Super-rigid UV in semiconductor lithography: difficult, long, expensive — and necessary

Super-rigid UV in semiconductor lithography: difficult, long, expensive — and necessary

In May 2003, Intel, to the great surprise of many experts in the semiconductor industry, announced that it was no longer going to wait for lithographic machines with a working laser wave of 157 nm to appear in their production. Instead of reaching the production rate of 65 nm in 2005, it was with the help of 157 nm aggregates that the global chipmaker then decided that he was stimulating his partners — suppliers of photolithography equipment — to continue to develop well-established DUVs with 193 nrs.

In the same year of 2003, the DUV machines had already been successful in manufacturing the 90-year-old SBIS process, and the engineers of the ASML, Canon and Nikon companies, where these units were being developed, had been asked to modify them to 65 and then 45-nm production standards; and as for the 32-nm process that Intel planned for 2009, various options were considered for its implementation, among them the most promising.

As we now know, with the task of refining the DUV-litography up to the marketing 7 nm, at least one of the listed chip-maker companies, the Dutch ASML, did well. At the same time, while EUV technology has developed in laboratories since the early 1990s, it was only in the late 2010s that it was possible to implement commercial machines for the serial production of SBIS, but it was only in the late 2010s that it was possible to overcome all the problems of 157-nm photolithography, which Intel, as the owner of lithographic installations, had realized as early as 2003, that engineers had not yet succeeded — at least outside the laboratory walls, in an annex to the serial semi-conductor production.

Why, then, would it be easier for chipmaker manufacturers to go over 180 nm than 36 nm on the way to the next technological milestone, even though it took almost two decades to do so?

♪ The purpose of the question ♪

The first series of ASML TWINCKAN NXE:3400B, which is still used for the 7 nm and 5 nm process, and has the capacity to build a silicon plate on the surface with a distance between adjacent closures of 13 nm, is 180 tons, consists of more than 100,000 knots and components and consumes more than 1 MW of electrical power. The developer announced its readiness to produce serially in 2017.

It cost customers about $120 million at the start of sales, while the more modern TWINSIAN NXE:3600D model, which is used for the production of 5-Nm and 3-Nm SBIS, is almost double the cost these days. DUV-aggregates are now available at $40-60 million per unit — one of which is a significant indication of the qualitative gap in the magnitude of the problems that EUV developers have had to solve.

In the early 2000s, in theory, the same optical systems as those used for 193 nm lasers would be appropriate for the engineers to solve only a number of purely technical problems associated with a stronger 157 nm absorption, and the new lithographic machines would quickly replace 193 nm of the predecessors on the front edge of the microprocessor front.

Still, moving quickly from 193 nm of argon fluoride, although it has problems with double beaming, and growing a suitable lens takes up to three months.

One of the main nightmares of 157-Nm photolithography, which the developers have not been able to get rid of until now, is photons of this kind of radiation. As we will see later, even higher-energy EUV phototons are also causing a lot of blood to create lithographic installations for engineers, but in the case of a refractive optical system, the situation is simply hopeless.

The problem is that a 157-nm laser beam, falling on a photoresisted plate, inevitably blows the molecules of an organic photoresist to a significant height. As a result, they are deposited on a nearby focus lens -- and alter its optical properties, thereby disrupting the speed of the beam and causing the observed production to become unserviceable. And it is not possible to increase the distance between the lens and the production: the numerical aperture of the optical system will be reduced -- it will not be possible to reach the necessary resolution.

Another delicate point is that we all know that lithographic machines work in clean rooms, where the air is often cleaned to a level of 0.5 μm 10 units or less, 300 nm or less, 200 nm or 75 or less, 100 nm or 350 or less. The finer the production standard, the stronger the tolerance, but inside the lithographic machines the atmosphere is kept even more free from contamination.

Now, for 157 nm lasers, it's not only possible to have any significant concentrations of so-called oxygen, carbon dioxide, water vapour in the air, but all because 157 nm is absorbed by these substances three decimal orders of magnitude more than 193 nm, so the presence of these molecules, which are basically perfectly normal to atmospheric air gases inside the lithographic unit, must be calculated from the total number of molecules filling the internal area of the 157 nm optical tract.

With EUV, the situation is fundamentally different, because ultra-shortwave UV radiation is absorbed almost without the rest by the thinst layer of glass. This makes it pointless to apply a refractive scheme with curved mirrors, which certainly complicates the task of developers, but at the same time eliminates the need to try to pour what is called young wine into old furs.

As the history of the last two decades of microprocessor technology has shown, from an economic point of view, it has become a reasonable choice to invest heavily in the development of a forward-looking 13.5 nm EUV-litography with the continuous progressive improvement of the DUV process on 193 nm lasers. This is how, in fact, at the same time, in 2018-2019, the TSMC chips produced by the "7 nm" technology standard on the EUV-litographs entered the market.

♪ Technology in reflection

To readers of the previous material on the difficulties and achievements of the semiconductor industry, it is certain that the linking of the minimum physically attainable — only optical means, without further masterminds such as LELE or SAMP — is the extent of the detail of the final image:


The sizeless parameters and the real systems have been shown to be generally close to one, and only by the incredible efforts of engineers and optics they are able to deviate from this value in order to reach the ratio of E ×/2. It seems clear that reducing the wave of the radiation photoresist used to display is the simplest and most appropriate way to obtain increasingly miniatural parts on the surface of the semiconductor.

So initially, when work on 193 nm photolithography was just beginning, it was logical to view 157 nm sources of coherent radiation as the next stage in the evolutionary miniaturization of the production process, but as the detail of the BIS was approaching the limit of ×/2 for 193 nm lasers, it became increasingly clear that the prowling problems with the deposition of photoresist molecules on the right side of the lens and the need for deep cleaning of the gas environment within the optical tract made it virtually impossible to change the sources of radiation to a slightly shorter wave.

Even more so, as early as 2003, the leading developers of semiconductor technologies were convinced that it would be easy to bring to mass commercial use a much more promising EUV technology within 5 to 6 years; indeed, the pattern of a reflective optical tract for a photolithographer using not even UV-violet, but essentially a soft X-ray, is not at first glance inherently complicated.

Yes, there are about 10 mirrors, but there's actually a parallel beam coming out of the EUV, and in multiple re-reflections on complex mirrors, it's almost non-dispersive, so there's no loss of light because there's no loss of light, there's almost no loss of dissipation either, and you can start showing it through the mask, but if it were that easy!

The engineers, moving along a "dear" revolutionary path to reduce the wave length of radiation used for photolithography, began to discover that the only transition from lens to mirror presents them with a radically more complex task than before. For example, what does it mean in practice to make a mirror that reflects a super-rigid UV? The problem is that the radiation with a wave length of 13.5 nm is practically any substance, even if it is polished to the required smoothness in one eighth of the wave length.

The main problem for engineers implementing the above multi-spectral pattern is that absolute reflectors — especially in the soft X-ray range of wave lengths, to which the 13.5 nm of interest to us is actually attached — do not exist, and this is due to the very high energy of photons of super-rigid ultraviolet radiation.

Classic optics deal with electromagnetic waves from the visible human eye range, and here any contact of the luminous flux with the boundary of the two environments whose refrigeration rates are different creates two beams: reflected and absorbed, by classical definition, let's recall, is the ratio of the speed of light diffusion in a given environment to the speed of light in a vacuum.

True, a certain effort can be made to ensure that the amount of energy in the light flux reflected by the substance is negligible compared to that reflected; for example, in recent years there has been intensive work to create an "ideal mirror" based on photon crystals, but the walls of labs will not be left for a long time.

The phenomenon known to any aquariumist of full internal reflection for the EUV photolithography is also unlikely to be activated: it only works when light from a denser medium tries to move into a less dense and not at any angle. In principle, filling the optical tract with a certain transparent liquid with a high refrigeration rate is possible, but -- let's recall, it's about an EUV radiation that's absorbed so that it doesn't exactly reach the photomask and much less the silicon plate of a photoresist.

By the way, the optical tracts of today's EUV machine are vacuumed for that reason: even an inert gas like neon is an insurmountable barrier to 13.5 nm radiation.

♪ Like an X-ray doctor ♪

Good reflecting properties of many materials — smooth polished metal, glass, even flat liquid surface — in the case of long-wave radiation, which can reflect an average of 88% of the light falling on them: this is more than a technically acceptable indicator.

The specifics of the hard-core UV, much less X-ray radiation, overcoming the resistance of the denser environment, are either to pass through the sample or to absorb it by transferring its energy to its substance.

The word "simple" in this descriptive explanation has a very strict physical meaning, either as a reflection of dispersion, or as a function of refrigeration from wave length. For the visible range of electromagnetic radiation, the dispersion is illustrated by Newton's canonic experience of decomposing the light flux from the sun through a prism into a fan of coloured rays: purple is weaker.

It would seem that the variance has to do with the EUV photolithography if there is a special generation of 13.5 nm radiation? But in fact, it becomes necessary to take into account the interaction of the electromagnetic wave with the substance when moving from visible light through UV to X-ray, and the classic Newton dispersion just suggests that it's not going to be easy with ultra-shortwave radiation.

The substance consists of structural elements — atoms, molecules — in a size that is quite comparable to the wave length of the electromagnetic EUV stream. And these elements themselves are the pairs of interconnected differing electrical charges, and these dipolies themselves produce electric fields that, in turn, affect the environmental flow of the EUV radiation. Yes, this is the way: the dipole is electrically neutral — the positive and negative charges in it are mutually compensated, but it creates a field around it.

By spreading in any environment other than an ideal vacuum, the electromagnetic wave leads to forced fluctuations in the resulting dipole — and transmits part of its energy to that environment. The fluctuations of the reference wave and the observed secondary waves are building up in a complex way, in which it already has to distinguish the phase and group speeds of the waves. In other words, the spectrum is bound to blur.

By studying the refrigeration indicators for visible light in transparent environments, dispersion in general can be neglected, even though it often becomes anomalous there. But for severe ultra-shortwave radiation, this phenomenon has to be systematically considered in a clear way. In this case, the refrigeration indicator is defined as a non-dimensive physical value that characterizes the difference between the luminous waves in the two environments and becomes an integrated number:

== sync, corrected by elderman == @elder_man

The imagining part with the blackout factor β is responsible for a given wave length, and the actual part .

That's why X-rays and the super-rigid ultraviolet adjacent to it pass through material barriers, so medical, let's say, X-ray machines don't actually need the means to focus: the image is already clear, with minimal, subtle diffusion on the edges of the objects.

*Bragg meaboutit

But in order to carry out a semiconductive photolithography, it is still necessary to ensure that the EUV radiation is reflected from mirrors with at least the minimum reasonable efficiency. The reflection factor -- in other words, the share of energy reflected from the section of the two media with the refractation indicators 1 and 2, when the beam falls normally -- determines the known formula of Frenelle:

= 2

Usually, by illustrating this formula, they look at visible light for which 1 = 1.0 .

So, of course, that's why the mirrors of mankind first were made whole metal, and then they found a way to put a thin layer of well-reflected metal on the back of the glass plates, first silver, and closer to our day, aluminium, but that's for visible light, and what about the soft X-rays/super-hot UVs?

Here, the situation could be described as catastrophic. For a molybdenum, which is the best representative of 13.5 nm wave radiation available to engineers, the material part of the complex refraction indicator is 0.076 and, in the sense of β, is barely 0.0064. For an ideal vacuum, of course, β = β = 0, and therefore the reflection factor calculated for the backfall of the EUV-ray on the molybdenum mirror using the Frenelle formula, using complex refraction indicators, is a pitiful 0.14 per cent.

In terms of the organization of the production process, if the optical tract of each of the 10 mirrors consistently absorbs 99.86% of the falling radiation, the photoresist would at best occasionally be filled with individual photons, and if the electromagnetic radiation was a particle stream, it would be possible to keep an extreme UV for semiconductive photolithography for a total lack of perspective.

Fortunately, the nature of light is dualistic, and it's quite simple for the EUV equipment developers to be helped by a wave phenomenon such as Da-yes, the same one whose parasitic influence in DUV photolithography is fought by fascinating masks. It's quite simple: the light waves from a common monochromatical source are essentially identical, and therefore, in addition, reinforce one another -- that is, if the minimums and maximums of each one are the same.

The source of light in the EUV machine itself does not produce synphasic radiation — just below, we will mention how it works and what difficulties engineers have to face in order to get this very 13.5-nm stream; however, this is a good luck! — there is a chance to take the known wave phases away if a mirror is set up in a special way. "He who prevents us is going to help us" is actually the motto of the engineering approach and of the EUV-litography, too.

It is well known that if the light wave is reflected from the boundary with a substance whose refractive rate is greater than that of the primary distribution medium, the reflected beam phase will change to pi/2, and vice versa, if the original beam was in an environment with a higher refractation rate and reflected from the boundary beyond which the substance with a smaller phase of the light wave will remain unchanged.

On this simple principle, you can build a fairly complex structure -- a fine-filtered Bragg mirror named so similar to the Bragg law of 1913 -- in the first detailed studies of crystal structure using X-rays. Such a mirror will consist of alternate layers of media with relatively high and low refractation rates, such as molybdenum.

To be precise, it is necessary to take into account the complexity and complexity of the refrigeration values of both environments and the non-perpendicularity of beams falling from the beam composition to the reflecting surfaces, and even the curvature of mirrors. In other words, in the manufacture of such reflectors, it is necessary to maintain a strictly defined pattern on a fairly macroscopic scale. It is not surprising that each single mirror from the optical tract of the EUV-litograph at a price comparable to the new sports car.

What happens in such a multilayered structure with a light beam? It falls on the first layer of molybdenum and barely two, the same 0.14% of its total intensity, reflects on it, changing the phase to pi/2. The beam is then reflected on the edge of the molybdenum, but because it has crossed the column of the first layer of molybdenum \4 -- and will come out into the surrounding vacuum as the first reflected portion. et cetera.

As a result, if you put one on the other 40-50 layers of molybdenum and silicon in a proper manner, the interferential interfermentation between the same phase of weak wave intensity can achieve 70 per cent of the reflectance of the Bragga mirror as a whole at 13.5 nm wave length. Even better results are shown by the Molibden-Berry Braggs mirrors, but due to the high toxicity of the beryllium, they are not applied in practice. The 70 per cent value in any case is incontrovertible, given that it started with 0.14 per cent for a polished molybdenum bolt.

It would seem that engineering happiness is the only thing left to put under this powerful luminous flux a mask and a photoresisted set!

♪ The haircut just started ♪

It wasn't here: the optic scheme of the EUV-mounting includes at least a dozen spherical and aspherical mirrors necessary for both the photo mask and target radiation to fall in strict parallel and at clearly defined angles; otherwise, it will not be possible to achieve even resolution of the extent of lithography across the entire large area exposed.

And once the Bragg mirrors are 10, and each one of them reflects the 0.7 intensity of the flow falling on it, the target will eventually reach 0.7 *0.7* *0.7 = 0.710 * 0.0282, i.e. 2.82% of the initial power of the beam. This is certainly better than 0.14% of the single molybdenum mirror, but still scarce.

The main problem, let's recall again, is that the ultra-rigid UV rays are extremely energy-intensive. So if the radiator's power is unmistakably increased, it's going to heat up more and more energy on the mirrors; the thermal deformations will inevitably distort their shape — and the focus will be broken, preventing the lytographing of the chips with the necessary level of detail. Of course, the EUV-litograph optical system is actively cooling, but it does not make it easier or cheaper.

Another problem element in the optical tract of an extreme ultraviolet machine is that actually shows the semiconductor structure of the future chip. The production of photo masks for DUV photolithography is an excellent process: usually it is a quartz plate with chrome spray in areas designed to block the luminous flux.

In order to protect the expensive mask from the deposition of dust, which inevitably distorts the pattern of lines that will lead to the deterioration of the future chip directly during the manufacture process, the masks cover — transparent polymer membranes about 1 um thick, raised above the surface of the mask by 5 to 10 mm — known to be outside the focal plane of the luminous flux, which coincides with the surface of the template.

When the dust falls on a spleen that passes through the optical system the light is slightly diffused, the "image" of the dust on the surface of the mask comes out thin and wide blurry, and the distortions thus introduced into the pattern projected through the photo mask are minimal. When sufficient dust is stored, the membrane is changed, it is cheaper and easier than to dismantle and purify the quartz mask itself.

Since the EUV radiation is absorbed, as has been said, almost any medium, or any reading template for a 13.5-nm lithography, is out of the question. The masks for such machines are the same Bragg mirrors with the structure of the future chip on them. In forming this structure, it is necessary to take into account that the extreme ultraviolet flux will fall on the photo mask. This means that the partial shadowing of the reflected light will have to be compensated by the tantalum-based areas above the surface of the first layer of the molybden.

Not only does the Braggov mirror reflect 70 per cent of the radiation falling on it, the sink is 0 per cent, which means that the transmission of a photo mask is very unevenly distributed across its outer surface, which, of course, can also lead to thermal distortions which, for understandable reasons, are not acceptable even at the level of the nanometres, so that both the correct material selection and the sophisticated heat-extinguishing system are required on this section of the optical tract.

They can make the EUV phototomas of a company in the world on a worldwide basis: Applied Materials, Asahi Glass Co, one EUV, which also requires a large number of them, estimated at 300,000.

In the early stages of the EUV series production, chipmakers hoped that no dust in the carefully vacuumed optical tract of the lithographic machine would simply occur. However, practice has shown that these expectations have not been met, and masks have had to be changed frequently. The development of a protective membrane for EUV photolithography has been taking place since the mid-2010s as ASML, IBM, Samsung Electronics and other companies, but for a long time it has not been possible to achieve a comparable price and quality result.

It was not until the beginning of 2021 that ASML declared its readiness to use a 50 nm 50 nm polysilicon wreath designed to withstand heating up to 600-1000 °C and passing more than 90 per cent of the 13.5 nm wave radiation. It should be borne in mind that the EUV flow at 200 W passes through the protective membrane when it falls on and is affected by the mirror mask. This is why the development of the ASML is unique in that it not only freely passes through the lion ' s portion of the radiation, but also sustains 400-W power, while maintaining 99.8 per cent of its original geometric characteristics, namely equal thickness and smoothness of the surface.

♪ Hot shop ♪

The high energy of extreme ultraviolet photons is the true scourge of EUV-litography. At the outset, while mastering this technology, Samsung Electronics and TSMC used non-film-free matrices, and were able to generate high-power radiation streams. Today, the need to protect the photo mask from dust has been shown to be materially proven, but the chipmaker's chipsticks are only applicable when the EUV flow is just over 200 W. Imec and Canatu are actively working on a mold of carbon nanotubes that can withstand up to 600 W, thus increasing the source's photolithography power to 300 W.

Why is high power so important, because it's just trouble, the more time it takes to form an acceptable contrast.

The higher intensity takes less time to expound; the faster the surface of each plate is processed; the higher the rate of release of the finished plates; and this is very important because the cost of a single chip is determined by the volume of the series; the cheaper the EUV machines will produce the BIS, the more accessible the end-user will be with 5 nm processors, and the less time it will take to return investment to an extremely expensive EUV infrastructure, the sooner the chipmaker will be able to start mastering the next phase of miniaturization of the process, and the more likely it will be that the "Moore Law" will be saved again.

But here's the thing: ASML TWINSCAN NXE:3600D, the most advanced EUV-litograph today, produces the finished product at 160 plates per hour, which is one and a half times slower than the almost crowning of the perfection of DUV technologies, the TWINSCAN lithograph NXT:2000i, which bakes 300 mm plates with a physical resolution of 38 nm at a rate of 275 units per hour.

What's the problem here? It's enough just to increase the power of the EUV-rayer -- and the speed of plate release is going to increase immediately! But, unfortunately, as the reader must have noticed, there's nothing in the EUV photolithography that's "simple".

Note that the energy intensity of the EUV process itself is simply incredible. To begin with, the efficiency of converting the electric energy consumed by the lithograph into a ray is ridiculously small — not more than 0.02 per cent for the first sample of such machines. In other words, in order to generate a luminous flux of 200 W, the modern EUV machine costs just over 500 kW, but the typical DUV-grigate with 193 nm lasers needs less power in order!

By the way, the 13.5-nm luminous flux node is rarely a complex design that uses impulsive carbon dioxide droplets that are melted and cleaned of the smallest contamination of tin. It is in the hot state of supersealed plasma that the tin emits a super-rigid UV with the required wave length.

Each part of the EUV radiation is formed into two receivers: the laser beam first splashes a drop of metal into a flat disc, thereby increasing the target area, and then the second pulse turns the disc in part into a plasma cloud, thereby increasing the effectiveness of plasma generation from 0.33 per cent for single exposure to 10 per cent for double exposure. It is clear that at least closer to the point of formation of the EUV-flow, the Bragg mirror has to be changed at least once a year, as both tin drops and high-energy ions damage the surface.

Of course, microprocessors want to accelerate the release of finished plates from the most advanced EUV photolithographs so far, increasing the power of the radiation source. already, EUV-litographers are only 70-80 per cent of the lifetime, while DUV-machines have long achieved almost continuous 90 per cent productivity.

However, in addition to the lack of established high-temperature protective membrane production, a number of other factors have to be taken into account, especially the duality of photons, which characterizes the nature of light. In radio electronics, this nature manifests itself as a fractional noise — stochastic fluctuations of signal intensity. The more photons in the stream, the more smoothed the effect of fractional noise on measurements, but it never disappears completely.

In the case of photons, 13.5 nm of radiation, the situation is reverse: the photoresist is extremely small. Formally, this does not prevent the light of the sensitive layer: the energy of 13.5 nm of the photon is about 14 times greater than 193 nm. However, if a square nanometer of the target in the DUV machine reaches an average of 97 photons, the same area in the EUV-aggregate is only seven. Consequently, each element on the surface of the silicon plate will be formed literally by a small number of photons per square nanometer, so the error of the number of photons entering a single site increases from about ±10 per cent to ±40 per cent.

Moreover, the powerful EUV phototons, when crashed into a layer of photoresist, beat the electrons from its constituent molecules, which are also very high-energy, which is actually the principle of the action of the light-sensitive layer: primary , modified molecules, having lost the ability to hold in place, are washed out.

However, in the case of 13.5 nm radiation, the energy of the photons is so high that the electrons produced by them are themselves powerful enough to beat the cascade out of the surrounding substance; they in turn initiate a series of chemical reactions, sometimes quite bizarre, both in and outside the area directly lit. This further disrupts the clarity of the image on the edges; the "shadows" of the surface-formed chips appear to be more blurry than on the template.

All of the effects listed are that the shapes of the individual components of the transistor on chips made by the EUV-mode are not clear, rough lines , which further increases the probability of defects on the finished product.

In this connection, there is also a problem of defects in the entire growth of developers: how can 300 mm in the diameter of a plate covered by conductive transistors with a density of 130 to 230 million units per square millimeter be identified quickly and with a good coverage to detect unserviceable areas of future BISs — and thus guarantee a certain proportion of known chips? For this purpose, ASML, KLA, NuFlare, Tasmit and other companies are developing electro-radio-fractory defects capable of mass verification of the expected characteristics of many substantially vertical structures, because when moving to marketing standards of 3 nm, 2 nm and less, FinFET transistors are being used by even more vertical transistors with ring closures, GAAFET.

Semiconductive photolithography based on 13.5 nm radiation as a field of engineering research is now in about the same position as DUV technology two decades ago. The first generations of EUV chips have successfully solved a number of technical problems, but there are even more crows of completely new ones.

The ways to improve the EUV are many: it's also the creation of new photoresists, and the increase in the number of apertures of the optical system, and the expansion of the radiation power to increase the speed of processing the unit plate, and many other things, and while the demand for computing equipment continues to be at least as high as it has been during the last half-century, to further miniaturization of the process, the state and the private business will invest more and more mind-blowing tools, because they will be repaid as the capabilities of new generations of computers grow at the same speed.