Is it possible and feasible today to reduce the size of transistors below the 3 nm mentioned? The industry is opening several fronts with ASML at the forefront, while many others wonder how we are going to create masks for such small wavelengths. What are the problems?
New scanners, new resistance processes and masks with EUV High-NA
There are three main pillars of wafer engraving today and for years to come: scanners provided by the aforementioned ASML, resistors for the masks, and the masks themselves.
Why are scanners a problem? Mainly because reducing the wavelength of 193nm is really tricky because you need a digital aperture (NA) lens of 0.33 with resolutions of 13 nm and with a yield per tranche as high as possible.
EUV is a technology that works with laser pulses that hit small tin droplets at a very high speed, which generates photons and these bounce off various mirrors inside the scanner to reflect off the mask and then on the wafer, shaping the transistors.
The problem is that the photons don’t end up on the wafer, but stay in the mask, or are directly lost at the edges. The solution would be to use films that protect the wafer, but the technology is not ready so, pushed by the competition, everyone has been shot to record at 193 nm without them, which is doable at 7 nm and future 5 nm.
And this is the case because a single, not multiple pattern is used, as it can have either a 32nm or 30nm step gate, but reducing it involves changing the scanner to 3nm or less. The answer will again be given by ASML with a 0.55 NA capable of recording at resolutions of 8nm, a giant leap forward.
These scanners will be referred to colloquially as High NA and they will allow a return to the uniquely designed model, reducing costs and gradually increasing platelets per hour. The problem is, for them to work, you need two masks and not one, which must be sewn together for a wafer due to their complexity.
This will skyrocket the cost, which is already prohibitive today, because, for example, a current EUV scanner exceeds the $ 153 million, while the High-NA will be on 318 million. Does this mean that the price of processors will increase, as well as GPUs? It is very likely if they cannot find a solution.
The resistors for the wafers are also at an impasse
Before creating a pattern for a wafer, it is inserted into a system called a coater, after which photoresist materials are poured into the wafer, which rotates to leave an even layer to protect it.
After that, the engraving will arrive in the lithographic scanner, so this is a very important preliminary process. There are currently two systems for doing this in EUV: chemically amplified or CAR and those based on the call metal oxide.
The problem with CARs is that they suffer from the blur of the scanner lens, which doesn’t allow such high resolutions in them, let alone the ones that come with NA. Metal oxide resistors are more suitable for these failures, but they are less stable, so there is no 100% reliable and above all profitable system.
However, a system called a dry photoresist is being developed, a system to remove moisture and improve resolution and stability by coating it in a dry processing chamber before exposure to photons. This will be valid for EUV and EUV NA, but it will take a few years to happen.
Photomasks are not spared either and need to improve
A current EUV mask consists of 40-50 thin layers of silicon and molybdenum on a substrate, resulting in a total thickness typically between 250nm and 350nm. Each mask has a component called an “absorber” which prevents light from bouncing off the wafer and allows for more perfect etching.
The problem is that in an EUV system, the light hits the mask at an angle of 6º, which reinforces the fact that it is reflected and that the absorbent cannot do all the work, causing possible changes of location in the patterns, with the consequent deterioration of the wafer.
How to solve this? With finer absorbents. The current ones are 60nm and the better and more expensive 50nm, but they are not able to solve this problem and less at increasingly higher wavelengths. For this reason, researchers are creating two new types of masks for future lithographic processes in EUV: the so-called High K y las phase shift
The latter are the simplest, as it suffices to change the absorbent material for a material with Low-N properties, leaving aside the tantalum. What is wanted is to block the light and prevent it from being reflected, to achieve dark areas and therefore to improve the precision of images with a more basic contrast.
Very little is known about the High-K, but it appears that they can be repaired with a laser, which lowers the final manufacturing costs, as the masks are extremely expensive and one is needed for each design.
In short, the 7nm and 5nm are guaranteed, but go down to 3 nm It won’t be that easy despite Samsung, TSMC and Intel claiming to have them ready by 2025 in the worst case scenario.