news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

some time ago, i suddenly heard a lot of people asking:

china's own 7nmlithography machine, is it really made?

the cause was a notice issued by the ministry of industry and information technology on september 9: "guidelines for the promotion and application of the first set of major technological equipment (2024 edition)".

someone noticed that there were two lines in the catalog that were wrong. very wrong.

see picture.

what do these two lines mean? did they suddenly and quietly announce china's own new lithography machine without even an adjective?

why is there a "≤8nm" in the description of the lithography machine below? oh my god, doesn't that mean it has broken through the bottleneck of "7nm"?

soon, someone said: great! the boat has crossed the mountains, it's confirmed. china finally has its own 7nm lithography machine and can make its own 7nm chips, so it doesn't have to worry about being stuck anymore.

however, some people said: don't get excited. it's just a misunderstanding. the "8nm" is not the point, but the "65nm" above it is. domestic chips are still only at the 65nm level. if they work hard, they can only reach 28nm at most, which is still far from 7nm.

two sounds, two rhythms.

i don’t know, how do you feel after listening to it?

what does it mean to "build a 7nm chip"? is it really a big deal to do this? what do those few words in the "catalog" of the ministry of industry and information technology mean? are we still stuck in the chip market?

my feeling is that maybe we can “feel” it later, because for most people, chip manufacturing is too unfamiliar.

for example, take a random news article:

"the domestically produced lithography machine officially announced this time is a duv lithography machine with an overlay of ≤8nm, a resolution of 65nm, dry type, a wavelength of 193nm."

the sentence is not long, and there is not a single uncommon word. but, if you are not a professional who knows this field, how many people can understand it? and how many words can they understand?

perhaps, in order to really understand this matter and not be easily led by the rhythm, we must at least understand 10 things first.

let’s start with the one that went viral a year ago.

a thunderclap

on august 29th a year ago, the huawei mate60 pro mobile phone was suddenly put on sale without any publicity.

then, in those few days, from all the hot search lists to my circle of friends, they were all flooded with the same word: 7nm chip.

many of the first batch of people who bought this phone, both at home and abroad, did the same thing: opening it.

take out the kirin 9000s chip in the mobile phone, run the score, test the performance, and see what level it has achieved.

the conclusion is: this might really be a 7nm chip.

a loud thunderclap.

many people are sighing: "the most difficult time has passed, and the boat has sailed through thousands of mountains."

why do you say that? how difficult is it to make a 7nm chip? is it really amazing?

coincidentally, during that period, my live broadcast room invited the author of "chip wars", mr. yu sheng. i also took this opportunity to read some materials and consult some friends.

after learning more about it, i have a growing feeling:

to produce a 7nm chip, one really needs to overcome many difficulties.

if you can get past it, it would be amazing.

this is really amazing and worth knowing.

so today, i helped you sort it out.

the information is a bit hard-core, so i will try to speak mandarin to you.

first, let’s talk about the “7nm” that made many people stand up and exclaim.

7nm

first, a basic question: what exactly does this 7nm refer to?

why are people so concerned about this number? is it impressive?

this matter has to start with you.

when you buy a mobile phone, you want it to have strong performance, long battery life, and be thin and light?

these three requirements, when passed on to the chip world, have become three "ultimate kpis":

PPA。

performance performance, power power consumption, area size.

this ppa has become a "small goal" for chip manufacturers:

put more transistors into more in a small chip.

the main feature is that we have employees who can do more work and help you complete more and bigger projects, while consuming less of your electricity and taking up less of your space.

but what if there are too many employees and we can’t accommodate them all?

the solution was drastic: ask employees to lose weight.

there is a "groove" in the structure of the transistor, which provides plenty of room for weight loss.

so, please note that when we first talk about chips and say "your chip is 28nm" or "my chip is 14nm", 28nm and 14nm do not refer to the same thing. it is not the size of the chip, the size of the transistor, or the distance between transistors, but the "channel width" in the transistor.

but then, as we talked, the topic got tangled up. 28nm, 14nm, 7nm…

when it comes to "7nm chips", whether the "channel width" is really reduced to 7nm is no longer the point. there are different opinions, but the essence has not changed:

smaller nano-process means better ppa, which means more "employees" can be crammed into a smaller "office".

how much is enough?

to make a 14nm chip, more than 30 million transistors must be packed into every square millimeter.

to make a 7nm chip, nearly 100 million transistors must be packed into every square millimeter.

twice as capable. but also, twice as hard.

and this is just the beginning of "crossing thousands of mountains".

because, just being able to cram in space is not enough. how can you arrange tens of millions of employees clearly in a space as small as a fingernail?

lithography

that’s right, it relies on a method that sounds expensive: photolithography.

how to carve relying on light?

this is a very complicated matter. a photolithography machine has more than 100,000 parts and costs hundreds of millions of dollars, not including shipping. it costs more than a boeing 737. just to get this done.

but it is also very simple to say. have you seen the movie?

when a traditional film movie is shown, a beam of light is first emitted, passed through a lens like a magnifying glass, and then passed through a layer of film, so that the pattern on the film can be projected onto the screen.

photolithography is similar. it also emits a beam of light, passes through a set of lens systems, and then passes through a layer of mask, and then the circuit diagram engraved on the mask can be projected onto the substrate for making chips, that is, the wafer.

the only difference is that in film projection, a "magnifying glass" is used to project a small image into a large one, while in photolithography, a "reducing glass" is used to project a large image into a small one.

how clever to use the projection of light as a lever.

however, at this point, we have only clearly outlined the edges and know where to start next.

but how to start?

the circuit diagram of a 7nm chip must clearly arrange tens or even hundreds of billions of transistors and other electronic components.

moreover, everything from transistors to the wires connecting them are refined to the nanometer level, which is 100,000 times thinner than the blade of your kitchen knife.

someone in the industry once described it as: this is equivalent to carving out the whole of shanghai in an area the size of a fingernail, without missing a single room or carving a single road crooked.

this is crazy. how do you carve this? how can you carve the grooves of this circuit diagram "quickly, accurately, and steadily"? with a laser?

at the beginning, it’s not that no one has tried it.

however, after trying various methods, such as laser direct writing and nanoimprinting, some are very expensive, some are very slow, and some are easily scrapped. it is difficult to commercialize it, and anyone who does this will lose money.

until someone found a very imaginative method:

take a roundabout way to save the country. use photoresist.

photoresist

what is photoresist?

photoresist is something that is very sensitive to light.

once exposed to light of a specific wavelength, a chemical reaction occurs.

it was originally very tough, but it became timid when exposed to the light and could be easily washed off by chemical solvents.

grasping this point, lithography has a new approach to solving problems:

it is not done by physical carving stroke by stroke, but by chemical corrosion layer by layer.

although there are many processes involved, the idea is roughly the same as "putting an elephant in a refrigerator", with four main steps:

the first step is to apply a layer of photoresist evenly on the raw material of the chip, that is, the wafer.

the second step is lighting. let a specific beam of light pass through the mask with the circuit diagram drawn on it.

where the lines cover, light cannot pass through, and the photoresist retains its original properties.

light passes through the areas not covered by lines, and when it comes into contact with the photoresist, the photoresist changes into a different state.

the third step is to wash the photoresist. put the wafer covered with the two types of photoresist into a specific chemical solution.

those photoresists with relatively softer properties will be dissolved, and the circuit diagram will be displayed on the photoresist layer.

step 4: etching: place the wafer into the etching solution.

the places where the photoresist has not been dissolved are equivalent to being covered with a layer of protective film, while the places where the photoresist has been dissolved will directly come into contact with the corrosive liquid and be "quickly, accurately and fiercely" etched into grooves corresponding to the circuit diagram.

light, mask, photoresist, wafer, and various chemical solutions.

the physics problem that i thought was extremely difficult suddenly became an ordinary chemistry problem and was solved.

this is the current mainstream lithography method:

first, project the circuit diagram onto the substrate like a movie;

then, just like developing a photo, etch the circuit diagram onto the chip.

it seems that photolithography is not that difficult.

it doesn't look like it.

but there is a key difficulty here: the wavelength of light.

wavelength

to carve out nanometer-level circuit diagrams, at least, the knife in your hand should be delicate enough.

how to get a finer knife?

when your knife is made of stainless steel, you just need to sharpen the blade.

but what to do when your knife is a beam of light and you can’t sharpen anything?

solve from the source of knife materials: the shorter the wavelength of light, the sharper its blade.

because the shorter the wavelength of light, the smaller the diffraction diffusion angle. in other words, the light will go in a straight line more obediently, without being blurry or moving around, and it will hit wherever you point it.

that’s easy. just open the spectrum and find the light with the shortest wavelength and use it.

spectrum diagram (image source: www.asml.com/en)

it’s not easy, because you can’t use short-wavelength light whenever you want.

do you have the ability to produce it stably and continuously under the premise of controllable costs? does your photoresist react with it? are your other process flows compatible with it?

these are all difficult problems. we have to explore them.

after exploring till now, there are mainly two “light knives” that can help people to achieve stable efficiency and control costs:

duv and euv.

duv is the name of a type of light: deep ultra-violet. the wavelength can be as short as 193nm.

many people believe that the lithography equipment using this "light knife" can basically only produce chips with a process of more than 20nm.

euv is also the name of a type of light: extreme ultra-violet. as the name suggests, this type of light is more intense, with a wavelength as short as 13.5nm.

whoever possesses this knife will have the opportunity to take another step forward. , engrave 7nm, or even more advanced chips such as 5nm and 3nm.

great. doesn’t that solve the problem of finding short light waves?

if you want to make 7nm chips, use euv.

the technical problems were solved. but other problems arose.

someone is choking me.

stuck in the neck

currently, there is only one company in the world that can produce euv lithography equipment: asml in the netherlands.

in 2018, china's semiconductor manufacturing international corporation (smic) spent 120 million euros, equivalent to its annual profit, to order china's first euv lithography equipment from asml.

a big order.

asml was also very happy and even the export license was ready.

however, the united states spoke out, claiming that 20% of euv lithography equipment contains american parts, and that exporting them requires their consent, but they disagree.

a ban.

what should we do? if we can’t use euv, which can engrave 7nm chips, can we still make 7nm chips?

can we try using duv which can only engrave chips above 20nm?

there is hope.

there are two technologies that offer hope: immersion lithography and multiple exposure.

immersion lithography

what is immersion lithography?

very simple, translated, it is: soak in water and carve.

known: the wavelength of your "light knife" should be as short as possible.

it is also known that the shortest wavelength of duv light can only be 193nm.

an idea for carving a more advanced chip emerged: can the wavelength of duv be made shorter?

yes, add water.

a layer of ultrapure water is added between the wafer surface and the lens. the water is so pure that it does not contain any impurities such as minerals, particles, bacteria, microorganisms, etc., and only contains hydrogen ions and hydroxide ions.

then, let the light refract in the water.

the refractive index of 193nm deep ultraviolet light in water is 1.44, and the wavelength can be further shortened to 134nm.

the "blade" becomes sharper in this way.

so smart.

this method brings duv lithography equipment directly from the dry era of "etching in the air" to the immersive era of "etching in water".

but that’s not enough.

by iterating the "blade" in this way, you may be able to advance to the top of your class and improve the manufacturing level from 28nm process to 22nm process, but it is still difficult to get into tsinghua university and master the 7nm process in one go.

what to do?

another method can be added: multiple exposure.

multiple exposure

what is multiple exposure?

it is also very simple. to translate it, it is: carve several times.

for example, combing your hair.

question: how can i comb all my hair so that each strand is clearly distinguishable?

comb more often.

is there any way to be more efficient and get all the hair done in one go?

it's difficult, but it's not impossible. you can go to yiwu and ask the boss to customize a new comb.

there are hundreds of thousands of hairs on a head. if you want to comb all of them at once, you need to make a comb with at least hundreds of thousands of teeth.

but what if the boss in yiwu hears you and says he can’t make it, or can’t sell it to you even if he can?

then don't pursue efficiency for now. just comb your hair a few more times to make sure it's all in place.

the same is true for multiple exposures.

if the lines on a "shanghai map" are too fine and too difficult to "carve", then carve it a few more times.

split it into three "layers" with more sparse lines, and then make three "mask plates" and "carve" them one by one. in the end, can't we also overlap them to form a complete "shanghai map"?

hair can be combed over and over again. circuit diagrams can also be carved layer by layer.

the so-called lele process, lfle process, and sapd process are essentially methods of multiple exposures and multiple engravings.

smart. if we want a 7nm chip, can't we just expose it a few more times?

theoretically, yes. but in reality, this method has its limits.

first, they use one mask and expose once. you use three masks and expose three times. who is more competitive in terms of cost and efficiency?

secondly, you need to comb your entire head at least once. move your hands and aim the comb to another position, right?

but when combing your hair again and again, how can you ensure that it is 100% accurate every time you change to a new position?

when "carving" layer by layer, how can we ensure that the last few pieces will be 100% perfectly matched when stacked together?

there is no guarantee. there will always be errors.

this error value is called "overlay".

the “≤8nm” that many people have highlighted in the “catalog” corresponds to the overlay value.

cost, efficiency, yield.

chip manufacturing is not only a technical issue, but also an economic issue. in addition to "can it be done", we must also consider "is it worth it".

using duv lithography equipment to manufacture 7nm chips through multiple exposures may help reach higher places, but there are also costs and ceilings.

so today, many sources believe that, taking all factors into consideration, even with the addition of immersion lithography and multiple exposures, making 7nm chips is almost the ceiling for duv lithography equipment.

to move forward and manufacture 7nm chips, or even more advanced 5nm and 3nm chips, we still have to rely on euv lithography equipment.

it's too difficult. either you can't buy it, or it's not worth it.

so, is it possible to take the route of "constant self-improvement" and build an euv lithography machine by ourselves?

well, you have courage.

euv lithography machine

how difficult is it to build an euv lithography machine?

a friend of mine answered me:

if the difficulty coefficient of "making a 7nm chip using a duv lithography machine" is "crossing thousands of mountains" , then the difficulty coefficient of "building an euv lithography machine" is"more than ten thousand mount everests".

lithography machine

why? how big a difference can there be between euv lithography machines and duv lithography machines, with only one letter difference?

isn't it just emitting light, casting a shadow, and carving some grooves? how difficult can it be?

that is correct. let us discuss these points one by one.

level 1: “shine a light”, how difficult can it be?

the light source of duv is just excimer laser, which is similar to the light used in laser surgery to treat myopia.

however, the light source of euv is light that does not originally exist on earth.

it doesn't exist? how to send it?

the current method is to "beat" a metal, tin, until it shines.

this is a rough but not simple process, which can be divided into three steps:

the first step is to drop a liquid tin bead from mid-air.

the tin beads should be small. it is so small that it is only 20 microns in diameter, about the same size as one of your cells.

the second step is to use a high-energy laser to continuously bombard the dripping tin beads.

act quickly. the same tin bead was bombarded at least twice, flattened the first time and vaporized the second time.

hit it so that its atoms are ionized, emitting very angry radiation and the light you want.

the third step is to continue bombarding and continue emitting light.

the hands can't stop. you have to keep bombarding it continuously for at least 50,000 times per second to ensure that it keeps collapsing, keeps ionizing, and that you always have light and can carve it steadily.

do you have the ability to do these? if you can, you can enter the next level.

level 2: “cast a shadow”, what’s so great about that?

light with a shorter wavelength has an unreliable characteristic: it is easily absorbed and is almost completely dispersed before it reaches the photoresist and starts working.

what to do? you have to rely on the "mirror".

current euv lithography machines are equipped with many "mirrors", or focusing reflectors, to ensure that euv light is less absorbed along the way and reaches the photoresist more safely.

how flat do these "mirrors" need to be?

in technical terms: the peak-to-valley value of the surface accuracy is 0.12 nanometers, and the surface roughness is 20 picometers.

translated into mandarin: if this "mirror" were enlarged to the size of the earth, only a protrusion as thin as a hair would be allowed on it.

no wonder some people say that this kind of "mirror" probably the smoothest man-made object in the universe.

now, even if you could make such a mirror, photolithography would only be the beginning.

the third level: "carve out gullies and trenches", how many mountains do you have to cross?

how can we carve out the corresponding grooves with such extreme precision?

in addition to an extremely sharp knife, you also need to have an extremely stable working environment.

take the clean room of asml as an example. the air inside needs to be 10,000 times cleaner than the air outside.

to do this, you would need at least a ventilation system that can purify 300,000 cubic meters of air per hour.

in addition to the air, the water and light used in the working environment need to be ultra-clean and require special treatment.

schematic diagram of the photolithography machine

"emit a light that does not exist on earth."

"use the smoothest mirror of mankind to cast a shadow."

"carving grooves in an environment where the air is 10,000 times cleaner."

this means that in order to build an euv lithography machine that is similar to the ones others are using now, there are at least a few mountains to climb.

oh my god. take a deep breath.

but i still can't help but want to see where we have climbed today?

future

remember that first line of introduction?

now, watch it again, how do you feel?

"the domestically produced lithography machine officially announced this time is a duv lithography machine with an overlay of ≤8nm, a resolution of 65nm, dry type, a wavelength of 193nm."

what does that mean?

"overlay ≤8nm" only refers to the error when "combing", not the level of "being able to produce 7nm chips".

"resolution 65nm" means that there is a chance to carve out a 65nm chip. if multiple exposures are done at all costs, perhaps even 28nm chips can be achieved.

“dry” means there is a “submerged” mountain to climb ahead.

"duv lithography machine with a wavelength of 193nm" means that there is still a mount everest to climb ahead, which is "euv lithography machine with a wavelength as short as 13.5nm".

why are there so many more mountains? when will we finish climbing them?

when will we be able to truly produce a 7nm process, or even more advanced domestic lithography machines that can compete with the world's level, and no longer be stuck in the bottleneck?

there are many theories. perhaps you have heard some of them. for example:

a few years ago, some people said it was impossible. “even if we gave them the blueprints, they would not be able to build a photolithography machine.”

some people have said that it is still a long way to go. “it may take more than ten years, because asml, the world’s most advanced company, took more than ten years to complete this journey.”

but soon, some people said, it’s hard to say. “asml spent more than ten years to develop this product, which involved the cooperation of dozens of countries around the world and the cooperation of thousands of domestic and foreign suppliers.”

yes, i have heard of it. but everyone has their own opinions, so how can i judge? is there any opinion from the front line?

huawei did not say much at this year's mobile phone launch conference. but on september 19, huawei's vice chairman and rotating chairman xu zhijun briefly said two sentences at another huawei conference:

1. "china's chip manufacturing technology will lag behind for a long time, and we need to prepare long-term computing power solutions."

2. "huawei's strategy is to carry out systematic innovation and improvement based on available manufacturing processes."

what about the "catalog" of the ministry of industry and information technology? to put it more simply. look at the title:

"guidelines for the promotion and application of the first set of major technological equipment (2024 edition)".

what is “significant”? a breakthrough that is critical. and critical things often continue to break through.

what is "promotion"? very advanced and mass-producible. in addition to the factories that put into mass production, there are often more advanced laboratories.

so, what about outside the lab? is there anything else?

a few days ago, i visited mexico and saw many chinese new energy vehicle factories being built there;

a few days ago, i saw the hot search for "huawei", and the one below it was the mass production and delivery of the domestically produced large aircraft "c919";

what about before that? the national bureau of statistics announced the national economic operation in the first half of 2024. among them, investment in high-tech industries increased by 10.6% year-on-year, 6.7 percentage points faster than total investment...

innovation, improvement. continue to break through, continue to research. more expansion, more investment.

the story of 7nm chips is not just about chips and computing power, but also about technological development and competitive games.

this is a major change that occurs only once in a century.

in this changing situation, there are always people shouting: the boat has passed through thousands of mountains.

indeed, from no 7nm chips to having 7nm chips, from duv to euv, from a new document to a new computing power.

both are difficult, both are possible.

but there are mountains beyond mountains.

in addition to 7nm, there are 5nm, 3nm, and even 2nm and 1nm...

in addition to chips, there are also artificial intelligence, new energy, aerospace, marine engineering...

what to do?

the canoes seldom answered. they just kept on sailing.

go forward, go forward.

bless.