It is difficult to get an idea of the difficulty of developing EUV technology. To get a rough idea of what this important industry milestone means, we have to remember that its development started in 1997 and now, 23 years later, it is ready to be used on a larger or smaller scale.
A simple example of the difficulty at the start of the EUV journey is the bottom figure, which reflects our galaxy, where we are the orange point (solar system) while the EUV technology It would be about 100,000 light years away, at the top of the spiral, inaccessible to humans. Progress has been difficult, costly and complex, so the big guys are fighting to get there first, so how is the battle going?
ASML holds the global key to EUV technology
As we have already seen in other articles related to ASML and EUV technology, only this European company has the necessary technology and high performance scanners capable of bringing wafers made with this technique to life.
The problem is logically to build each scanner in record time and without the slightest failure in precision. On November 8, 2018, ASML on its Investor Day said that in 2020 it had two predicti ons for its scanners:
- High demand -> 35 EUV scanners provided.
- Low demand -> 33 EUV scanners provided.
But all forecasts have been truncated, as of this 2020, the company will close with 40 units shipped, including 4 High-NA units for Samsung of the latest generation. This gives us a clear picture of the demand and pressure on ASML from TSMC and Samsung, as the latter’s customers push even harder to have the densest and cheapest EUV nodes, increasing their profits and s ‘imposing. to competition.
A clear example is Apple, where TSMC almost exclusively manufactures its 5nm EUV for Bitten Apple chips. After EUV, as we have seen, High-NA will arrive, the next step of this technology, with which ASML, in high demand, plans to deliver 9 scanners + 55 traditional EUV until 2025, while in low demand , they plan to provide 7 + 43.
For next year, TSMC and Samsung are together expecting 80 EUV scanners, but it is possible that this high number will not be reached, because as it happens in consoles or GPUs at the moment, the demand is higher than the offer, much higher, and costs are enormous (318 million per scan on average and lower).
TSMC steps on the accelerator and already finalizes its 3 nm and less
The TSMC data is brutal, since in 2018 it was selling 7nm wafers and since then until today this lithographic process represents 35% of the total, and if we add to that the 5% goes up to at 43%, indicating that the business relies heavily on its new nodes.
Next year these two processes are expected to account for over 51% of shipments and this production will finally focus on EUV. Demand is such that when Huawei kicked out of the conglomerate of companies that need TSMC, its gap was instantly closed by rivals (-15%).
What will TSMC need for years to come? By 2020 and at the start of the year, it was calculated that the company would need 35 new EUVs to start producing the 5nm for Apple and the 3nm pilot tests, which turned out to be very accurate. as a prediction because it overflows with ASML. .
By 2021, 5nm will transition to high performance with AMD at the helm and 3nm will, in theory, enter risky production, so the number of scanners is expected to be increased to at least 54, a number which will be seen lightly. surpassed by 2022. That year the 3nm is expected to go into mass production and start with the 2nm testing, or at least with the new 3nm +, as it looks like they will have issues with the GAA.
The years 2023 and 2024 will need 62 new scanners for each of them and therefore a minimum number of 292 units is expected until 2025, an average that if all goes well, it should approach 60 scanners per year, 5 for each month.
Samsung tests its high-performance NAND flash with EUV
As strange as it may sound, Samsung is using its NAND Flash memory chips to test its EUV technology, and while this is not an announcement as such as the company has made almost no statement about it, they have left a pearl about it long ago with their 4th gen 10nm DRAM, saying it was made with EUV
Samsung’s bet is astronomical for EUV, literally. In 2019, he invested 133 billion won in a business strategy for 2030, where it has hired 15,000 professionals, from teachers to doctors, including engineers, in all sectors and fields, its objective is therefore to overthrow TSMC as a leading foundry, there is nothing .
Samsung’s problem is precisely the TSMC itself, as it was previously to order scanners. Therefore, this 2020 could not have 9 scanners, and the forecasts are not much better, since it is estimated that they will reach 20 units each year until 2025, which is almost 3 times less than TSMC.
The most realistic predictions are that by 2025 TSMC would have around 353 scanners and Samsung 119, but, since there is always a but, Samsung would be given the first High-NA units, which TSMC will not have.
Intel is upside down right now, they are looking for a new CEO because Bob Swan hasn’t set clear goals and where he has they haven’t been met. Surely the 7nm setback was his condemnation in the company …
In any case, we are currently discussing the quantity of production that will be allocated to TSMC and especially for which products. They talk about the new High performance Xe GPU and low performance, Pentium and Celeron processors and maybe a new batch chipset thinking of the LGA 1700 socket.
All very volatile for the moment, nothing confirmed and therefore too much fog in the blue giant that it must dispel if it wants to compete in the years to come as a foundry.
ASML is and will be overwhelmed despite the investments made
The reality is ASML goes as fast as it can get, but scanners have to be assembled and manufactured by engineers in what’s called clean rooms, known internally as Cabin.
Therefore, ASML must create a series of additional buildings to increase these cabins and thus meet the demand, where by the end of this year 35 of them. But how long does it take to assemble a complete and functional scanner? Well almost 12 months per unit and of course that brings up an adjacent question as ASML is not going to create 80 cabinets to meet customer demands, ergo, how do you expect to get shipping 80 high scanners EUV performance?
Shorten the assembly period of each unit. To achieve this number of units, each scanner must be assembled in less than 6 months with the current infrastructure, so for this, all parties and companies involved in creating the necessary parts must tighten their belts.
The main one would be Carl Zeiss, as he is in charge of the optics and the EUV super reflective mirror of each scanner. It seems that the company has increasingly refined the technique of creating these mirrors, which, combined with the fact that other parties are also ramping up production, could allow ASML to increase production to a volume of 65 or 70 units per year.
Complicated? A lot, as predictions are usually insufficient, but there is no one in the industry who can do what they do with the technology required for EUV and on-chip etching, so from there all customers will need to understand that units are limited and therefore the number of wafers will be the one available each quarter, which they are already used to as this has been happening since 2018.