Samsung and TSMC Roadmaps: 8 and 6 nm Added, Looking at 22ULP and 12FFC
by Anton Shilov on May 5, 2017 9:30 AM ESTBeyond 10 nm at TSMC: 7 nm DUV and 7 nm EUV
As noted previously, TSMC’s 7 nm node will be used by tens of companies for hundreds of chips targeting different applications. Initially, the company plans to offer two versions of the manufacturing technology: one for high-performance, and one for mobile applications, both of which will use immersion lithography and DUV. Moreover, eventually TSMC intends to introduce a more advanced 7nm fabrication process that will use EUV for critical layers, taking a page from GlobalFoundries’ book (which is set tp start 7 nm with DUV and then introduces second-gen 7 nm with EUV).
TSMC’s first-generation CLN7FF will enter risk production in Q2 2017 and will be used for over a dozen of tape outs this year. It is expected that high-volume manufacturing (HVM) using the CLN7FF will commence in ~Q2 2018, so, the first “7-nm” ICs will show up in commercial products in the second half of next year. When compared to the CLN16FF+, the CLN7FF will enable chip developers to shrink their die sizes by 70% (at the same transistor count), drop power consumption by 60% or increase frequency by 30% (at the same complexity).
The second-generation 7 nm from TSMC (CLN7FF+) will use EUV for select layers and will require developers to redesign EUV layers according to more aggressive rules. The improved routing density is expected to provide ~10-15-20% area reduction and enable higher performance and/or lower power consumption. In addition, production cycle of such chips will get shorter when compared to ICs made entirely using DUV tools. TSMC plans to start risk production of products using its CLN7FF+ in Q2 2018 and therefore expect HVM to begin in H2 2019.
| Advertised PPA Improvements of TSMC's CLN7FF Nodes Data announced by TSMC during conference calls, press briefings and in press releases |
||||||
| 7FF vs 16FF+ |
7FF vs 10FF |
7FF EUV vs 7FF |
5FF EUV vs 7FF EUV |
|||
| Power | 60% | <40% | 10% | lower | ||
| Performance | 30% | ? | lower | higher | ||
| Area Reduction | 70% | >37% | ~10-15-20% | tangible | ||
| HVM Start | ~Q2 2018 | - | ~H2 2019 | ~H2 2020 | ||
As it turns out, all three leading foundries (GlobalFoundries, Samsung Foundry and TSMC) all intend to start using EUV for select layers with their 7 nm nodes. While ASML and other EUV vendors need to solve a number of issues with the technology, it looks like it will be two years down the road before it will be used for commercial ICs. Of course, certain slips are possible, but looks like 2019 will be the year when EUV will be here. In fact, keeping in mind that both TSMC and Samsung are already talking about their second-gen EUV technologies (which they call 5 and 6 nm) that will use more EUV layers, it looks like the foundries are confident of the ASML TwinScan NXE manufacturing tools (as well as of the Cymer light source, pellicles, photoresists, etc.) they are going to use.
Facebook
Twitter
RSS
89 Comments
View All Comments
melgross - Saturday, May 6, 2017 - link
A lot of chip experts don't believe that a true 5nm is possible. Not because we can't build it, but because the laws of physics are closing in. At that point, we have no substitute for FinFET, which doesn't work below 7 nm, and the three technologies that have been considered as a replacement aren't working either.When you begin to have features that are just 10 to 12 atoms wide, Heisenberg's Law hits you hard. As many electrons that travel through the feature, escape it. That's a death hell. So I expect that 7nm will really be 10 to 14 for most fabs, the way 14 is really 16 to 20 for most now.
The next step is expected to be carbon nanotubes, which both Hp and ibm have been working for years, and have shown limited success. That hoped to be ready, in limited complexity, by 2005 to 2030.
But there will be a wide gap between any silicon technology and that, even assuming they can get it working on a commercial basis at all. There are still too many steps for that, and they don't yet know how to climb them, or even if they're there.
melgross - Saturday, May 6, 2017 - link
Oops, too many typos. I meant, ready by 2025 to 2030, of course.Meteor2 - Sunday, May 7, 2017 - link
You're right, but I think between EUV and further development of the new gate concepts we'll make 5 nm happen. Although as really it's a question of whether commercial interests will fund the R&D, rather can 'science' make it happen, I suppose there's a risk 5 nm won't happen as designing such chips will be fantastically expensive. Will we be prepared to spend the $$$ for the performance which would be delivered?lefty2 - Friday, May 5, 2017 - link
Intel already has lost it's process advantage. Samsung's 10nm is currently in HVM and denser than Intel's 14nm. Intel say they will launch 10nm in 2017, but the yields are so bad they can hardly be considered production yield. By the time it reaches production yield TSMC will have 7nmDrumsticks - Friday, May 5, 2017 - link
Intel's 10nm is going to be denser than Samsung or TSMC's 7nm imo, going by the numbers we see here. Intel's 14nm is already denser than their competition by somewhere in the realm of 30% (per the hard numbers Intel released a few weeks back, and nobody has contradicted them). Intel's jump to 10nm is going to provide ~2.7x higher density than their 14nm node, and I think they've said several times they plan to ship 10nm this year.Even with a 70% area reduction on 7nm vs 16nm at TSMC, I don't think that overcomes a 30% lead + a 2.7x increase in density on top of that lead.
For another comparison, Intel's 10nm measures 100M transistors / mm^2, versus their competition at 50M / mm^2 at 10nm. Assuming TSMC's transistor density is around the "Others" metric, a 37% reduction in area from 10nm to 7nm would still leave them short of Intel's process node. I suspect everybody else will need a 5nm node to temporarily jump ahead of Intel's 10nm, before Intel's 7nm rolls around in 2020 or something and puts everybody behind again.
Numbers come from https://newsroom.intel.com/newsroom/wp-content/upl... which based on most of what I've seen has been accepted as a well done report. I'd love to see everybody switch to a more objective metric, since process node is now just a marketing game.
vladx - Friday, May 5, 2017 - link
They obviously can't compete with Intel head-to-head so say they have to resort to marketing gimmicks to make it appear they're coming ahead.SuperMecha - Saturday, May 6, 2017 - link
Intel's 14nm density advantage compared to GF/Samsung's 14nm process is only 23% not 30%. Also Intel's recent presentation only compares up to their competitor's 10nm not 7nm. In 2018 Intel will lose it transistor density advantage.https://www.semiwiki.com/forum/content/6713-14nm-1...
Drumsticks - Saturday, May 6, 2017 - link
I know Intel compared to the 10nm products; I was just extrapolating from that based on what TSMC stated (in the article) about their 7nm vs their own 16nm and 10nm. With everybody using different statistics now for each of their parts, it's not surprising that Intel gets passed every now and then, considering the time between their nodes is getting longer and longer.lefty2 - Saturday, May 6, 2017 - link
Scotten Jones did a detailed analysis of various leading edge nodes and concluded that TSMC's 7nm is slightly denser than Intel's 10nm and Samsun/TSMC 10nm is slightly desnser than Intel's 14nm:https://www.semiwiki.com/forum/content/6713-14nm-1...
Intel's 2017 launch of 10nm is virtually a paper launch. They are only going to release a couple of low volume SKUs at the very end of 2017, just so they can claim that they have the process lead. It's not till late 2018, or 2019 that the bulk of their products go to 10nm. Also the first iteration of 10nm performs worse than Intel's current 14nm+ process.
melgross - Saturday, May 6, 2017 - link
I don't believe it. First of all, neither he, or anyone else outside those companies actually knows enough about the actual chips to know the true density. Evaluating these by making some basic mathematical calculations doesn't tell us anything about the actual processes. It's all theoretical.