Samsung 2nm process

[Erinyes]

I want to see it. It's going to be the only mobile RDNA4 launch, and despite headlines RDNA in Exynos has matched Qualcomm's offerings every single year (you need to set power limits to be the same for comparing otherwise identical phone chassis, duh).

I'm not quite so sure it will be RDNA4 as IP may not have been ready in time for the SoC. So it might be RDNA 3.5, AMD mentioned that they leveraged their collaboration with their mobile partners and optimized for power efficiency.

 

[DegustatoR]

Samsung Withdraws Its Personnel From Its Taylor Plant Located In Texas Due To 2nm GAA Yields Unable To Improve Beyond The 10-20 Percent Range

A report states that due to abysmal yields of its 2nm GAA process, Samsung has little choice but to withdraw personnel from its Taylor plant

wccftech.com

Samsung Reportedly Scrambling To Boost 3nm GAA Chip Yields, Which Are Said To Be Three Times Less Than The Recommended Threshold To Start Mass Production

A report claims that Samsung is frantically finding solutions to boost its 3nm GAA yields, or risk losing out on more orders

wccftech.com

 

[MfA]

As per Ming Chi Kuo, Qualcomm may be the sole supplier for the Galaxy S25 as Samsung may not use the 3nm Exynos 2500 due to low yields

ASML has royally screwed the pooch, TSMC is the only one with the pellicle technology to actually make their machines work.

All the other process improvements are just fucking around in the margins as long as they can't match TSMC pellicles.

 

[Erinyes]

ASML has royally screwed the pooch, TSMC is the only one with the pellicle technology to actually make their machines work.

All the other process improvements are just fucking around in the margins as long as they can't match TSMC pellicles.

How is that ASML's fault? Pellicles are made by the foundry right.

And Samsung has been using EUV from 7nm, and while they might not have matched TSMC initially, the latest 4nm process almost does. So what is wrong with their pellicles?

 

[DegustatoR]

It has been confirmed that Samsung Electronics has completely postponed the start of construction and orders for its 4th factory in Pyeongtaek (P4) and the 2nd Taylor Foundry factory in Texas, USA. It is interpreted that Samsung Electronics has chosen a conservative approach to semiconductor investment, and it is expected to have an impact on the ordering schedules of its partners.

[단독] 삼성전자, 평택 P4⋅美테일러 2공장 발주 전면 연기 통보

삼성전자 테일러시 공장 구축 현장. [ⓒ경계현 미래사업기획단...

ddaily.co.kr

 

[trinibwoy]

Clearly the aliens have only shared their technology with TSMC.

 

[MfA]

How is that ASML's fault? Pellicles are made by the foundry right.

ASML developed pellicles which Mitsui now sells, TSMC uses their own. Samsung is also trying to develop its own.

Even though the mere suggestion sends ASML engineers into conniptions, I think it's possible TSMCs pellicles are active components. If you can electrostatically charge the tin debris, preferably before it hits the pellicle, you might be able to keep it away from the pellicle in the first place. Even debris on the pellicle can cause problems.

 

[Rys]

Plasma debris is much more likely to contaminate the optics, rather than the mask or pellicle. Mask contaminants are much more likely to be larger particles that make their way into the scanner. Plus, you don't always need to run a pellicle depending on what you're making.

You're talking like fancy pellicles are the only way to make leading edge nodes yield well.

 

[MfA]

TSMC is the only way to make leading edge nodes yield well, since the introduction of EUV.

The processing after exposure didn't see massive changes, unless Samsung suddenly became less competent that leaves some novelty in TSMC's EUV mask creation or their pellicles as the likely distuingishing factor, that's where the room is for in-house innovation in EUV. Their lead was too sudden and persistent to just chalk it up to skill.

 

[Deleted member 2197]

Samsung Reportedly Fails To Secure NVIDIA As An "HBM Customer", Witnessing Massive Business Impact

Samsung has addressed public about the firm's disappointing performance in the AI markets, especially its "failure" in the HBM segment.

wccftech.com

The Korean giant is having a tough time when it comes to securing the major industry players, such as the likes of NVIDIA, due to which Samsung has revealed that their Q3 2024 earning guidance would come in below market expectations. While there are several factors accounting for the reason behind this "economic downfall", Samsung's late entry into the AI markets, coupled with growing competition from the likes of SK hynix and TSMC, has pushed back the Korean giant, which they have admitted in a post addressing the general public.
...
Samsung says that they failed to secure a significant AI "anonymous" customer (via Reuters). While they didn't give us any names, the leading candidate is none other than NVIDIA, given the track record of developments we witnessed with NVIDIA and Samsung collaborating for an HBM partnership.

The Korean giant hasn't explicitly confirmed this, but based on what the firm claims, it looks like NVIDIA has pushed Samsung off the list. Samsung says the supply of high-end HBM3E AI chips has been delayed, putting them far behind the competition.
...
Not only this, but Samsung claims that the supply of traditional semiconductors from Chinese firms has hindered the company's business. Currently, the production capacity of China's semiconductor industry has grown massively, which has attracted companies looking to acquire lower-margin chips but in higher quantities. So, Samsung has seen a hit in this segment as well, which is why the firm's foundry business has been hit massively.

However, Samsung isn't expecting to undergo "radical" changes. In a recent statement by Chairperson Jay Y. Lee, the Korean firm won't spin off the foundry business but will continue to be indulged in the race. Right now, Samsung is faced with severe issues, including ineffective 3nm yields and troubles in the production of next-gen processes. Hence, the foundry division isn't really in the best of position for now.

 

[del42sa]

Intel & Samsung Are Reportedly Inking A “Foundry Alliance”, Sharing Production Facilities Along With Process Tech

Intel & Samsung Are Determined To Combat TSMC's Semiconductor Dominance Through An Extensive Collaboration​

​

 

[DegustatoR]

Samsung Foundry could abandon its 1.4 nm (SF1.4) process node initially targeted for 2027 production, according to industry leaker @Jukanlosreve. This decision comes amid ongoing yield problems with the company's 3 nm SF3 node and follows the shutdown of underutilized 5 nm and 7 nm production lines. This could significantly impact Samsung's technology roadmap, which had positioned SF1.4 alongside automotive-focused SF2A and SF2Z nodes.

Samsung Foundry in Trouble, Might Cancel 1.4 nm Node High-Volume Manufacturing

Samsung Foundry could abandon its 1.4 nm (SF1.4) process node initially targeted for 2027 production, according to industry leaker @Jukanlosreve. This decision comes amid ongoing yield problems with the company's 3 nm SF3 node and follows the shutdown of underutilized 5 nm and 7 nm production...

www.techpowerup.com

 

[MfA]

IMO ASML, Intel, Samsung, the US and the EU need to get together to buy TSMC trade secrets for an ungodly amount of money. TSMC's technology is clearly one step beyond ASML and necessary to run their machines economically.

Having ASML as a single source of EUV steppers is bad enough for the market, but TSMC becoming the only user worse.

PS. still convinced it's ion assist, apart from claiming vastly lower contamination of pellicles, TSMC also claims huge reductions in power ... the most likely source of power reduction is vacuum pumping, again something where ion assist can help.