You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
How do you feel if a customer would like to use 7nm from Intel? Would they block or delay the process even if it's available just to use 10nm for the time being?
Also, does anyone find it strange that Samsung 7LPP EUV doesn't have any announced customers? When TSMC who seems to be behind them has said to have 100 customers by 2019 on 7nm and 7nm EUV?
I thought I saw something from Dan that Intel exited the foundry business.
Qualcomm has been quoted as saying Samsung 7nm is easier to design to with EUV versus multi-patterning so clearly they have done some work. I have heard that now that GF is out of the 7nm race a lot of people are asking for Samsung's 7nm PDK because they don't want to be sole source. Having said that there is no question TSMC is currently the 7nm leader.
I thought I saw something from Dan that Intel exited the foundry business.
Qualcomm has been quoted as saying Samsung 7nm is easier to design to with EUV versus multi-patterning so clearly they have done some work. I have heard that now that GF is out of the 7nm race a lot of people are asking for Samsung's 7nm PDK because they don't want to be sole source. Having said that there is no question TSMC is currently the 7nm leader.
Yes I saw where Intel has exited the foundry business but I don't read this anywhere else. Something like this I felt would be public knowledge. However, if true, do you feel Samsung 7nm LPP EUV is the best? They seem to have Technology not even Intel was able to achieve. Maybe this could have a strong performance showing over TSMC 7nm EUV View attachment 22741View attachment 22742View attachment 22743View attachment 22744
We will see, I have sources who say they are trying to stay with FinFETs for 3nm and given that foundry 3nm is really around a 5nm process that is certainly achievable.
Yes I saw where Intel has exited the foundry business but I don't read this anywhere else. Something like this I felt would be public knowledge. However, if true, do you feel Samsung 7nm LPP EUV is the best? They seem to have Technology not even Intel was able to achieve. Maybe this could have a strong performance showing over TSMC 7nm EUV View attachment 22741View attachment 22742View attachment 22743View attachment 22744
Yes I saw where Intel has exited the foundry business but I don't read this anywhere else. Something like this I felt would be public knowledge. However, if true, do you feel Samsung 7nm LPP EUV is the best? They seem to have Technology not even Intel was able to achieve. Maybe this could have a strong performance showing over TSMC 7nm EUV View attachment 22741View attachment 22742View attachment 22743View attachment 22744
Samsung decided to go with EUV for their first 7nm process and now they are struggling to ramp it. Meanwhile TSMC's first generation 7nm is optical, yielding well, getting all the design wins and has a slightly higher transistor density over Samsung although Samsung has a sightly smaller SRAM cell size.
In my view TSMC is the current technology leader.
TSMC will ramp EUV early next year around when Samsung will probably ramp their 7nm EUV process for real and TSMC's 7nm EUV process is even denser than their optical process. TSMC 7nm EUV is denser than TSMC 7nm optical is denser than Samsung 7nm EUV.
Late next year TSMC is due for risk starts of a 5nm process that will be a lot denser than the 7nm processes and a lot denser than Samsung's competing 5nm process that is only a modest shrink from 7nm. I have heard development of TSMC's 5nm process is going very well.
TSMC is moving their process development forward in a very measured, well thought out way and they are executing extremely well.
So even with Samsung Single diffuse break and mixed diffused break you suspect they are still behind? This is supposed to increase performance and density.
Up until about 5 years ago, Intel wouldn't comment about technology development like this, they would simply demonstrate it with a SRAM test chip. Proof of life basically. Generally in the last few months of the year.
Note there is no test chip, this year. Most likely next year.
I'm curious to see how AMD 7nm Ryzen will perform against Intel 14nm+++. Intel said that PC uses burst performance where the server processors are more performance and power per watt. We should see what TSMC 7nm process lead does in the real world.
"Beyond FinFET, TSMC is planning horizontal nanowire, what they call gAA for gate-all-around. As the name implies, the source/channel/drain is built out of a number of wires running through the center of the gate giving even better control than FinFET. This is planned for 3nm. It has superior electrostatics for enhanced energy efficiency. Beyond FinFET they are also looking at germanium with Ge hGAA pFET that they have built having record performance (the picture showed a transistor using 4 nanowires"
TSMC did say that in 2017 but that may not be true in 2019. I do not recall details of 3nm mentioned at the 2018 Symposium. Things really did change for TSMC when Apple became a partner/customer 10 years ago. Apple has a preference for new processes every year thus the TSMC half node strategy. Apple will not delay iPhone introduction so TSMC will chose the path of least resistance so to speak. This is why you have N7 and N7+. This is why TSMC uses the same fabs for 20nm and 16nm, 10nm and 7nm. 5nm will definitely be FinFET so the question in my mind is: Will 5nm and 3nm share the same fabs? If so, it will most likely be FinFETS.
What Samsung provides is also a factor. Without Apple as a customer Samsung can take risks to be the "process technology leader" which is why Samsung bet on EUV for 7nm. Samsung may bet on gAA for 3nm but Apple and the other top fabless semiconductor companies may not take that risk.
AMD is not tied to TSMC like Apple and AMD has an existing relationship with Samsung at 14nm so I would not be surprised to see AMD at Samsung Foundry in the future. Maybe AMD can take risks with Samsung unlike Apple? AMD could also reap the benefits of being Samsung's #1 customer with close collaboration and custom processes.
How long it will take for Intel to work with leading technology from TSMC or SEC? There must be some plans to switch if they loose too many market shares to AMD ? beside time to market, when the cost of developping a new technology and the corresponding fab will be too much for Intel ? at 5nm ? if Daniel is correct about intel Foundry unit closing, all the revenues will come from internally for the foundry, which was exactly what Intel Foundry unit was trying to avoid ...
How long it will take for Intel to work with leading technology from TSMC or SEC? There must be some plans to switch if they loose too many market shares to AMD ? beside time to market, when the cost of developping a new technology and the corresponding fab will be too much for Intel ? at 5nm ? if Daniel is correct about intel Foundry unit closing, all the revenues will come from internally for the foundry, which was exactly what Intel Foundry unit was trying to avoid ...
Just to be clear, the rumour was not that Intel is closing their foundry/fab lines, but that they're closing down the Intel Custom Foundry program which was supposed to bring in ASIC customers from outside Intel (competing with TSMC and Samsung) but basically failed to do so.
From the look at Intel new time-line it seems a 2020 release for 7nm. "A new transistor optimization" sounds like 7nm. At least to me. View attachment 22764
Just to be clear, the rumour was not that Intel is closing their foundry/fab lines, but that they're closing down the Intel Custom Foundry program which was supposed to bring in ASIC customers from outside Intel (competing with TSMC and Samsung) but basically failed to do so.
So customers can still use it? Synopsys seems to still be helping customers. With the technology they have. I wouldn't be surprised if they did shut it down for outside customers to use. Now that they're entering the GPU market to fight AMD and Nvidia.
From the look at Intel new time-line it seems a 2020 release for 7nm. "A new transistor optimization" sounds like 7nm. At least to me. View attachment 22764
Sunny Cove (2019) is an improved architecture CPU in 10nm -- that'll be the downgraded but hopefully now working 10nm process, might be called 10+ (but it's really 10-...). Willow Cove (2020) is a further optimised Sunny Cove with process tweaks, presumably Intel will call this 10++ but it's based on the same process, it's certainly not 7nm. Golden Cove (2021) is another 10nm CPU with more features, and maybe (going by Intel strategy in the past) some more process tweaks, maybe this will be 10+++?
Timescales for 7nm are (according to Intel) completely decoupled from 7nm; logically you'd expect the first 7nm CPUs to emerge in 2022 according to the above dates, but maybe if 7nm really does deliver on-time (unlike 10nm...) the later 10nm CPUs will get canned in favour of 7nm. Unless 7nm is lower power/higher density but lower speed -- who knows, it's all guesswork/spin/marketing right now. Given how badly customers have been burned with Intel's 10nm delays I suspect nobody will believe any of their 7nm predictions until they start shipping CPUs in quantity with good yield...
So customers can still use it? Synopsys seems to still be helping customers. With the technology they have. I wouldn't be surprised if they did shut it down for outside customers to use. Now that they're entering the GPU market to fight AMD and Nvidia.
I don't see how anyone can use it, because to commit to an ASIC design you need a process and libraries that you know will work and Intel don't have that in 10nm right now. The "new" 10nm process will need new layouts and libraries and still isn't proven to work, you'd have to be crazy to choose this over TSMC 7nm because it's riskier, doesn't have the IP support, and is more than a year behind.
Similarly I don't see how a 10nm Intel GPU can hope to compete with 7nm GPUs from AMD and Nvidia; Intel have never managed to deliver a competitive architecture, and they're now behind in process technology which is critical for GPUs because power and chip size are crucial. It doesn't matter how much Intel would *like* to be competitive, clicking their heels together and wishing ain't gonna make it happen...
TSMC did say that in 2017 but that may not be true in 2019. I do not recall details of 3nm mentioned at the 2018 Symposium. Things really did change for TSMC when Apple became a partner/customer 10 years ago. Apple has a preference for new processes every year thus the TSMC half node strategy. Apple will not delay iPhone introduction so TSMC will chose the path of least resistance so to speak. This is why you have N7 and N7+. This is why TSMC uses the same fabs for 20nm and 16nm, 10nm and 7nm. 5nm will definitely be FinFET so the question in my mind is: Will 5nm and 3nm share the same fabs? If so, it will most likely be FinFETS.
What Samsung provides is also a factor. Without Apple as a customer Samsung can take risks to be the "process technology leader" which is why Samsung bet on EUV for 7nm. Samsung may bet on gAA for 3nm but Apple and the other top fabless semiconductor companies may not take that risk.
AMD is not tied to TSMC like Apple and AMD has an existing relationship with Samsung at 14nm so I would not be surprised to see AMD at Samsung Foundry in the future. Maybe AMD can take risks with Samsung unlike Apple? AMD could also reap the benefits of being Samsung's #1 customer with close collaboration and custom processes.
Daniel the transition to GAAFET from FINFET is not as complicated as you make it out to be. Atleast Samsung is claiming that 90% of the fabrication process is similar to FINFET.
According to the paper abstract the GAA transistor channels comprising the horizontal nanosheets that are completely surrounded by gate structures. Samsung calls this a Multi-Bridge-Channel (MBC) architecture, and says it is highly manufacturable as it makes use of approximately 90 percent of the company’s existing FinFET fabrication technology, requiring only a few revised photomasks.
With TSMC 5nm set for 2020 and a TSMC 5nm+ likely in 2021 its fairly possible for TSMC to bring 3nm GAAFET to HVM in 2022. Samsung 5LPP which is 7LPP with SDB and 6T is a 2020 process and still Samsung is aiming for a 2022 launch of 3nm GAAFET. I highly doubt TSMC will let Samsung overtake them to GAA from so far behind as it currently stands today.
I don't know about that. I just can't see Intel letting AMD run away with it for that long. It has to be 2020 when their 7nm is coming out. When the ex CEO spoke to President Trump last year. He brought up the new technology and creating jobs.
Do you think it will compete with TSMC and Samsung 5nm in density?
I doubt Intel 7nm is going to arrive in 2020 or 2021. The most likely timeframe is late 2022. TSMC is already having customers run test chips on 5nm with risk production and first tapeouts planned for Q2 2019 and high volume manufacturing in Q2 2020. If Intel 7nm is anywhere close to TSMC 5nm in development stage they would not have 3 CPU generations on 10nm. Moreover even though ICL will launch in late 2019 it would be in ultra low power tablets followed by mainstream notebooks in H1 2020. 10nm ICL for desktops is launching in Q4 2020 as there is a 14nm Comet Lake for desktops coming in Q4 2019.
ICL-SP is likely to arrive in Q1 2021 given that CSL-SP is expected to launch in late Q1 2019 / early Q2 2019 and CPL-SP in Q2 2020.
btw its not about Intel letting AMD run away. Its about Intel being able to keep up with TSMC's steady pace of leading edge nodes. Intel gifted away their leadership position by failing to deliver 10nm on time. TSMC 5nm is on track for risk production in Q2 2019 and HVM in Q2 2020. TSMC 3nm is likely to arrive in 2022. TSMC execution is steady and management is good. I think TSMC will gain huge silicon share based on leading edge HPC products from Intel over the next 3-5 years.
That is good information, thank you. TSMC generally goes down multiple technology paths to see which one fits the time-to-market requirements. Since TSMC did not cover Gaa at the 2018 Symposium it may not have progressed enough to update customers publicly. We will know for sure when the 3nm PDKs come out, or maybe it will be covered at the next TSMC Symposium. I did not see any papers on it at IEDM.
