Is Intels fabrication process advantage shrinking?

[tipoo]

Just thinking about the Apple event yesterday. They said 1 billion transistors on the same die size (100mm2), which means they must have shrunk it to 20nm.
It's kind of amazing really, I remember when Intel was nearly two whole years ahead of the universe in process technology, now they're on 22nm while others are hitting it now. Sure, they'll be the first to 14 for sure, but the lead is shrinking, it must be in the months rather than year+ range now. I wonder if AMD can tap into that 20nm plant, and who is making it?

 

[Exophase]

There's no way Apple is releasing a phone in September 2013 that's on TSMC's 20nm. That's totally unprecedented, it'd be like 6 months earlier than expected. And no one else is close to ready with 20nm either.

Even then TSMC's 20nm isn't really comparable to Intel's 22nm since it doesn't have FinFETs. They won't be using a similar process until their so-called 16nm hits.

 

[ShintaiDK]

Why would it mean its 20nm? Nothing points to it being 20nm.

If anything, Intels process advantage keeps expanding. They already have a 2year+ lead.

 

[blackened23]

Just thinking about the Apple event yesterday. They said 1 billion transistors on the same die size (100mm2), which means they must have shrunk it to 20nm.
It's kind of amazing really, I remember when Intel was nearly two whole years ahead of the universe in process technology, now they're on 22nm while others are hitting it now. Sure, they'll be the first to 14 for sure, but the lead is shrinking, it must be in the months rather than year+ range now. I wonder if AMD can tap into that 20nm plant, and who is making it?

Intel is still 3 or so years ahead of everyone. TSMC does not have FinFETs, and they are actively trying to correct that situation - as mentioned above even TSMCs 20nm won't be comparable to intel's 22nm because of the lack of FinFETs. Intel's 14nm is ready, by the way.

There's a lot of marketing BS associated with TSMC, they won't have intel's low leakage characteristics due to the lack of FinFETs even when their 20nm is ready - and it won't be reading until 2H 2014. A7 will not be 20nm.

1 billion transistors on 28nm silicon, which the A7 is 28nm - isn't an impressive feat nor is it unprecedented.

 

[Imouto]

You guys should wonder why Samsung isn't giving up on their foundry business.

 

[VengenceIsMine]

Just thinking about the Apple event yesterday. They said 1 billion transistors on the same die size (100mm2), which means they must have shrunk it to 20nm.
It's kind of amazing really, I remember when Intel was nearly two whole years ahead of the universe in process technology, now they're on 22nm while others are hitting it now. Sure, they'll be the first to 14 for sure, but the lead is shrinking, it must be in the months rather than year+ range now. I wonder if AMD can tap into that 20nm plant, and who is making it?

There are an awful lot of design decision that affect density besides process tech, there could be significant amounts of cache for example, RAM density is waaaay higher than most processor area, it's just too early to know if Apple has really pulled off a coup and managed to get TSMC to deliver 20nm. Sure it's possible, and if it is, there will an absolutely FURIOUS Qualcomm for TSMC to manage but it is just to early to conclude that.

 

[SiliconWars]

Is Intel pulling away from the rest in terms of time to market new nodes? It looks like they might be, slightly. But, each new node has lesser gains so the gap shrinks at each. If we assume it evens out overall, that basically means Intel is maintaining their advantage. The big issue is, the cost goes up every node.

Intel's problems aren't on the ability level, but simple economics will bring this advantage to an end - or a crawl - eventually.

 

[Phynaz]

Intel's fabrication advantage is increasing, not decreasing.

 

[Khato]

Intel's fabrication advantage is increasing, not decreasing.

Exactly. TSMC's rough equivalent to Intel's 22nm process that they're calling '16nm' likely isn't going to show up until 2015... and then most of what I've seen indicates that they're hoping to have EUV tools available in order to make the next shrink past that. If they actually end up waiting for tools to be available they could easily fall even further behind.

 

[Fjodor2001]

Well what about the claim that some have been making that Intel's 22 nm process is not actually 22nm, but 26 nm or something?

 

[redhotiron2004]

Just thinking about the Apple event yesterday. They said 1 billion transistors on the same die size (100mm2), which means they must have shrunk it to 20nm.
It's kind of amazing really, I remember when Intel was nearly two whole years ahead of the universe in process technology, now they're on 22nm while others are hitting it now. Sure, they'll be the first to 14 for sure, but the lead is shrinking, it must be in the months rather than year+ range now. I wonder if AMD can tap into that 20nm plant, and who is making it?

AMD graphics would follow next with 20nm process. Actually, intel is not concerned with other players that much as long as it knows that it have better processors than the competition. It have the technology to go to 14nm and then 10nm pretty fast. But, it wont do it as long as they wont feel threatened. 14nm/10nm are there trump cards. It's something that makes then quite comfortable. They want to make maximum profit at every level knowing that they could easily beat the competition anytime they want.

By going to 14nm/10nm directly just because they can. Would only result in huge profits in the short term. As once that would be out others too would gain it after some time. And eventually intel would loose it's advantage. They want to make consistent long term profit. And are in no hurry to get to the finish line.

AMD is catching up fast and I hope it makes intel release 14nm processor in 2014 itself according to the original plan. Instead of 2015 which was post pond later on!

 

[Phynaz]

Well what about the claim that some have been making that Intel's 22 nm process is not actually 22nm, but 26 nm or something?

What about it?

Some people claim the moon is made out of blue cheese.

 

[Phynaz]

AMD is catching up fast and I hope it makes intel release 14nm processor in 2014 itself according to the original plan. Instead of 2015 which was post pond later on!

I don't know where you are getting your info, but 14nm CPUs from Intel (Broadwell) are shipping this year (2013).

In 2015 Intel will be shipping CPUs produced on a 10nm manufacturing process.

 

[USER8000]

I don't know where you are getting your info, but 14nm CPUs from Intel (Broadwell) are shipping this year (2013).

Intel demonstrated both Ivy Bridge and Haswell long before they were actually available for retail. This is nothing new and people said the same thing about Ivy Bridge being released in 2011,replacing Sandy Bridge in under a year.

It is doubtful we will see Broadwell entering retail this year,unless you really think Haswell is the shortest tock Intel has done in the last decade,being replaced in under six months. What do you think the OEMs are going to do with all the Ivy Bridge and Haswell laptops that they still have significant inventory off?? The PC market is shrinking,not expanding. Even Ivy Bridge and Haswell were delayed due to OEMs wanted to shift unsold stock of the previous generation products,without having to resort to fire sale pricing.

 

[Arachnotronic]

Intel demonstrated both Ivy Bridge and Haswell long before they were actually available for retail.

It is doubtful we will see Broadwell entering production this year,unless you really think Haswell is the shortess tock Intel has done in the last decade,replace in under six months.

First Broadwell chips will likely be LGA chips for the Xeon E3 space, followed shortly by the Ultrabook parts in August/September. The latter part I can confirm from roadmaps I've seen (I'm at IDF now and some of my peers here are very good at getting these leaked roadmaps). Speculating a bit on the first part, but it makes sense given how Intel does things.

Excavator is 28nm and will be coming in 2015.

 

[ShintaiDK]

Intel demonstrated both Ivy Bridge and Haswell long before they were actually available for retail.

It is doubtful we will see Broadwell entering production this year,unless you really think Haswell is the shortess tock Intel has done in the last decade,replace in under six months.

Broadwell starts production this year. Just like Haswell started production last year.
http://news.softpedia.com/news/Inte...ture-Begins-This-Quarter-Q4-2012-299953.shtml

You might be unfamiliar with it. But besides the time to build up inventory to supply all the OEMs from day 1. It also takes around 3 months from production start till you have the first finished ready to use consumer chip in your hand.

 

[Arachnotronic]

There's no way Apple is releasing a phone in September 2013 that's on TSMC's 20nm. That's totally unprecedented, it'd be like 6 months earlier than expected. And no one else is close to ready with 20nm either.

Even then TSMC's 20nm isn't really comparable to Intel's 22nm since it doesn't have FinFETs. They won't be using a similar process until their so-called 16nm hits.

You're absolutely right. The likely explanation is that Apple's chip is mostly GPU, which can be made much more densely than CPU.

 

[TuxDave]

There are an awful lot of design decision that affect density besides process tech, there could be significant amounts of cache for example, RAM density is waaaay higher than most processor area, it's just too early to know if Apple has really pulled off a coup and managed to get TSMC to deliver 20nm. Sure it's possible, and if it is, there will an absolutely FURIOUS Qualcomm for TSMC to manage but it is just to early to conclude that.

This man speaks the truth.
1) Design methodology affects density
2) Relative proportion of register files vs combination logic
3) Voltage/frequency targets (which can fatten up your wires/devices)

 

[Khato]

Well what about the claim that some have been making that Intel's 22 nm process is not actually 22nm, but 26 nm or something?

Eh, node shrink sizes have never been perfectly accurate. If you use SRAM cell sizes to compute node labels, then using Intel's 130nm process as reference would yield a label of approximately 27.5nm for Intel's 22nm process.

I figure a more meaningful comparison though is that Intel's 22nm process has an SRAM cell size of 0.092um^2 while TSMC has reported an SRAM cell size of 0.081um^2 for their 20nm process... which is technically slightly larger than it should be if using Intel's 22nm size as reference. (0.092*20*20/22/22 = 0.076um^2)

What matters more is that Intel's published transistor performance characteristics have pretty much always been leading TSMC's comparable node by a fair margin. For example, Intel's 32nm process has ~20% higher NMOS and ~30% higher PMOS drive currents than TSMC's 28nm HKMG at the same 100nA/um leakage current and 1V. Which means that Intel could reduce voltage and still obtain comparable drive currents.

 

[USER8000]

First Broadwell chips will likely be LGA chips for the Xeon E3 space, followed shortly by the Ultrabook parts in August/September. The latter part I can confirm from roadmaps I've seen (I'm at IDF now and some of my peers here are very good at getting these leaked roadmaps). Speculating a bit on the first part, but it makes sense given how Intel does things.

Excavator is 28nm and will be coming in 2015.

All roadmaps have shown Broadwell to be retail in 2014,which would make sense because Haswell was only released in June this year.

Intel has previewed SB,IB and Haswell long before retail availability. This is primarily done for investors of all people. All the time people on the internet get over excited and start thinking availability will be then. Look,at the last couple of years Intel has always waited around 14 to 18 months,between ticks and tocks since 2006:

http://en.wikipedia.org/wiki/Intel_Core

It just makes financial sense.

Even AMD showed off working Trinity reference laptops long before the official release date,and the same goes with the Jaguar based ones. Loads of companies do it.

Broadwell starts production this year. Just like Haswell started production last year.
http://news.softpedia.com/news/Inte...ture-Begins-This-Quarter-Q4-2012-299953.shtml

You might be unfamiliar with it. But besides the time to build up inventory to supply all the OEMs from day 1. It also takes around 3 months from production start till you have the first finished ready to use consumer chip in your hand.

He said shipping this year,not production. Big difference.

Moreover,Haswell took at least six months to nine months to enter retail for various reasons,on a proven process node.

Guess what,it is the same for all companies,including the ARM licensees. Just showing prototype silicon off,is no indication of whether you are able to buy products containing them anytime soon.

At least with new generation Atom,products are coming soon,not with Broadwell.

However,old Atom was not released a mere three to six months before and it is on a proven process node too.

 

[blackened23]

AMD is catching up fast and I hope it makes intel release 14nm processor in 2014 itself according to the original plan. Instead of 2015 which was post pond later on!

You have an interesting definition of "fast".

 

[Fjodor2001]

What about it?

Some people claim the moon is made out of blue cheese.

Well is there any truth in it, and in that case are the reasons for that valid for Intel's 14 nm process too in a similar way?

 

[redhotiron2004]

I don't know where you are getting your info, but 14nm CPUs from Intel (Broadwell) are shipping this year (2013).

In 2015 Intel will be shipping CPUs produced on a 10nm manufacturing process.

http://www.guru3d.com/news_story/in...have_pci_express_4_ddr4_and_sata_express.html

I just read some latest developments and intel would be releasing the 14nm for consumers by 2014 only. As far as 10nm is concerned then looking at the way intel had managed these things in the past. I really doubt that they would release 10nm before 2016 for consumers. It would require change in material for that. And they would really want to squeeze out the last bit of profit before jumping to that one. Untill or unless AMD would do something about that.

 

[redhotiron2004]

You have an interesting definition of "fast".

fast in terms of APU'S. AMD value for money proposition are simply unbeatable. Until or unless you are an enthusiast.

 

[Idontcare]

Well is there any truth in it, and in that case are the reasons for that valid for Intel's 14 nm process too in a similar way?

There isn't any truth to it is the bottom line. Someone grabbed onto the fact Intel didn't scale their M1 pitch at 22nm as aggressively as they did for 32nm and decided to try to make a mountain out of a molehill before everyone else published their 22nm pitch design rules...once the other's published theirs the picture got quite a bit different because Intel actually has tighter pitch than the 22nm competition (IBM).

But the real reason why there is nothing to the original argument is the simple fact that nodes have no definition. There is no rule or law or industry standard for defining what constitutes a 22nm node, a 20nm node, etc. It is a label that is used solely at the discretion of whoever wants to use it.

You could, right now, declare to the world you have invented a 7nm node if you wanted. What are the design rules? Could be a breadbox layout for all it matters, there is no definition of what a 7nm is such that your claims would become technically invalid.

It is just silliness really, marketing silliness.