the world. If you make semiconductors in volume, then you have a strong interest in the machines that you need to manufacture the semiconductors.
The Silicon Repeater
The story of ASML is largely one of its technology, and that technology starts in Natlab with the development of the first Silicon Repeater. The Silicon Repeater was a machine that would move a silicon wafer so that an image could be repeatedly projected onto it by an optical mechanism held above the wafer. To make sense of the Silicon Repeater, though, we have to understand how integrated circuits were being made before it arrived.
In the 1960s and 1970s, integrated circuits were manufactured using a technique known as ‘contact lithography’. First, a ‘reticle’ with an enlarged version of the pattern to be created on the integrated circuit would be made by cutting into a thin plastic sheet, known as Rubylith. This pattern would then be reproduced at a smaller size on a master mask – typically glass coated with a thin layer of photo resist – by exposing the mask to blue light shone through the reticle. This process would be repeated across the whole of the mask to create a series of identical patterns. The photoresist would then be developed to create the required patterns – a series of images across the master mask.
The master mask would then be replicated onto a number of ‘working plates’ coated with silver halide. These working plates were then used to ‘contact print’ the required pattern onto a resistive film on the final silicon wafers. The pattern on this resistive film would then enable the required circuits to be etched onto the surface of the wafer.
This approach had some significant limitations. The contact between the working plates and the wafer led to damage to the plates over time, so that after as few as ten impressions the plate would become useless. Firms often had to create very large numbers of working plates, which made the process very expensive. Even when the working plates were quickly discarded, the yields from these wafers would be as low as 10%.
This all changed with the development of the Perkin-Elmer Micralign. After a long gestation period with research, in part, funded by the US Air Force, the Micralign came onto the market in 1973.
The Micralign used a series of mirrors to project an image of the working plate onto the surface of the wafer. Part of the image of the plate would be projected onto the corresponding part of the wafer, with a motor moving the source and the target (on the wafer) of the projection around whilst the mirrors remain stationary.
The fact that there was no contact between the plate and the wafer meant that the plates were no longer damaged in the process. This extended their lifetime massively, thus reducing costs. At the same time, yields increased significantly, to 50% or higher.
At around this time, the team at NatLab were investigating a different approach. Rather than start with an image of the complete wafer on a mask and then scan this onto the wafer section by section, they would take a single image of the integrated circuit. They would then move the wafer around so that this image would be reproduced on different parts of the wafer as required.
This had several advantages. The design of the Micralign limited the smallest feature size to be at least 2 micrometers. The Natlab machine could shrink the integrated circuit image down to a much smaller size on the surface of the wafer, which in turn allowed much smaller feature sizes.
The technical challenges of doing this were huge. First, the wafer had to be precisely located before each image of the integrated circuit was created. This in turn generated two formidable requirements. The machine had to locate precisely where the wafer was placed and, depending on this information, quickly move the wafer to the correct place for the next image to be projected onto its surface.
The precision was vital. If an image was misaligned relative to other images on the wafer, then the resulting integrated circuit would be useless.
The machine consisted of an optical column containing an arrangement of lenses through which light was shone from the die image and onto the wafer, together with a table to move the wafer around.
Two Natlab engineers, Herman van Heek and Gijs Bouwhuis, took up the challenge after seeing how wasteful the contact lithography approach was at Elcoma. In May 1971 they proposed their approach to Philips management and gave their new machine a name – the ‘Silicon Repeater’ – the generic term for this type of machine would be ’stepper. It took more than two years to develop the technology and build a working machine. The first Silicon Repeater (known as SiRe 1) was completed by the engineers of Natlab at the end of 1973.
Meanwhile, the Micralign, with its much improved yields, had entered the market. It would become a huge commercial success. It allowed a three-inch wafer to be exposed in around one minute and enabled the development of Very Large Scale Integration (VLSI) integrated circuits. Perkin-Elmer sold over 2,000 of the machines to all the biggest firms in semiconductor manufacturing. Famous microprocessors such as the 6502, 8086 were all manufactured using the Micralign and it was the Micralign that allowed these new designs to be sold cheaply. It thus formed part of the foundation of the emerging home and personal computer markets.
Over this period development of the Silicon Repeater was progressing but slowly. A team led by Steve Wittekoek refined the technology of the SiRe 1 to develop the SiRe 2. Key among the changes were fundamental improvements in how the wafer was handled. The repeater could now measure and rapidly correct not only the wafer’s location, but also its rotation. This in turn meant that the optical column hanging over the wafer didn’t need to move as the wafer moved underneath it.
But there was a fundamental challenge in gaining market acceptance for the SiRe 2. The hydraulics in SiRe 2 were contained in a sealed system, rather than having the oil run freely and be caught in a tray. But the use of hydraulics still meant oil under pressure. Oil in turn means the potential for oil leaks, and oil once released into the super-clean atmosphere of a fab spells disaster.
So the engineers of Natlab started to develop a better alternative, with electric motors powering the mechanism that moves the wafers.
Commercialising SiRe
Development of the SiRe was originally a research project, but by 1978 the time had come for it to be commercialised. That meant handing it over to one of the business units within Philips Science and Industry Division (S&I). The unit that accepted it was headed by Wim Troost. Troost would become a key supporter of the development of steppers at Philips over the coming years.
The handover was by no means straightforward, with a clash of cultures between the two parts of the Philips empire: Natlab focused on research and development and S&I on production. The machine passed to Troost’s unit had limited documentation, and so the S&I team had a lot of work to do to get it ready for mass production.
By now, other firms were also looking at similar approaches. In 1978 US firm GCA David Mann introduced the first commercially available stepper. With a list price of $450,000 the DSW4800 (Direct Step to Wafer) uses Zeiss optics and although it was less sophisticated than the Philips machines it’s an immediate commercial success. The first GCA stepper was soon sold to Texas Instruments, and by 1981 GCA had over $100 million in revenues from its stepper products. Other firms soon followed suit. In 1980 Nikon introduced the first stepper manufactured in Japan.
Philips had the most advanced technology, with greater precision than the competition, albeit one with a serious drawback, in the shape of its hydraulics. It had zero presence, however, in the market beyond Philips subsidiaries whilst others are selling hundreds of machines. Troost has to set about changing this and to find the first external customer for the Silicon Repeater.
At the end of the 1970s, IBM is at the forefront of semiconductor manufacturing, achieving milestones such as being the first to produce 64Kbit DRAMs at their manufacturing site at Burlington, Vermont. Keen to expand his customer base beyond Elcoma, Troost saw IBM as a natural customer for equipment as advanced as the Silicon Repeater. Troost approaches IBM, who agree to take a SiRe 2 with one condition: it must be delivered by June 1982.
Getting the SiRe 2 working well enough to deliver to IBM was challenging. Troost had to bring in the Natlab team to work with his own to make progress. The machine was finally delivered a few weeks behind IBM’s original deadline, on 1 July 1982.
With a potential external sale, the SiRe 2 needed a name that could be used with customers. It became known as the PAS 2000 (PAS for Philips Automatic Stepper).
Meanwhile, in 1982 Philips started on a programme of divestment of non-core businesses. The Silicon Repeater, with its ongoing problems, scepticism from Elcoma and no immediate prospect of meaningful external sales, is one of the products targeted for disposal.
Troost started to search for external partners to keep the technology alive. One by one, possible partners - Cobilt, Perkin-Elmer and Variant – were approached, express interest and fell away.
But then a partner appeared on Philips doorstep. Advanced Semiconductor Materials International (ASMI) was based in the village of Bilthoven, northeast of Utrecht. ASM was run by Arthur del Prado who had overseen rapid growth in the company’s sales and profits. Del Prado had been pushing Elcoma for years to let his firm make more of their production machinery that they use. Finally, Philips came to him with the stepper business. After some discussions del Prado proposed that he takes over the wafer stepper business. But despite the perilous state of the business, Philips refused. The reason? ASMI was just too small.
Finally, Troost and other members of Philips management realised that ASMI was their last chance to save the stepper business. They met del Prado again in 1983 and work to hammer out a deal. Del Prado was insistent on one point throughout: that the new company should continue to have access to Natlab’s expertise.
In September 1983 an announcement was made that ASMI and Philips were launching a joint venture. But at this stage there was just a memorandum of understanding and not a final contract. The two sides still have to agree the final terms.
It takes until March 1984 to finalise the contract between the two firms. The two firms would each own 50% of the new firm – now called ASM Lithographic Systems B.V. – and would each contribute $2.1m to the joint venture. There was one catch, though. Only ASMI would be putting the full cash amount into the new company. Philips would be counting the physical assets it's contributing as part of the $2.1m. This includes 17 PAS 2000 machines, most of which were still under construction.
Forty-seven staff would transfer from Philips to the new company. However, after a battle between Philips and the works council, and in a clear indication of their lack of confidence in the new venture, they gained the option over the next four years to return to Philips if things don’t work out at ASML.
An Unpromising Start
We end the first part of the story here. ASML starts off in a perilous position. It has some fairly unhappy employees, many of whom are based in a few wooden huts outside a Philips office in Eindhoven.
The seventeen PAS 2000 machines it has received from Philips look likely to be impossible to sell due to the risk posed by the hydraulics to any fab where the machines are installed.
It has one external customer with a PAS 2000 on trial. IBM isn’t entirely happy with the PAS 2000, but the speed of the machine has impressed the team there, who want ASML to fix its flaws.
The new company also has access to Natlab, who are continuing to work on an electrical system to replace the hydraulics in the PAS 2000.
What it doesn’t have is enough cash to successfully continue development of the PAS 2000 into a product that stands a chance of competing in the stepper market. ASML’s management would soon estimate that they need $100m to catch up with their rivals.
It also doesn’t have a CEO. Filling that role would be key if the company was to make progress.
It’s hard to imagine a more unpromising start for a new venture. How did ASML turn the situation around? Please subscribe to get the next part of this story and other updates direct to your inbox.
Thanks for reading. Please comment either below or on Hacker News.
Postscript
I’ve played down somewhat the politics of the ASMI / Philips relationship. There is an interesting post on bit-chips.nl which looks at ASMI’s lobbying effort to get Philips to work with the smaller firm.
References
This post has drawn heavily on ‘ASML’s Architects’ by René Raaijmakers, which is a hugely detailed history of the first years of ASML and of the development of this technology at Philips in the years before ASML was founded. It’s hard to imagine a more comprehensive account of this story. The flip side of this is that it's a long read and, currently at least, only available in a (relatively expensive) Kindle edition. It’s strongly recommended if you like your semiconductor history in depth.
Raaijmakers also has a really excellent article at bits-chips.nl where he goes into more detail about the fact that electric table technology was not included in the assets that Philips provided to ASML at the start of the joint venture even though this technology was in development at Natlab. It’s a great piece that provides a detailed exposition of the problems that the joint venture partners gifted to the new company, but also some spoilers for the second part of this post.
Chip History has a short note from Richard George who worked on the installation of the PAS 2000 at IBM including some photos of the workings of the machine.
For more on the Micralign see:
https://www.chiphistory.org/landmarks/micralign-projection-mask-alignment-systems-1970
Photo Credits
Natlab
Johan Bakker, CC BY-SA 4.0 <https://creativecommons.org/licenses/by-sa/4.0>, via Wikimedia Commons
Great writing!
I know this is skipping ahead, but how would you characterize ASML's current relationship with ASMI?