As we all know, the current chip process is a photolithography process, that is, through the principle of optical-chemical reaction, the circuit diagram is transmitted to the photoresist-coated silicon wafer with light to form an effective circuit pattern.
The chip is very small, the circuit diagram is very complex, a chip even dozens of layers of circuits, and in terms of transistor density, under the current 3nm process, every square millimeter has reached nearly 200 million transistors, so the lithography process is very complex and fine.
Therefore, the resolution of the lithography machine determines the process of the chip to a certain extent, so for the resolution of the lithography machine, there are several lithography machines, such as I-line, G-line, KRF, ARF, ARFI, and EUV.
Among them, the ARFI lithography machine, also known as the immersion lithography machine, is the most widely used lithography machine, because it is only an improvement of the ARF lithography machine, the light source is the same, all use 193nm, but a layer of water is added before the wafer as the medium, after the light passes through the water, after refraction, the equivalent is 134nm, the wavelength is short, and the resolution is high.
Therefore, ARF can theoretically support up to 65nm process, while ARFI can theoretically support up to 7nm in everyone's mind.
But in fact, what everyone is not clear is that for the immersion lithography machine, 7nm is far from the end, if you don't count the cost, the immersion lithography machine, up to 3nm, 2nm is no problem.
There is a multi-faceted, misalignment issue involved.
For example, if a chip has 10 layers of circuit diagrams, we can split it into 30 layers, or 40 layers or more, so that through multiple lithography processes, we can finally achieve higher resolution. But every time you lithography, there can be no error in the middle, which is the problem of alignment.
Just like if you want to rub a "king" character on a relatively small piece of paper, it is definitely not possible to rub it directly at one time, because the paper is too small, so you will be divided into n times to rubbing, the first time only to rub a horizontal "one", and then slowly move the paper, and then extend a horizontal "one", and then extend a vertical "|Then it became "dry", and finally it became "king".
Every time you extend a stroke, you have to move the paper slowly, but the error must be controllable each time, which is the difficulty.
The multiplicity of the chip is actually the same reason, as long as the error is small enough each time, in fact, a chip with a very small process can be photoetched onto the silicon wafer N times.
But under this method, the cost is very high, after all, one lithography used to be enough, but now a lithography becomes 10 times, or even 20 times, and the time spent is 20 times more, which is equivalent to 20 times higher cost.
Considering that every time more than once, the yield is reduced, so in fact, the cost is higher than 20 times, the chip is commercialized, to be marketized, the cost is too high, it cannot be used, so the general immersion lithography machine, after manufacturing 7nm chips at most, will no longer move forward, because the cost is so high that it cannot be commercialized.
But it doesn't mean that its limit is 7nm, in fact, theoretically speaking, there is no so-called limit, as long as it costs at any cost, it can be tossed all the ......time