[EETimes] Newsflash: Samsung 7nm EUV SRAM size of 0.030um2 (probably ND)

witeken

Diamond Member
Dec 25, 2013
3,899
193
106
Samsung SRAM size of 0.030um^2

In a sign of the status of the underlying process technology, TSMC will unveil what organizers called “the smallest SRAM bit cell published to date.” The 0.027μm2 256 Mbit SRAM is made in a 7nm FinFET process using write-assist circuitry for low Vmin applications.

The Taiwan foundry’s rival, Samsung, will not be far behind with a 7nm FinFET SRAM less than 0.030μm2 in area. Interestingly, the part is made using extreme ultraviolet lithography for peripheral repair analysis, an approach said to improve Vmin by 39.9 mV based on failure analysis.

Samsung announced earlier this year it plans to put EUV steppers into production earlier than rivals, in part because it is testing them in both logic and memory fabs. TSMC has said it will wait for 5nm before inserting EUV.

From the 'earlier this year' link:
As part of the news Samsung revealed it will skip the 7nm node with immersion lithography planned by rivals TSMC and Globalfoundries. Instead it aims to have in production a 7nm process with extreme ultraviolet lithography in late 2018

It's not clear of the reported 0.030um^2 number is the high density or low density cell, but it's a bit meager given that Samsung will indeed not bother with immersion after the 10nm node.

The normal density and high density numbers of Samsung 10nm are:

0.049 / 0.040

If the 0.030 is the normal density cell, then that is a 1.63x shrink from 10nm.

You can read the article for more of the presentations that will be given at next year's ISSCC, but there will be no 3D XPoint one.
 
  • Like
Reactions: Ajay

FIVR

Diamond Member
Jun 1, 2016
3,753
911
106
And intel's supposed "process advantage" continues to dissipate. Even if we pretend to know intel's actual density numbers by assigning their advertised SRAM density, they still come up short.


Really I wonder if Intel will pull AMD's move of selling off their foundry. They really have no practical or theoretical advantage here anymore, so why bother? I guess it all comes down to their lack of actual work on their architecture... not much they can do after sinking $B after $B on their foundries while ignoring their architecture and merely working on marketing tactics to get market share and increase margin. They are great business people, not so much a great CPU design company. Now they aren't even a premier foundry, falling behind TSMC and Samsung both.
 
  • Like
Reactions: lefty2

witeken

Diamond Member
Dec 25, 2013
3,899
193
106
And intel's supposed "process advantage" continues to dissipate. Even if we pretend to know intel's actual density numbers by assigning their advertised SRAM density, they still come up short.


Really I wonder if Intel will pull AMD's move of selling off their foundry. They really have no practical or theoretical advantage here anymore, so why bother? I guess it all comes down to their lack of actual work on their architecture... not much they can do after sinking $B after $B on their foundries while ignoring their architecture and merely working on marketing tactics to get market share and increase margin. They are great business people, not so much a great CPU design company. Now they aren't even a premier foundry, falling behind TSMC and Samsung both.
And people continue saying Intel's process lead is dissipating. At least when they say that they also admit Intel *currently* has a process lead :D.

Have you looked at the production dates? End of 2018, and that's only if ASML does not fail to meet their targets.
 

FIVR

Diamond Member
Jun 1, 2016
3,753
911
106
At least when they say that they also admit Intel *currently* has a process lead :D.
Actually the scare quotes along with the "supposed" part were indicating that I do not, in fact, believe intel has any process advantage... but I do believe many forum members here (like yourself) think they do.
Have you looked at the production dates? End of 2018, and that's only if ASML does not fail to meet their targets.
Are you taking intel roadmaps seriously? Why wouldn't they just tweak the architecture on 14nm, and create "Tea Lake" or something to pretend they still have something new to offer customers instead of actually producing a 10nm product? They've done this how many times now (Kabylake, Coffee Lake)?

It's pretty clear intel won't even be taping out cannonlake before TSMC and samsung have 10nm products in the pipeline. On top of that, they already are at a disadvantage to TSMC on 16nm, their process advantage is a myth.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
I think we'll have to wait and see what Intel's 10nm process actually looks like to make any judgements. Even if Intel process lead is shrinking - they still have and advantage with their CPU lines because they can simultaneously tailor their designs to nodes under development - so it makes no sense for Intel to divest as they hold a competitive advantage. I think that node advantages are likely to shrink as consolidation continues - those who remain have increasing larger R&D budgets to compete against Intel.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
It's pretty clear intel won't even be taping out cannonlake before TSMC and samsung have 10nm products in the pipeline.

Cannonlake pretty much has to be taped out by now, if not for some time.

If SS 7FF has EUV, even partially, that's a big problem for Intel since Intel's 10 nm likely doesn't have EUV.

I think we'll have to wait and see what Intel's 10nm process actually looks like to make any judgements. Even if Intel process lead is shrinking - they still have and advantage with their CPU lines because they can simultaneously tailor their designs to nodes under development - so it makes no sense for Intel to divest as they hold a competitive advantage. I think that node advantages are likely to shrink as consolidation continues - those who remain have increasing larger R&D budgets to compete against Intel.

There was always a concern that Intel would slow releases not because of technical capability but because it would have cost too much. Especially now that PC sales continue to tumble, Intel doesn't have the volume advantage anymore.
 

VirtualLarry

No Lifer
Aug 25, 2001
56,229
9,990
126
Especially now that PC sales continue to tumble, Intel doesn't have the volume advantage anymore.

What kind of volume advantage does the ARMy have over Intel, if any? I think cell-phone sales and tablet sales have started to slide, too.
 

Rifter

Lifer
Oct 9, 1999
11,522
751
126
What kind of volume advantage does the ARMy have over Intel, if any? I think cell-phone sales and tablet sales have started to slide, too.

are you serious? since i purchased my last intel CPU i have purchased 9 Items with ARM CPU's(2 phones, 2 tablets,1 router, 3 R Pi's, and 1 TV) I'll eat my hat if ARM isnt producing more CPU's vs intel on at least a 3 to 1 margin.
 

teejee

Senior member
Jul 4, 2013
361
199
116
What kind of volume advantage does the ARMy have over Intel, if any? I think cell-phone sales and tablet sales have started to slide, too.
There were 4 billion ARM CPUs sold in Q4 2015... About 15 billion CPUs a year. This is about 40 times more than Intel. Intel is actually a small niche player compared to the ARM camp.

Sent from my LG-D855 using Tapatalk
 

witeken

Diamond Member
Dec 25, 2013
3,899
193
106
There was always a concern that Intel would slow releases not because of technical capability but because it would have cost too much. Especially now that PC sales continue to tumble, Intel doesn't have the volume advantage anymore.
It's not abou
There were 4 billion ARM CPUs sold in Q4 2015... About 15 billion CPUs a year. This is about 40 times more than Intel. Intel is actually a small niche player compared to the ARM camp.

Sent from my LG-D855 using Tapatalk
It's not about the volumes (and you've got to measure mm^2 sold, not chips sold), it's about revenue/income. Intel is still number one on the semiconductor list, so no worries there.
 
  • Like
Reactions: Ajay

ehume

Golden Member
Nov 6, 2009
1,511
73
91
There were 4 billion ARM CPUs sold in Q4 2015... About 15 billion CPUs a year. This is about 40 times more than Intel. Intel is actually a small niche player compared to the ARM camp.

Sent from my LG-D855 using Tapatalk
Does this include server CPU's?
 

oak8292

Member
Sep 14, 2016
81
67
91
The number of processors does not necessarily relate to a volume advantage and the revenue is only an indication of a long held advantage.

The advantage has to do with the lowest cost transistors. Intel had a very large transistor cost advantage with both die volume and wafer volume on the most advanced node. High volume wafers improve process and die yields to reduce transistor costs. High volume designs spread the non-recurring engineering cost to reduce transistor costs. About five years ago Intel had both of increasing wafer volumes and die volumes.

Without arguing extensively about the node density differences between Intel and TSMC the cost and the performance of the transistors is now close enough that Samsung and TSMC transistors are competitive if not lower in cost. Intel has a very expensive SG&A and R&D structure. TSMC is producing about 100K wafer starts per month on the 16/20nm node to give very good yields. Apple is producing over 200 million die on a single die design, this is probably higher than the highest volume Intel 2+2 die. If TSMC keeps reducing transistors costs at the projected pace it will be a very competitive environment going forward.