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!
According to the rumor mill Intel 10nm has been delayed yet again and is now expected at the end of 2018. If true, this is the third slip for 10nm: 2017 to end of 2017 to 2018 and now end of 2018.
View attachment 20443
Remember, Intel is known for their tick-tock model where they either release a new process or new architecture every 18 months. Currently Intel has been stuck on 14nm since Broadwell was released in Q2 2014 (Tick). Instead Intel has been optimizing 14nm in preparation for the AMD onslaught (AMD is also on 14nm).
TSMC is now on a 12 month tick cycle releasing a new process version every 12 months. TSMC 7nm will be in production Q1 2018 (which has a slight advantage over Intel 10nm) and TSMC 7nm EUV will be in production in Q1 2019 (which has a significant advantage over Intel 10nm.
Given Intel has slipped 10nm I would not be surprised if Intel 7nm also slips into 2021.
We are at the GF Tech Symposium today and will update on GF's process roadmap tonight.
Dan, with nodes shrinkage coming to an end, do you see materials or architectures as the next area of progress? Are there any other ways of getting significant improvement?
For the mobile SoC I see density as the main driver. They need to pack more devices onto their chips. Look at the Apple benchmarks from TSMC 16nm to 10nm. The performance delta is huge versus the process delta. The benchmarks I am seeing on the new Nvidia GPU (12nm) versus the older version (16nm) are also huge! The reason being TSMC is in VERY close collaboration with Apple and Nvidia so the new process are more interested in Apple's and Nvidia's Law than Moore's.
Node shrinkage isn't coming to an end. Intel is just run by incompetent fools who have married themselves to the wrong business model. TSMC will continue to deliver on shrinks.
Yes, I just heard from AMD at the GTC event today. I would not expect AMD 12nm parts until later in 2018 though. It should get them a 15% die shrink and a 10% performance boost. The GF 7nm process is doing very well and should be out towards the end of 2018. It will get AMD a 40% performance boost and a 65% shrink over 14nm.
If AMD 7nm is out the same time as Intel 10nm that should give them a decent market share boost. All very positive stuff. The GF CTO Gary Patton did a great keynote. His sound byte was the best one: "This is a different GF!"
Yes, I just heard from AMD at the GTC event today. I would not expect AMD 12nm parts until later in 2018 though. It should get them a 15% die shrink and a 10% performance boost. The GF 7nm process is doing very well and should be out towards the end of 2018. It will get AMD a 40% performance boost and a 65% shrink over 14nm.
If AMD 7nm is out the same time as Intel 10nm that should give them a decent market share boost. All very positive stuff. The GF CTO Gary Patton did a great keynote. His sound byte was the best one: "This is a different GF!"
Dan its looking quite positive for AMD and GF. Intel Cannonlake seems further delayed to end 2018 which pushes Icelake into 2019. I would like to hear your thoughts on GF 12LP and GF 7LP and any feedback from the ecosystem. How competitive are these processes against TSMC 12FFN and TSMC N7. btw can we expect AMD Zen on 12LP in H1 2018 or is it more likely to be H2 2018.
According to the rumor mill Intel 10nm has been delayed yet again and is now expected at the end of 2018. If true, this is the third slip for 10nm: 2017 to end of 2017 to 2018 and now end of 2018.
Intel has reportedly rescheduled the releases for some of its next-generation Cannon Lake-based processors, mostly ones with an integrated GPU, to the end of 2018, which has already affected notebook brand vendors' new projects and their suppliers, according to sources from the upstream supply chain.
For now, it’s unclear when we’ll see Intel’s desktop-class Coffee Lake processors, though we do know that the company’s 10nm die shrink, code-named ‘Cannonlake’ is set for 2018 at the earliest, and the aforementioned, 10nm+ Ice Lake won’t be here until at least 2019.
Does anyone know if GF 12nm is similiar to TSMC 12nm, ie. an "optimised" 14nm with 6 track library? And if this is the case, how much work is it for to convert a 14nm 7/9 track tape out to 6 track?
The Digitimes article reports Intel reportedly rescheduled some Cannon Lake to end of 2018. Intel said in response to these reports that it will be shipping 10nm products near end of this year, starting with lower volume SKU, with a ramp in H1 2018.
Does anyone know if GF 12nm is similiar to TSMC 12nm, ie. an "optimised" 14nm with 6 track library? And if this is the case, how much work is it for to convert a 14nm 7/9 track tape out to 6 track?
That is a very good article, thanks for the link. The big difference here is that GF is making processes specifically for the AMD products. This has not happened since AMD owned their own fabs (32nm). Intel has always had this advantage of course. Apple and NVIDIA have this advantage at TSMC and that partially accounts for the industry leading products they have been releasing. Let's hope AMD follows suite.
The real test will be 7nm and I truly hope AMD get's their chips out on time and exceeding expectations. If so the AMD stock should go through the roof. Especially if the latest rumors about Intel 10nm are true.
The only 7nm downside is EUV. GF was not as positive as TSMC so that will be an interesting story to follow next year. After my last trip to Hsinchu I'm very confident TSMC will have 7nm (N7+) EUV out in time for the 2019 Apple products. I also believe current TSMC 7nm customers, which is just about every top semiconductor company, will move to N7+. It is easier to design to and you get a 1.2x density and a 10% performance or 20% power improvement. So hopefully GF will up their EUV game...
I also believe current TSMC 7nm customers, which is just about every top semiconductor company, will move to N7+. It is easier to design to and you get a 1.2x density and a 10% performance or 20% power improvement.
Xilinx, Broadcom, Huawei, MediaTek.... everyone except Qualcomm as they are back at Samsung. GF could be capacity constrained as well if the 7nm AMD parts take off.
This is old news. QCOM moved from TSMC 20nm to Samsung 14nm and 10nm. QCOM moved back to TSMC 7nm and is now using Samsung 11nm and 8nm. QCOM has always bounced around foundries since the beginning of time. The SVP of QCOM was at the GF Conference this week so they are probably using GF as a second source.
Yes, it will come to an end, at some point you will hit a brick wall. The van-der Waals radius of a SiO2 molecule is 0.3 nm, resist molecules are much larger (line edge roughness!) and you won't be able to split molecules into half. I predict that 3 nm will be the practical limit of what is profitable. It may be lower, but there *will* be a limit. Assuming a roughly 30% reduction as usual, that means, we will have some 3-4 technology nodes left before we hit 1 nm. Or one to two tech nodes until we hit 3 nm below the 5 nm currently in R&D...
