Home » Technology » The Revolution of Chip Manufacturing: Europe’s Role in Fabless Production and EUVL Technology

The Revolution of Chip Manufacturing: Europe’s Role in Fabless Production and EUVL Technology

When you think of chip manufacturing, everyone thinks of a Taiwanese factory TSMC. In practice, it is more complicated – the integrated circuit is gradually formed into a finished commercial product in many other factories – but for the sake of simplicity of the example, let’s stick with this scenario.

So let’s have an engineer from, say, Nvidia, create a design for a new chip in the design software, zip it up, send it by email to Asia, and in a few months a palette of graphics accelerators will arrive H100which power the most demanding AI applications today, including ChatGPT.


The Nvidia H100 accelerator powers today’s most demanding AI tasks

We call it fabless manufacturing and it is one of the pillars of the digital revolution in the last few decades. The vast majority of sensors, processors of all kinds and other integrated circuits are often developed by relatively small companies that, thanks to the outsourcing of production, did not have to invest billions of dollars in their own factories.

To make a new chip we need three things

So here we have:

  • top brainswho invent the chips
  • and brute forcewhich will give the chips their final shape
  • In order for integrated circuits to become ever better, smaller and more energy efficient, brute force must constantly miniaturize production. It therefore needs better and better machines that burn conductive paths a few nanometers narrower into the raw semiconductor material, which then together form millions and millions of interconnected chip transistors.


    Inside TSMC: It’s made here, but who supplied the machines?

    And that finally brings us to the third pillar, thanks to which we have any processors at all today: To the factories that assemble the chip production machines themselveswhich then serves, for example, in Taiwan at the TSMC plants.

    You can’t make any next-gen chip without the Netherlands

    While TSMC is one of the leaders in brute force, the European’s heart will be warmed by the discovery that the hegemon in the field of machines for the smallest production processes is the Dutch holding ASML.

    Until relatively recently, only the professional community knew about it, but the spark between the West and China in recent years nevertheless put ASML on the front pages of (not only) the technological media.

    Most recently, for example, due to suspicion of industrial espionage and prohibition of the export of machinery to China. And it’s no wonder, because ASML is the first manufacturer in the world that managed to commercialize the technology years ago EUVL (Extreme UltrViolet Litography).

    What is photolithography?

    What exactly is EUVL? First, let’s explain in all simplicity how the chip is actually manufactured. The most widespread technique is photolithographyusing which on a thin semiconductor wafer (wafer) we apply various chemical layers and finish everything with a so-called photoresist. As its name suggests, it is already photosensitive, so incident light changes its chemical parameters.


    Sandwich composite ready for photolithography

    If we wanted to burn, for example, the word ŽIVĚ into such a wafer, we take a quarter of paper, cut out the characters Ž, I, V, Ě in it with scissors and place it on the wafer. When we then shine a lamp on the sandwich, the light passes through the holes of the letters, the photons hit the exposed photoresist and change its properties in such a way that we can somehow remove these irradiated areas later.


    Light passes through the holes in the mask and hits the photoresist, which changes it chemically (purple)

    If the irradiated photoresist was soluble, for example, in water (so-called wet etching – wet etching), we immerse the wafer in a pond for a few hours, and the next morning we find the chemically engraved inscription ŽIVĚ in it. If, on the other hand, it reacted to plasma (so-called dry etching – dry etching), we choose a more sophisticated method.

    We carve chips like stone sculptures, but with light

    The production of chips is understandably somewhat more delicate, ponds are therefore not used, and current technologies have long been dominated by dry etching. Today, ultraviolet radiation is used as a light source, and the mask is somewhat more sophisticated.


    We removed the irradiated photoresist

    Depending on the type of photoresist, a negative or positive of the original appears on the wafer after irradiation and cleaning, which reveals the substrate layers under the upper photosensitive component.


    Etch the exposed lower layers

    By plasma etching, we then disrupt these exposed lower layers and finally mine microscopic valleys in them, which separate the individual conductive paths of the microscopic circuits.


    Tramtadadada, we have a wafer with chemically carved circuits – die

    In the last stage, we get rid of the rest of the upper photoresist and we have in front of us a wafer with baked and etched chip circuits (we call it die), which are then cut, encased in protective material, installed on boards with contacts and other auxiliary electrical circuits, and we have the final product.

    He does this and he does this

    These next stages are often carried out somewhere else, so the production of any processor is usually a highly global – geographically strongly decentralized – process.


    An etched circuit on a wafer is the beginning, this is not what the processor in your laptop looks like

    During the covid pandemic, problems in Malaysia, which specializes in testing and the aforementioned encapsulation, disrupted the entire process, and many car manufacturers, including Škoda Auto, had to suspend production for a while.

    The 2nm manufacturing process is (mainly) marketing

    There are many photolithographic techniques and it is an extremely complex matter, but one of the basic principles is that the size of the individual conductive paths is determined by the wavelength of the irradiation beam.


    Semi-transparent mask for photolithography. However, it is far from sufficient for the current smallest production processes of integrated circuits

    Analogously speaking, if you want to mill a tiny pug with dimensions 10×10 mm in the highest possible detail from a piece of aluminum, you will need a much smaller drill than if the poor dog has dimensions 100×100 mm.

    However, the width and pitch of conductive paths on chips manufactured by the smallest current method are calculated at tens of nanometers! Wait, tens? After all, everyone is talking about 10nm, 5nm or even 2nm production technology. So shouldn’t the dimensions be calculated in units of nanometers in that case?

    In an ideal world, probably yes, but since the mid-1990s, the designation of the production process is mainly a marketing simplification, with which Intel, Samsung, TSMC and others express a technological leap forward. And usually everyone in their own way and according to their own internal rules. For the 5nm process, the smallest pitch is actually around 30 nm, for 2nm around 20 nm and for 1nm somewhere around 15 nm.

    Even in the 1980s, it was unattainable science fiction

    In any case, it is true that at this already hard-to-believable scale, existing methods of photolithography cease to work. We need light with an extremely short wavelength, but at the same time some photoresist reacts to it.

    The shorter the wavelength, the smaller structures we can irradiate. But after a certain level, we can no longer use semi-transparent masks, because they absorb even the light they are supposed to let through

    And so, decades ago, scientists began to focus on the already mentioned EUV on the border between ultraviolet and X-ray radiation. For EUV lithography, a wavelength of approx 13,5 nmhowever, even in the 80s of the last century, the technology itself was pure science fiction.

    The development of EUVL was paid for by the American taxpayer

    But then a few scientists from Bell Labs put their heads together and published in 1991 articlethat it might be possible. Well, it will work if someone pays for it.

    Similar primary research is usually paid for from the state budget, and the USA is no exception. They have had a subsidy system for this for decades CRADA (Cooperative Research & Development Agreements) for financing joint projects between the private sector and federal institutions.

    Renowned national laboratories were therefore involved in the development Lawrence Livermore a Sandia and in the following decade together they refined the technology to a usable state.

    Who do we license? Well, not really Japan!

    In the last step, someone finally had to turn it into a practical machine, and now the ASML photolithography expert finally appears on the scene, who thanks to several acquisitions in the USA and simple geopolitics, got the stamp for production from Congress.

    Why from Congress? As we explained above, the development was largely paid for by the American taxpayer, so it was up to Washington to give the green light. And because at first he did not want to hand over sensitive know-how to the then very predatory competition from Asia, in the end it was Europe and the Dutch holding company from Veldhoven that made the most of it.


    The most advanced chips are produced on these machines today

    Its engineers, in collaboration with a bunch of other labs from both sides of the Atlantic, gradually developed the most popular EUVL machine for the mass 5nm and 3nm manufacturing process Twinscan NXE:3600D.

    The one-hundred-and-eighty-ton crumb fetched a good $200 million in 2022, and ASML sold 140 on the same date. His successor is Twinscan EXE:5000which can also handle the 2nm production process.


    The most advanced chips of tomorrow will be produced on these machines. If you have a rich uncle and enough workshop space, call Veldhoven

    All in all, if we move to the present, it is not surprising that the US has enough leverage to forcefully explain to the Netherlands which systems the Dutch factory should not send to China. Holding made them, but the basic know-how is American.

    Instead of high vacuum mirror lenses

    And it’s really groundbreaking know-how, because EUV is not just another UV with an even shorter wavelength. Engineers had to contend with a host of challenges, including how to make the mask itself. This can no longer be made of glass as in the case of older and simpler photolithography, because such a material dampens the EUV and the radiation would not shine through the mask.


    Instead of a semitransparent mask, in the case of EUVL, an array of mirrors and a reflective mask are used

    To make matters worse, extreme ultraviolet radiation is also absorbed by the normal atmosphere, so the photolithography process inside the Twinscan NXE:3600D and other machines takes place not only under absolute cleanliness, but mainly in the most perfect vacuum.

    Instead of semi-transparent glass masks, reflective masks with a system of mirrors are used for the same reason, which was designed by the former American research consortium Sematech in collaboration with Carl Zeiss. The EUV beams are simply reflected from the photolithographic template of the chip to be manufactured, and depending on whether it was a depression, the beam will acquire the right parameters to burn the photoresist on the wafer.


    The EUV beam bounces off the sample and hits the photoresist wafer. Therefore, the pattern must be incredibly perfect and the chamber must have the best possible vacuum

    But for this to work, the plastic template must be incredibly clean. Any unevenness at the nanometer level can cause a bad projection on the wafer and, as a result, invalidate all subsequent processes, which we have simply shown in the diagrams above.

    Pallets full of wafers, on which they would be photolithographed, would then leave the TSMC factories milled graphics accelerators H100, which would, in the best case, only short-circuit everything else.

    Carefully guarded European gold

    So it really is a miracle of science and engineering that it works at all, and Sematech briefly covered it in its thirty-three-page presentation (PDF). Although it is of an older date (the consortium effectively disintegrated in the meantime), it is still valid.


    Headquarters of ASML Holding NV in Veldhoven, Netherlands

    ASML is European gold with American know-how, which protects its production technology better than most banks with their vaults, and at the same time provides the old continent with a certain stamp of inviolability. That is, at least until the moment when Asia also gains ground with the same technology and on a large scale.

    2024-03-12 18:55:51
    #wont #single #highend #chip #European #ASML #gold #Dutch #factory

    Leave a Comment

    This site uses Akismet to reduce spam. Learn how your comment data is processed.