TSMC 12nm node

[lefty]

According to a rumour at Digitimes, TSMC will introduce a 12nm process.

The upcoming TSMC 12nm process is actually a smaller version of the foundry's 16nm technology, which is already offered in three process variants, said the sources. TSMC originally planned to launch it as the fourth generation of its 16nm technology that will come with lower leakage and better cost characteristics, but has decided to introduce it as an independent new process technology, the sources indicated.

TSMC planning 12nm process technology

 

[Daniel Nenni]

This is true but their explanation is ridiculous:

"Taiwan Semiconductor Manufacturing Company (TSMC) is planning to roll out a 12nm process node to further enhance its competitiveness in the 28nm and more advanced technology segment, according to industry sources."

I guess this is to compensate for the 16nm vs 14nm debate that raged when it was first announced. They may be over compensating naming it 12nm however. IEDM is next week so we will know more then. Scott Jones will be at IEDM so it will be interesting to see if he updates his Standard Node Value by Node chart:

The 2016 Leading Edge Semiconductor Landscape

This was the most read blog on SemiWiki in 2016 by the way. Seriously, it went viral.

 

[simoncc]

I think this is a reaction to the 12nm FDSOI from GF which is on a 14nm BE. this is just marketing. I don't believe the BE will be different, maybe some further optical shrink and optmize libraries like smaller tracks for the cells and new bitcells.

If we look back at all the flavors in 28nm, from 28LP SiOn to 28HPC+ HKMG ( HPL, HP, HPM. HPC etc ... ) It was probably a mistake to use 28nm for all these nodes from a marketing point of view ( well it's also giving confidence on the technoloogy maturity ) knowing that the improvement was probably more than a node ( 30% speed/ 30% leakage reduction/30% area saving) between the first flavor and the last one of 28nm.

 

[Staf_Verhaegen]

I think this is a reaction to the 12nm FDSOI from GF

Also my thinking, if it is for marketing driven node names GF has been the clear leader and the rest following; starting with their 20nm FinFET node they called 14nm.

 

[IanD]

Didn't Intel start the "it's FinFET so we'll call it a smaller node" race-to-the-bottom?

 

[Daniel Nenni]

Didn't Intel start the "it's FinFET so we'll call it a smaller node" race-to-the-bottom?

Yes, Intel first called FinFETs "22nm Tri-Gate" both of which proved otherwise and now their marketing people are paying for their dirty deeds because based on this new process naming paradigm Intel has lost the process lead at 10nm to both TSMC and Samsung.

I would be surprised if Intel did NOT push their 10nm out in the first half of 2017 versus the second half to reclaim their "Process Lead" title before TSMC releases 7nm in Q4 2017.

 

[Jozo035]

I am not sure, if it is good idea releasing their 10nm ahead of shedule. Yes it should be good for their foundy buisness image but foundry is just minority of what Intel is doing.

In core market they plan to release Kabylake, Skylake-EP and KnightMill in 2017 and Coffelake in 2018. All of these are 14nm products and releasing 10nm before releasing these products should make them kind of outdated, long before release.

So it is question what is the priority for Intel. Process lead (which is anyway questionable IMO) or biggest part of their renevue today?

 

[astilo]

Also my thinking, if it is for marketing driven node names GF has been the clear leader and the rest following; starting with their 20nm FinFET node they called 14nm.

Lol Staf, GF borrowed the 14nm from Samsung. It is hard to be the "clear leader" when you follow...

 

[lefty]

I am not sure, if it is good idea releasing their 10nm ahead of shedule. Yes it should be good for their foundy buisness image but foundry is just minority of what Intel is doing.

In core market they plan to release Kabylake, Skylake-EP and KnightMill in 2017 and Coffelake in 2018. All of these are 14nm products and releasing 10nm before releasing these products should make them kind of outdated, long before release.

So it is question what is the priority for Intel. Process lead (which is anyway questionable IMO) or biggest part of their renevue today?

If they do release ahead of schedule it'll probably be limited to a very small amount of low volume CPUs, like they did for the launch of 14nm. That way the low yield won't effect their operating margins.

 

[simoncc]

Lol Staf, GF borrowed the 14nm from Samsung. It is hard to be the "clear leader" when you follow...

Doesn't matter. When you answer an RFQ and you offer a technology for a COT, the management will only see : Oh GF is offering 12nm and TSMC 16nm, it has a big impact even if the schedule is not the same. Then the customer will check the density and realize the 12nm is comparable due to back end limitation, GF will be able to sell the fact that 12nm offer more perf or less leakage etc. That put GF de facto in the position of leader in the comparison. I really don't see any other reason for TSMC to introduce this node. And if you remember well couple of years ago when 28FDSOI was promoted, TSMC did some slides to prove that their technology is better, and 28FDSOi wasnt a mainstream threat at that time. So I believe TSMC marketing team takes very seriously every competitors...

 

[Staf_Verhaegen]

Didn't Intel start the "it's FinFET so we'll call it a smaller node" race-to-the-bottom?

No, they did scale the back-end layer also for this node. They were the first to be able to use FinFET in production, this does mean they started the node naming war; Intel has been the least aggressive party in this node naming war.

 

[Staf_Verhaegen]

Lol Staf, GF borrowed the 14nm from Samsung. It is hard to be the "clear leader" when you follow...

I was talking about how they put it in marketing and the public announcements of future technologies. I was not talking about the engineering side. These are total different worlds.
But my memory may be failing. How I remember it is that at the time or right after TSMC announced their 20nm planar node GlobalFoundries started to do their 14nm FinFET marketing to compensate for the fact they would not do a planar 20nm node and way ahead of actual release of the process. As I remember it, Samsung was not doing heavy marketing on nodes a few years in the future. I think at that time GF still had to prove to their investors that they could survive as a stand-alone foundry.

 

[Staf_Verhaegen]

Yes, Intel first called FinFETs "22nm Tri-Gate" both of which proved otherwise and now their marketing people are paying for their dirty deeds because based on this new process naming paradigm Intel has lost the process lead at 10nm to both TSMC and Samsung.

The reason TSMC is getting ahead is in my opinion not because of how Intel called their 22nm node but because of engineering at TSMC being able to do a better job than Intel at the moment.
If it is caused by having more resources available due to higher volume, or better management of resources, or more efficient use of resources by closer collaboration with R&D centers like imec I don't know. It's likely a combination of all three.

 

[mcmanus]

12FF is part marketing, to be sure, but I believe (read "guess") it is also a way to quickly apply lessons learned and real manufacturing ability that they have picked up in their advanced node development and put it back into the 16FF world. The result (in theory) should be something that will provide slightly better PPA than 16FF at minor cost.