news

the author is leslie wu, a former tsmc plant construction expert

edited by su yang

a document from the ministry of industry and information technology once again brought the development of domestic lithography machines into the public eye.

on september 9, the miit account "gongxin micro news" disclosed the notification document issued by the miit on september 2 regarding the issuance of the "guidelines for the promotion and application of the first major technological equipment (2024 edition)" (as shown below).

the first item of “special electronic equipment” in the notification document is “integrated circuit production equipment”, which clearly mentions the technical indicators of krypton fluoride (krf) lithography machines and argon fluoride (arf) lithography machines, especiallyargon fluoride lithography machine, the document indicates that its wavelength is 193nm, resolution ≤65nm, overlay ≤8nmthis was also understood by the outside world as a major breakthrough in domestic duv lithography machines, and there were even reports that domestic duv lithography machines had broken through the 8nm process.

so, what do the technical indicators of domestic lithography machines mentioned in this notice from the ministry of industry and information technology actually represent?

01 re-understanding the photolithography machine

due to special reasons such as export controls, lithography machines have been mentioned frequently in the past two years, and people in the technology industry have more or less some understanding of lithography machines.

to sum up in one sentence, the photolithography machine uses a special process to miniaturize the pattern, project it onto a silicon wafer and etch the transistor circuit, thereby realizing chip manufacturing.

depending on the light source, lithography machines can be divided into three types: uv, duv and euv.

each light source type is also differentiated according to the way it produces light. the wavelengths of various light sources can be referred to in the following table:

the two types of equipment mentioned in the notice of the ministry of industry and information technology correspond to the two types of duv lithography machines, krf and arf dry, which use deep ultraviolet light. the official documents only use the chinese characters krypton fluoride and argon fluoride to mark them.

different lithography machines have different light sources and corresponding wavelengths. the shorter the wavelength, the higher the resolution that can be achieved.for example, the krypton fluoride lithography machine uses a 248nm light source and supports the production of 0.11μm-0.8μm resolution chips, while the 193nm argon fluoride dry lithography machine can achieve a higher resolution of 65nm-0.11μm.

another key point is the numerical aperture (na) of the objective system. the reason why these two indicators are critical is due to a very well-known formula - the rayleigh criterion, that is, cd = k1*λ/na.

cd is the line width, that is, the minimum achievable feature size, λ is the wavelength of the light source used by the lithography machine, na represents the numerical aperture of the lithography machine objective lens, that is, the angular range of the lens to collect light, and k1 is a coefficient that depends on many factors related to the chip manufacturing process.

according to the formula, if chip manufacturing wants to achieve a smaller line width, that is, the smaller the cd value, the mainuse a shorter wavelength light source、objectives with larger numerical aperture (na), and find a waylower k1。

for example, the current euv extreme ultraviolet lithography machine has a light source wavelength of only 13.5nm. at the same time, asml is also continuously launching euv lithography machines with higher numerical apertures for the manufacture of 7nm or even higher process chips. but please note thata 3nm chip has about a hundred layers from bottom to top, and the resolution requirements are also from high to low. the euv lithography machine is only responsible for the bottom 20 layers, and the rest are coordinated by the duv lithography machine.

02 walking on two legs: high numerical aperture, immersion lithography

according to information we have learned from the industry, the lithography machine mentioned in the miit notice can achieve a k1 value of 0.25. according to the rayleigh criterion, 65=0.25 ×193/na, it can be inferred that the numerical aperture of the domestic lithography machine is 0.75.

the numerical aperture is relatively low, which is acceptable for the first generation of products. after all, there will be second and third generations in the future.

however, even initerate the numerical aperture on the existing arf light source lithography machine, from 0.75 all the way down to 0.93, the resolution is only increased from the current 65nm to the future 52nm, which is far less than the so-called "28nm lithography machine".

therefore, we iterate on the numerical aperture path.there are benefits, but they are not enough. we also need to try more breakthroughs in immersion lithography machines to achieve two-pronged approach.

the immersion arf light source has not changed in essence and is still 193nm (the light source power is the core of the mass production machine). it’s just that ultrapure water is added between the objective lens and the wafer of the lithography machine, and the refractive index is increased to 1.44, which in disguise reduces the 193nm wavelength to 134nm, thereby improving the resolution of the lithography machine.

why is this happening?

as mentioned earlier, the rayleigh criterionthat is, cd =k1*λ/na.now that we have the refraction of water, we can make a transformation.cd =k1*λ/nsinθ, where n is the refractive index of water, sinθ is the sine of the angle between the lithography machine lens and the imaging surface, and nsinθ is equal to the numerical aperture na.

the asml 2100i mentioned in table 2 is an immersion lithography machine, so n is 1.44, the sinθ value of the objective lens is 0.93, and the k1 value of this device is 0.28.

according to the deformed formula, the cd of the 2100i lithography machine = (0.28×193)/(1.44×0.93)=54.04/1.3392≈40nm, which is the resolution of what we usually call the "28nm lithography machine".

how will the domestic lithography machine perform if it is directly upgraded to immersion type without increasing the numerical aperture?

continuing to apply the formula, its cd = (0.25×193)/(1.44×0.75) = 48.25/1.08 = 44nm, which still does not meet the resolution requirement of the "28nm lithography machine".

so, back to what we said before, wenot only do we need to invest in the research and development of immersion lithography machines, but we also need to make breakthroughs in the lens, increase the sinθ value of the objective lens, and increase the numerical aperture.

the good news is,some companies are already working on immersion objective systems with a numerical aperture of 0.85.if the research is successful, the resolution of our lithography machine is expected to reach 39.41nm, truly breaking through the 40nm resolution required by the "28nm process".

in this document from the ministry of industry and information technology, the numerical aperture related to the objective lens was not disclosed, which deserves follow-up attention.

you know, the first generation of immersion lithography machines need to evolve from dry type. if the numerical aperture of the objective lens of the dry type lithography machine does not reach the first-class level, the immersion type lithography machine will be of no use.

as mentioned earlier, the principle of immersion lithography is to place ultrapure water between the bottom of the lens and the wafer. this is easy in theory but quite troublesome to implement in engineering.

first, we need to completely eliminate bubbles in ultrapure water. second, we need to eliminate the uneven surface of the liquid caused by the temperature difference between the light-transmitting area and the shielded area. the way to solve this problem is to make the ultrapure water flow quickly, but this will also produce vortices. it is a very difficult engineering problem to make ultrapure water flow quickly without producing vortices.

just for the immersion system, it took tsmc’s south taiwan science park factory dedicated to asml two years and 7-8 revisions before the breakthrough was achieved with lin benjian’s team.

after the alpha machine was completed, it entered the beta stage, and a huge amount of manpower had to be organized in the wafer factory to waste countless wafers in order to reduce the original thousands of defects to hundreds, dozens, and finally to zero. this was a difficult process.

03 multiple exposures are also "unsolvable"

if the resolution is only 65nm, is there any other way to improve it further? yes.

as mentioned above, the ruili criterioncd = k1*λ/na. in addition to the two indicators of wavelength and na numerical aperture, the continuous reduction of k1 can also achieve improved resolution.

reducing k1 is the top priority of wafer fab lithography process engineers. engineers have created many amazing technologies to reduce k1, including phase-shift masks, model optical proximity effect correction, over-etching, inversion lithography, etc.

according to lin benjian's lecture "optical micro-shrinking ics by a million times", to reduce k1, we must first "prevent vibrations", just like the anti-shake function of a mobile phone camera, and try to reduce the relative vibrations between the wafer and the mask during exposure, so that the exposure pattern is more accurate and the resolution lost due to vibration is restored. the next step is to reduce the "useless reflections" on the liquid surface during exposure.

by improving the above two items, k1 can basically be reduced to the level of 0.65.

to reduce k1 and improve resolution, dual-beam imaging methods can also be used, including off-axis exposure and phase-shifting masks.

off-axis exposure is to adjust the incident angle of the light source so that the light enters the mask obliquely. by adjusting the angle, the two lights interfere with each other to form an image, which increases the resolution and depth of field. phase-shifting masks are to make some adjustments to the mask so that the light passing through adjacent light-transmitting areas has a 180-degree phase difference.

both methods can reduce k1 by half and cannot be used in combination.

reducing k1 to 0.28 is almost the limit of what all the above technologies can achieve. if you want to reduce it further, you need to use more than two masks during exposure, which is the well-known multiple exposure (as shown below).

in the simplest terms, it divides the dense pattern into two or more masks with looser patterns, which are exposed on the wafer in turn to achieve improved resolution.

however, because the number of exposures is doubled, the wafer output efficiency is reduced by half while the wph (wafer output per hour) remains unchanged, and one more exposure will also lead to a decrease in yield.

through double exposure, k1 can be reduced from 0.28 to 0.14, or even 0.07 with quadruple exposure.

taking the 2100i lithography machine as an example, after various buffs are stacked, its theoretical cd = (0.07×193)/(1.44×0.93)=13.51/1.3392≈10nm. note that 10nm refers to the resolution, which corresponds to the 2nm process. in everyone's words, "a 28nm lithography machine produces 2nm."

since multiple exposure is so useful, can the resolution of domestic 65nm arf lithography machines be improved through multiple exposure? not yet.

multiple exposure is a technical means that needs to meet many engineering conditions, such as overlay accuracy, which can be simply understood as the error caused by exposure between different layers of the chip.

at present, the control window of single-exposure overlay accuracy is approximately 20%-25% of the resolution. therefore, a product with a 65nm resolution requires an overlay accuracy of at least 13nm. the overlay accuracy of domestic equipment is 8nm, which meets this standard.

however, it should be noted that 8nm is the factory standard, which is the result on the standard wafer. due to the errors caused by various processes in the wafer processing, the production line will be much lower than the factory standard. this is the same for asml and nikon. in other words, the 8nm standard indicator of domestic equipment is about 11-12nm in actual products.

to perform double exposure, the overlay accuracy must be reduced by half from 13nm to 6.5nm based on the resolution of 20%-25%. the current 8nm overlay accuracy indicator theoretically cannot meet the requirements.

therefore, it is necessary to improve the resolution through multiple exposures on this domestic equipment, and in future iterations, the overlay accuracy will have to be further improved.