ASML is a critical supplier of semiconductor
lithography machinery for foundries like Intel and TSMC. In my video discussing TSMC's $28 billion
capital expenditure, I briefly discussed their situation. Their CEO said in an earnings call that
they can make 50 high-end EUV lithography machines a year. That's it. Without those machines,
the foundries cannot churn out more 5nm chips. So why not make more of these machines?
ASML itself has thousands of suppliers making parts that end up into its machines.
Coordinating and integrating all of these parts together into a single smooth-running
machine is immensely challenging. In this brief video, we will continue our deep
dive into ASML and look at how the company puts together centi-million dollar lithography machines
for multi-billion dollar semiconductor companies. And how EUV makes it so much harder. You might want to watch my ASML explainer
video first before starting on this one, just in case you need a refresher. ASML is a global, sprawling company
- befitting the nature of its work. The factory is based in the
Netherlands but there are additional manufacturing and R&D sites
located in Connecticut and California. Its products are more like airplanes than you
might at first expect. In my video about COMAC, I mentioned that Boeing and Airbus no longer
make the majority of the parts that they put together. Instead they have evolved into having
a system integrator role where they select, procure, and put together the outputs
of different suppliers in their network. ASML is the same. Their lithography machines,
like those of the TWINSCAN variety, are built out of independent modules. Kind of
like the Megazord in Power Rangers. 90% of the components that go into these modules
are made by the 500-600 outside companies within ASML's outsourced supplier network. These
are extremely critical companies and differentiate ASML's supplier network from its competitors
Canon and Nikon, which tend to do things in-house. 300 of these suppliers are
located in the Netherlands. Another 100 are in other parts of Europe,
mostly Germany. And the rest are largely in the United States. Coordinating
within this global network is critical. Just as TSMC leans heavily on ASML to provide
training and advice on how to use their products, ASML in turn leans heavily on its suppliers
on how to best use their components. For instance, their strategic alliance
with German company Carl Zeiss AG. Zeiss is a specialist in precision optics and
helps produce the lens - a very critical part. So why outsource? Isn’t that the reason
why Boeing can’t make good stuff anymore? Isn’t outsourcing a scourge of western capitalism? Well, ASML might from time to time purchase
a supplier company to bring critical, unique expertise in-house. For instance, the acquisition
of Brion Technologies, a US firm specializing in computational lithography, and Cymer, another
US firm specializing in lasers. It happens. But for the most part, ASML wants to remain
in a system integrator role and leave as much of the actual manufacturing to its
suppliers. There are a few reasons for this. First, ASML can get its module components without
needing to learn how to be the best in the world in making that specific part. A single lithography
stepper can incorporate over 1,600 individual components. 200 of those are not cheap to
procure in terms of cost or production time. The company cannot focus on those components' minutiae while at the same time
maintaining a big picture view. Second, this allows ASML the flexibility to
change and adapt themselves to changing technology trends. If something happens and ASML has to
radically change the way it creates its machines, like as it did with EUV, then it has the
flexibility to make that change - without having to deal with the sunk cost of having invested
resources to develop a now-defunct technology. Third, like with Boeing and Airbus, ASML wants
the system integrator role because it puts them at the capstone of the entire massive
enterprise. The gateway that listens to the customer requirements and has
final say on which part goes where. This role allows for the most economic value.
Indeed, ASML is often its suppliers’ single most important customer. It might represent 50%
or more of their revenues (though this is not the preferred situation). Thus, ASML can exercise
strong influence on their operations and planning. Many of these suppliers are too
specialized to diversify away from ASML. How many customers are out there for
high-precision motors or a beam measuring unit? Okay, so now that we have a good
idea of what the supplier network is, let us look at how it works. At least, this
is how it worked as of the past few years. Might have changed a bit since then. This
is the latest information that I have. It starts with the client. Intel, Samsung,
or TSMC plan out their future customer and product demand. They need to do this a year or
more ahead of time, which can be challenging. They inform ASML that they would like to
purchase a machine to meet this demand and when they would like
this machine to be delivered. When making that order, the client chooses
from over 30 client-specific options, crafting the end product to meet their own
needs. This means that no single machine that ASML puts out is exactly the same
as the other. It also explains why TSMC, Intel and Samsung can get different
results despite sharing the same supplier. Once ASML knows that it has an order,
their Supply Chain Planning Department makes a Master Production Schedule. This MPS
clarifies when production for this specific, individual machine starts and ends.
It takes into account resource needs, available workforce hours, component lead times,
and of course, customer deadlines. After that, the suppliers are notified and issued
purchase orders to get working. As I explained earlier, each
machine is made up of modules. These modules are built independently in a process
step referred to within the company as "ASSY". The modules then are integrated together in a step
called "FASY" (standing for "Final Assembly"). After FASY, the machine is tested to see
whether or not it meets internal benchmarks. Configurations may be made to help improve
performance. This stage is called "Test". Once the machine is determined to have met
ASML standards for performance and reliability, the machine is disassembled and
packed for delivery to the customer. Over 80% of ASML's customers are located in
Asia, so this often will be a long flight. Once it gets to its destination,
ASML’s technicians work with the foundry’s technicians to install the
device into a fab clean room for use. It is a harrowing, delicate process. If you
want to know more about how foundries build and set up such rooms, I recommend watching
my video about TSMC’s fab construction work. ASML not only has to build and deliver the
product, but they also are responsible for its upkeep and service time. It is not like a
refrigerator where they set it up and go home. ASML signs service contracts with its foundry
customers after the sale with uptime KPIs. Such arrangements are normal for large
capital purchases between companies. These service contracts can hold ASML financially
liable for lost sales due to machine downtime. So the company offers 24/7 support coverage
with trained technicians and staff. These come out of their local technology
development and training sites in Japan, Korea, Taiwan, United States, and 16
additional countries around the world. When things do break, it is critical that spare
parts go out to customers ASAP. The company keeps spare parts at local service warehouses. If a
customer needs something that the local warehouse does not have, then the central warehouse has
to ship it over, with a target date of 14 days. Better than having fast turnaround on replacing
broken parts is to have them not break in the first place. To do this, ASML maintains
rigid quality control standards on their suppliers - a policy of "zero defects". Each
part is closely inspected on the factory floor and if anything can go wrong, then it is rejected. ASML's supply chain in action is a
symphony of chaos and nervous tension. Their planners have to account for parts being
rejected on the factory floor due to defects, possible disruptions, and transport issues. If a rejected or missing part is not all that
important, then workers can find a workaround - usually by replacing the item with a "dummy
part" - and continue on schedule. But if the part is mission-critical, then production can stop
entirely and the whole module can be put at risk. It is totally understandable from the
supplier's perspective that things can go wrong. It happens. But ASML has to deliver
a machine to its customers by the deadline. And a foundry like TSMC does
not want to have to go to Apple and tell them their iPhone has to be
delayed due to insufficient capacity. Furthermore, some of these parts - usually the
lens and lasers - need a lot of lead time. Longer than the actual lead time of the final product.
A single lens can take up to 40 weeks to be made. This means that some of the work needs to be
started far in advance of actual customer orders. So it is not like you can ring that up
overnight even with all the money in the world. Ah yes. Money. I have not mentioned
cost yet, but that matters too. Part of the reason why it can cost $18
billion to build a new leading edge fab has to do with the rising costs of
the latest lithography equipment. EUV equipment can cost $150 million each, as much
as a Boeing airplane. Individual components and replacements for those machines often cost
far in excess of a million dollars each. Oh, since we mentioned EUV ... Extreme Ultraviolet Lithography is the most
advanced lithography technology available today. Its commercialization with the TWINSCAN NXE series is expected to be more cost-effective
than other techniques for sub-10nm nodes. The technology has been around
since the 1980s. I have done a video about it earlier if you
want to learn more about that. I won't repeat myself recounting the long
journey. So let’s get into some gritty details. EUV lithography requires a re-engineering of many
traditional semiconductor fabbing principles. I have heard it described as a revolutionary, rather than evolutionary technique
and I think that is about right. ASML's entire supply chain has to be rejiggered
from the bottom up to accommodate these changes. Let me give you just one example of this in
action. A key challenge for EUV lithography is how to make sure that you can etch
enough wafers in a given amount of time while avoiding substantial yield defects.
But the EUV mirrors have relatively low reflective-ness, less than 70%. So
they need a very powerful light source. They achieve this by firing a
CO2 laser at droplets of tin in order to create clouds of
plasma and ultraviolet light. Why droplets? Because they need to reduce
the amount of tin debris from the plasma so to prevent the contamination of the
mirrors collecting and focusing the light. At first they experimented with a rotating
cylinder, then they moved to a thin target tape, and then a spray jet, and then a liquid filament,
before finally settling on tiny droplets. The margin of error when it comes to the
power source is really tight and this has had consequences across the entire supply chain.
It is common in older lithography methods to use something called a pellicle, a polymer film
just 1 micrometer wide, to prevent particles from casting a "shadow" discrepancy that gets
projected onto the etched wafers and ruins them. In the EUV world, however, even the super thin
traditionally-used pellicles can absorb and weaken the EUV light. So while ASML works to develop
a suitable pellicle, earlier generations of EUV machines all across the supply chain now have
to adhere to ultra-strict cleanliness standards. A particle just 52 nm wide, the size of a small
virus, can contaminate the EUV supply component. Maintaining such cleanliness pushes the
limits of what is technically possible. New proprietary detectors were
installed at all supplier premises. Particle flushing steps had to be added
throughout the whole build process. Such challenges are why EUV commercialization was delayed for several decades even after
proving the feasibility of the science in the 80s. Note. It has recently been announced that
pellicles are now available for EUV machines. So it is likely from now on that this cleaning
won’t need to be a part of the workflow. But before that, such cleanliness
standards were enforced for a few years. Having them now will improve
wafer yields and cost. ASML's market leadership in this
sector is defined in two ways. The first is in the technological superiority of
its products. ASML products are twice as expensive as competing products from Nikon or Canon, but
perform much better than the competition. They can print in more detail and at a higher rate.
That’s what the customers want more than price. The second has to do with its very
strong record of collaboration. The company works closely with both
its customers and its suppliers, conducting between all of them to deliver
a product as demanding as any airplane. This track record and constant
process improvement helps to keep ASML at the leading edge of the market,
and Moore's Law humming along.
