news

source:orange is not confused (wechat public account)

finally i have a chance to talk about semiconductors. this is quite valuable. it was published before, but i took a quick look at it and didn’t pay much attention to it. but i don’t know why it suddenly became popular ten days later.

core:

1. this is a dry lithography machine, which cannot produce 28nm process; the 8nm lithography machine is pure nonsense;

2. the 28nm lithography machine is actually an arfi immersion lithography machine. although it should not be called "28nm lithography machine", it is easy to understand because it is a lithography machine that can be used on the 28 production line.

the difference between this and the dry type is whether there is water as a medium under the exposure system module; when light passes through glass and water as a medium, the refraction angle is much smaller than that of air, so the wavelength of 193nm is equivalent to 134nm.

asml's 193nm immersion lithography machine can achieve a single exposure of about 22nm. the actual gate length is not 22nm, but this is not very important. therefore, 28nm does not require multiple exposures.

according to the public information, this is a dry lithography machine that can only perform 65nm process.

3. multiple exposure: used in immersion process technology, mainly used in 14nm-7nm "advanced process" technology. not to mention 28nm, 14nm is a bit reluctant to use multiple exposure now, expensive and complicated. if it is improved to 12 or 10, the performance will be improved and the price will remain the same. tsmc and samsung both vigorously developed multiple exposure at the 14nm node to compete with apple.

later, it was used at 7nm, but because it was too expensive and too inefficient, it only survived for one year at tsmc before switching to euv.

the multiple exposure technique was something intel invested a lot of money in back then, and later on, they became addicted to it…

there are three types of multiple exposure: lele, lfle and sadp/saqp.

lele, lith-etch-lith-etch, photolithography-etching-photolithography-etching. the original photolithography pattern is divided into two or more masks, and the image is superimposed. this is easy to imagine.

lfle, ltiho-freeze-ltiho-etch, photolithography-curing-photolithography-etching. it is essentially the same as lele, but it saves an etching process and reduces costs.

sadp is self-aligned double patterning and saqp is quadruple exposure technology (it seems that huawei has obtained a patent for quadruple exposure some time ago). the essential principles are similar.

the key points are:4 exposure was also done by intel after 10nm, and it was having a lot of fun. as a result, euv suddenly came, and intel cried out that it was cheated.i was practicing my archery skills, and you built a submachine gun for tsmc.

my point is clear:there is no economic value in developing advanced processes. even for china today, the process research and development and industrial value are limited.we are so far behind that it is impossible for the latecomers to follow moore's law, which is the pace of leading r&d.

it is indeed of great strategic significance, solving the problem of survival and breaking the blockade. however, the economic benefits are limited, but the pace of the industry will develop by leaps and bounds.

we can’t do it in 65, 40, 28, 14, 7 iterations. we must quickly complete arfi->euv, and achieve duv 14 to build a small-scale production capacity (optimistically, it may take 3-5 years), mainly through accumulation in the field of equipment manufacturing engineering. shangwei can cooperate with customers with good craftsmanship to expose more 7nm, and then we need to do euv. this process will take at least 10 years.

huawei can do it, because huawei is too advanced. it has reached the pinnacle of technology that is equivalent to 5nm at any cost. in fact, the technology does not have much significance for mass production... the yield rate must be low, the cost must be extremely high, and it is not a technological breakthrough. it is a kind of carving technology, but it is really amazing to achieve the goal. the production capacity will also be limited, but there is no way we don’t have euv.

4. overlay accuracy: 8nm, which has nothing to do with the so-called 8nm lithography machine. i will talk about overlay accuracy separately when i have the chance, which is quite interesting.

5. about progress:

i understand that these are all public information, and foreigners have already reported on them in various ways.

in fact, some senior industry friends have said that there will be significant progress this year, but at present...

when the lithography machine reaches the mass production stage, stability is paramount. we must ensure that this complex machine can work stably and efficiently for several years.
6. investment situation of lithography machines:

it’s a bit far-fetched… the market expects 28nm or even 14nm lithography machines to be mass-produced, corresponding to the level of 1970i and 1980i. so it’s been almost two weeks since this little piece of news was first trending on the internet.

this kind of thing needs continuous catalysis, or at least it must be impossible to disprove. but now it is too easy to disprove. so there is no market, and there is definitely no sustained market.

we really still have to wait for arfi’s official announcement, and a wave of confirmed super mainline opportunities will come.

(end of article）