How ASML and Nikon brought immersion lithography into customers’ fabs

In February 2007, Nikon announced that it had won an immersion lithography order after a year of evaluation against competing equipment at a customer’s factory. It also reported selections by several customers during the previous year. [1]

ASML was already expanding deliveries. Its 2007 annual report recorded cumulative shipments of more than 70 immersion systems. This was not a contest in which only one company understood the new technology. [2]

Where, then, should we look for meaningful differences?

This analysis focuses on the immersion transition of 2004–2007, with a follow-up on throughput improvements reported in 2008. Banseog’s view is that reconciling competing performance requirements and making a system work reliably in customer production mattered as much as its headline specifications. Equipment architecture, test methods and coordination with suppliers help explain how that happened.

Compare equipment from the same period

Immersion lithography places water between the lens and the wafer to print finer circuits. The principle is relatively simple. Turning it into fast, repeatable production without introducing defects is not.

Nikon shipped the NSR-S609B in January 2006 with a numerical aperture, or NA, of 1.07. NA affects an optical system’s ability to resolve fine features. Delivering an immersion system above 1.0 was a genuine technical achievement. [3]

Scroll the table horizontally to read all columns.

Nikon’s figures come from its shipment announcements; ASML’s come from its annual reports. These products belong to different generations and reached customers in different months. The table is not a controlled performance comparison. [4] [5] [6]

Reaching a specification first, shipping the next generation and achieving productivity in a customer’s fab are distinct accomplishments. A higher NA does not by itself establish better defect rates, availability or delivery times.

The useful question goes beyond “who was first?” What must a customer still verify before using the equipment, and who will help complete that work?

ASML separated what had to change from what could stay

TWINSCAN measures one wafer’s position and height while exposing another. When ASML introduced immersion, this architecture allowed water to be used at the exposure stage while retaining dry measurement. The arrangement appears in its 2006 annual report, not only in later recollections. [7]

The new technology did not require the customer to change an established measurement process at the same time. Separating the scope of change preserved what had already been validated and focused testing and development on the new uncertainties. That does not quantify a reduction in development cost, but it identifies where new problems entered the system.

Water still created a difficult trade-off. Bubbles and droplets could cause defects as the wafer moved quickly. Slowing it down might reduce defects at the expense of productivity.

ASML’s 2023 retrospective describes how fluid-dynamics research and supplier design work contributed to a hood that controlled water around the lens. The company reports that improvements to the hood helped increase wafer speed while reducing defects. This account covers several subsequent years of development; it is not evidence of a quantified improvement confined to 2004–2007 or a performance gap against Nikon. [8]

The significant connection is between the defect problem, a change in fluid-control design and improvement in two conflicting outcomes. That tells us more than a list of prestigious partners.

Nikon was addressing similar problems. Local Fill dealt with water behavior and defects, while Tandem Stage used stages with different functions to improve throughput and accuracy. Nikon’s 2008 report also described modularization and common components across models. Architecture and reuse were not capabilities unique to ASML. [9]

A wafer returned to the customer proved something different

An improvement inside the machine still had to work in the customer’s process.

In December 2004, TSMC announced electrically functioning 90-nanometer SRAM made with ASML’s immersion prototype. For one critical layer before metallization, wafers were split between dry and immersion exposure, then returned to TSMC to complete the remaining process. TSMC reported comparable yield, device characteristics and defect levels, with almost twice the yield-related depth of focus for immersion. [10]

This did not validate immersion volume production across every layer. It did move the evidence from an optical image to a working device that had passed through the customer’s remaining manufacturing steps.

That distinction matters commercially. A customer buys the prospect of preserving device function and yield in its own process, rather than an impressive image in isolation.

Adjacent processes also entered the evaluation. SCREEN’s November 2005 announcement described nine months of joint immersion-process evaluation with ASML using the RF3. The new RF3i added cleaning and drying functions to reduce watermark defects, with plans to connect it to ASML equipment and defect inspection. The joint evaluation had taken place; installation of the new configuration and customer demonstrations were still plans. [11]

The test boundary expanded from the lens to water, then to resist, cleaning and drying, and finally to the customer’s completed device. How many of the interfaces a customer would encounter had been tested before delivery? That is a more revealing question than the machine’s specification alone.

Nikon’s evaluation at a customer’s factory deserves the same recognition. Both companies had customer validation. A useful comparison must examine its scope, the stabilization of production performance after adoption and the subsequent expansion of supply.

Having suppliers is different from making decisions together

A further clue lies in the documents exchanged with suppliers.

A 2023 study by Taji and Enami, based on interviews with both companies, describes ASML’s separate requirements for an EPS specifying performance and a TPS specifying test methods for its light-source supplier. Instruments, measurement intervals and lower-tier supplier schedules were also considered. Nikon is described as documenting key target values while discussing others with its supplier. The study also examines the role of ASML system engineers in reconciling component requirements. [12]

The interviews took place in 2014–2017. The authors acknowledge that they could not obtain details of Nikon’s internal information sharing. The findings therefore cannot establish how every project operated in 2006, or that Nikon lacked coordination capabilities. [13]

Even with that limit, the comparison is useful. It directs attention from the existence of supplier relationships to whether the parties can judge results on a common basis.

Suppose a component meets its target but the complete machine underperforms. Someone must decide under which conditions it should be retested, and who can alter the requirements imposed on other components. This is Banseog’s interpretation of the evidence, not a reconstruction of a particular incident. Such coordination costs disappear from a comparison of headline specifications.

What matters is the ability to understand technologies beyond the company’s boundary and adjust conflicting requirements. Neither in-house production nor outsourcing solves that problem automatically.

Deliveries grew, but one mechanism does not explain the market

ASML’s 2007 report records cumulative immersion shipments above 70 systems and XT:1900i adoption by several customers. Nikon’s 2008 report likewise describes S610C deliveries to major regions producing advanced semiconductors. Both companies moved into customer adoption. [14] [15]

Our emphasis is on the connection between ASML’s retention of dry measurement, improvements to water control and adjacent processes, customer-device testing and follow-on equipment deliveries. Together, these describe a concrete route from immersion technology to a growing business.

They do not establish how much of the market-share difference that route explains. Installed customer relationships, investment timing, pricing, delivery and availability could also matter. Nikon’s wins in individual evaluations and ASML’s expanding deliveries must be read together.

A test win establishes competitiveness under those conditions. The next questions are whether the improvement transfers to other customers’ processes and whether adoption and delivery can be repeated. A performance advantage in one setting cannot simply be counted as a market-wide advantage.

What other choices were available at the time?

Advising Nikon to validate equipment jointly with customers would have repeated what it was already doing. The decisions ahead included both manufacturing capacity and stabilizing machine throughput after customer adoption.

In its 2007 results discussion, Nikon described a smooth introduction of the S610C and forecast sales of 21 units and shipments of 30 units, using full production capacity, for the year ending March 2008. Further expansion would depend on market conditions. Sales and shipments are different measures, and these figures were forecasts. [16]

Later in 2007, Nikon reported that throughput improvement had taken longer than expected and that corrective work was under way after identifying the causes. In May 2008, it said the problems had been resolved and planned throughput was being achieved. Customer selection and reliable production performance were separate milestones. [17] [18]

Capacity decisions therefore depended on both demand certainty and the ability to meet required throughput consistently. One path was to wait for greater clarity before expanding. Another was to secure customers with defined acceptance criteria and schedules, while advancing throughput improvement and capacity in parallel. Waiting reduced idle-capacity risk but could sacrifice delivery opportunities; expanding earlier increased exposure to delayed or canceled orders.

The evidence does not establish that the second path would have been better. That would require information on order certainty, progress in throughput improvement and equipment-production bottlenecks. It does show why stabilizing customer production and deciding when to expand supply were connected strategic choices.

The same standard applies to ASML. Rising shipments alone do not establish durable advantage if installation and stabilization take too long, or improvements do not transfer across customer conditions.

What should we look for at a corporate turning point?

Historical analysis is useful when it identifies evidence that could have changed a decision at the time. During the immersion transition, that evidence included what a new technology changed in the existing process, what device tests established, how component requirements were reconciled, and how throughput and supply developed after adoption. Insights from later interviews must not be treated as information available to outsiders then.

The question Banseog would carry into another company is straightforward:

As our strength improves, does adoption become easier for the customer—or do we pass new coordination burdens to other components and processes?

That question also changes the search for partners. It starts with the conditions blocking adoption and the parties able to change them, rather than a ranking of technical reputations.

ASML and Nikon’s immersion development does not diminish the importance of precision. It shows how much beyond the machine must be understood and coordinated before precision becomes customer production.

Independent analysis distinguishing contemporary public records from later recollections and research. Its scope is immersion development, validation and supply, not a single-cause explanation of the companies’ entire competitive history, including EUV.