>>515027426
When you're making a processor, lets' say, you take what is called a silicon wafer and process it. The wafer is bigger than the individual die of said processor, so usually there's multiple of them on one wafer. However because we're talking about building nanometre scale structures, any imperfection will result in chips that perhaps still run but not very well or are just outright useless.
The still running ones are often downgraded in various ways and that's why, for instance, Intel makes an i3, i5 and i7 processors. They are the same, they come from the same wafer and ideally Intel would love all of them to be i7's but they just can't be. As such they downgrade the faulty ones until they're stable and run without errors. Turn off some cores, lower the voltage and the clock speed etc. etc. until they're good.
The "yield" is a proportion of chips of desired spec to the number of chips on the wafer in total. Because improving it requires fixing all sorts of microscopic issue - let's say some cavity in the machine makes it easier to maintain it but also creates a very tiny amount of draft or moves the resonance frequency of the whole assembly by a tiny bit and the disturbances it causes reduces the reliability of the process by 0.5% - you can't really simulate these things in any way, you just have to diagnose and fix them one by one while iterating on the machines.
I assume the Chinese have do a lot to catch up here, because - a public information you can find by reading public newsletters - we know that the same Chinese memory chips that prompted ASML ban are now produced with less layers(memory in modern semiconductors is kind of 3-dimensional structure) which indicates that they reliability just isn't there yet.