By using this site, you agree to our Privacy Policy and our Terms of Use. Close
fatslob-:O on 06 May 2014
Edit Thread
Soleron said:
fatslob-:O said:
Soleron said:

...

@Bold Do you have a source for that ? It doesn't make any sense considering the fact that most of the gains come from using FinFet transistors and how would a different transistor technology be related to chip area scaling ? If that were the case it still wouldn't make any sense for globalfoundries to license samsung's 14nm technology. 

OK, I'm possibly wrong on the first part - it may be and not or. However, if you can tell me to which "industry planar 20nm process" the 15% figure refers to that'd be great. I think they're outright lying on that one.

It's true that globalfoundries may have had alot of issues and delays in the past but at this point they've likely cancelled their 20nm process node and I don't think their customers would like to be held out for that long but to claim absolutely NO PRODUCTION of commercial 14nm wafers until 2016 sounds a bit overly critical. 

Yeah it's just that I don't trust them to deliver when they failed to do so at 32, 28, 20 and their own 14. I agree, their customers will be very upset. But they need GF as a second source to Samsung. The rest of the industry only has a TSMC vs Samsung choice right now, because they have no access to Intel and wouldn't pick GF as a first source.

What amount of chip area scaling did Intel claim for their 14nm process ? The way I see it this isn't about minimum feature sizes anymore until extreme ultraviolet lithography comes into the picture as the semconductor manufacturing industry has redefined it to about surpassing limits of last generation process technology. I remember them only claiming a chip area scaling of 35% better than TSMC's 20nm process node so in that sense Intel's 14nm process node also isn't a real 14nm node.

This is the slide: http://www.extremetech.com/wp-content/uploads/2013/11/14nm-2.png

Intel is saying 22->14 scaling will be better than their own 32->22 scaling. That's the important takeaway.

TSMC has always been denser anyway, that's why "only" 35% vs TSMC. However, despite being less dense, Intel's always had a technologically better process than TSMC (superior frequency/power curve). It's a tradeoff for density vs power vs performance. TSMC can't run their chips at 3.6GHz like Intel, and if they did they'd be 200W monsters.

Intel's 14nm process node and Samsung's 14nm process node are closer than you think it is. 

I see Samsung 14nm as basically Intel 22nm.

 

Samsung and globalfoundries are clearly referring to TSMC's 20/16nm process nodes but I don't think samsung is lying because they tested this out with some SRAM cells or so ? After all TSMC's 16nm node has ABSOLUTELY no chip area scaling compared to their last generation 20nm technology. All TSMC's 16nm process node is 20nm + FinFET. 

I have a good feeling about globalfoundries that this time they'll actually deliver for the most part. Samsung practically footed all of the R&D bills for them which is a good thing for AMD and the rest of the customers. 

Intel was always better than TSMC when it came to semiconductor manufacturing but this doesn't change the fact that it isn't about minimum feature sizes anymore. 

Well to be fair samsung's 14nm node is more than just Intel's 22nm node. It's definitely more closer to Intel's 14nm node than you give it credit for seeing as how samsung isn't just some small player in the industry. 

Edit: Found it ... 

The only other foundry is going to launch their "Planar 20nm process node" while giving no chip area scaling in it's next generation wafers is TSMC. Since both Samsung and Globalfoundries are attempting to aim for mass production by the end of 2014 they will essentially have a 12 month lead on TSMC. Intel's lead on them is at best only 10 months. The reasons why samsung and globalfoundries are skipping 20nm is because they agreed that it will have a short life span. The saddest part about all of this is despite the fact that TSMC has a clear advantage of chip area scaling with it's 20nm process node when compared to Intel's 22nm process node all of it will have been for nothing due to the fact that TSMC has hamstrung that advantage by the fact that it didn't implement FinFET technology which is just as important as having better chip area scaling.