Chinese semiconductor industry

hvpc said:

When I was still in R&D I had worked with KLA tools and involved in many head-to-head evaluation of their challengers. Every time, the outcome is mostly the same. KLA has better application support, more thorough sets of tools to help us deal with false defects. All new comers always focus on detection sensitivity first, as this is the first prerequisite...a ticket into the boxing ring. But detection sensitivity is conducted on test wafers/reticles. Test patterns are small so the inspection area is relatively small, and because suppliers pick and chose test wafers/reticles with no random defects, the real vs false defect phenomenon is usually not observed on the initial head-to-head phase.

But, once we put new tool into production, that's when the problem comes up. When running inspection at the required sensitivity setting (setting that could capture all defects of interest), non-KLA tools would catch so much false defects that would overwhelm the system. Inspection would abort due to not enough memory to store/capture info on the plethora of false defects. When this happens, we'd need to run at lower sensitivity just to finish an inspection run, but this would be at the risk of missing real small defect of interest.

On 'normal' runs, where manageable number of false are captured, KLA has better auto-classification software that had been refined over the years with our (end user) input. User friendly features or customized features tailor to each specific fabs are not something new challenger SME would know about nor had come up with already. I would say, this is a big reason why we don't switch out from KLA even though we want a second supplier to keep KLA honest. Basically, I had not seen tools that beat KLA enough for it to make sense for us to adopt as a second vendor or to replace KLA outright.

On reticle inspection, KLA's forte is their die-to-database capability. Their ability to render a reference image from GDS file is unrivaled. Despite Lasertec having actinic light source that would give them better pattern and defect resolution...and hence better signal-to-noise, Lasertec was not able to achieve same level of GDS-to-reference image rendering. At best, Lasertec EUV reticle inspection is better than KLA's DUV die-to-die inspection. But for die-to-database inspection which is much more crucial in the R&D stage, KLA still wins out. This I was able to verify from talking to some former colleagues.

And at the end of the day, most of us prefer working with KLA because of the long history of collaboration and familiarity. That is also a moat that KLA enjoys against their challengers.

There's more reasons, but I'll just end right here. Signal-to-noise performance dictating real/false defect detection dynamic is what KLA had mastered during their multiple decade of dominance. It may sound simple, but it's what keeps KLA in the dominant position in the metrology/inspection/yield enhancement arena.

easier said than done. you don't see people rushing to buy Jincee inspection tools to replace KLA for 14nm...7nm.

with smaller resolution, the signal-to-noise issue gets worse. From what's happening at SMIC, they are still stuck at getting domestic tools to work efficiently at 28nm. I have not heard of major domestic inspection or metrology breakthrough that would enable replacing KLA at 14nm nor 7nm. This is why SMIC 14nm & 7nm capacity had been stuck at the same pre-Oct 7th sanction level....no change for a year now. This doesn't sound like domestic tools is making headway at 14nm/7nm, does it?

Click to expand...

Weaasel said:

Just make sure that you have alternatives that can replace any foreign equipment just in case that equipment is prevented from being sold to China.

Click to expand...

latenlazy said:

I respect Paul and I’m not saying you’re wrong here but officials who insist they would never sign up for something sign up for them all the time when the political conditions are right.

Click to expand...

Weaasel said:

Get it, they are the best at what they do and for a number for critical processes no one else comes even objectively remotely close. But as they are not a Chinese company, and especially much more so an American company, they just have to eventually be replaced and somehow China must be able to have entities that can replicate they measuring and detection instruments or any others that they produce... If they are as critically important in the supply chain as they are and have no peers from another country, it is then probably even a worse situation for China than it is with regards to lithography machines, as they are an American company...

Click to expand...

KYli said:

Please, Log in or Register to view URLs content!

Click to expand...

tokenanalyst said:

View attachment 120494

Click to expand...

Schmoe said:

As for ASML de-Americanizing, it will not matter for US sanctions because legislation would be enacted stating that any entity selling semiconductor equipment to China with certain performance parameters will be sanctioned. Iranian oil does not have any US IP or anything else to do with the United States, but yet the US can impose sanctions on any party importing it or involved in paying for it. US sanctions are simply what the US says are sanctioned. It is as simple as that, although the process the US follows to get there will have some variation.

Click to expand...

tonyget said:

In Jan 2023，havok said that fully domestic 28nm production line will take 2-3 years to complete. See these posts below

Chinese semiconductor industry

I heard stories about companies wanting to buy SMEE frontend machines, my guess is a least their I-line and/or KrF machines but their delivery times were simple just too long, so they just defaulted to ASML or Canon instead. It doesn't matter if the immersion lithography machine can process, as...

www.sinodefenceforum.com

Chinese semiconductor industry

I heard stories about companies wanting to buy SMEE frontend machines, my guess is a least their I-line and/or KrF machines but their delivery times were simple just too long, so they just defaulted to ASML or Canon instead. It doesn't matter if the immersion lithography machine can process, as...

www.sinodefenceforum.com

So the timeline will be 2025-2026

Click to expand...

tinrobert said:

I know about SMEE so it was not an oversight as to why I didn't include it in my article I published. I am waiting for actual facts on the company including actual node it can reach in FRONT END, not packaging, and revenues. There are more comments on on this Forum on SMEE than any other company, and as far as I'm concerned, its just speculation.
When I wrote a previous article about how SMIC got to 7nm about 18 months ago, it was using ASML's DUV immersion without the need for EUV. When I see SMEE metrics, I will include them.

Click to expand...