Xbit Labs: AMD-> Improvements of Next-Generation Process Technologies Start to Wane.

[cbn]

http://www.xbitlabs.com/news/cpu/di...ation_Process_Technologies_Start_to_Wane.html

AMD : Improvements of Next-Generation Process Technologies Start to Wane.

AMD Talks 14nm, 20nm: We Have to Approach New Processes Wisely
[12/16/2011 10:36 AM]
by Anton Shilov

The chief executive officer of Advanced Micro Devices said at a conference that advantages brought by the next-generation process technologies will get less noticeable than previously. As a result, AMD will need to transit to new technologies more wisely than before and attempt to innovate using proven fabrication processes.

"Now, let's talk about 20nm and 14nm. I think that we really flying hard in the path of subatomic environments. The price advantages as we move down nodes are starting to wane. The ability to [quickly improve] yields and ramp up our products (which have fixed amount of time) is under exceptional pressure. It costs huge amounts of money. I think we have to be strategic and think about how quickly we go down the node," said Rory Read, chief executive officer of AMD, during IT Supply Chain conference organized by Raymond James.

Traditionally, both vertically integrated makers of semiconductors as well as contract makers of semiconductors, introduce new process technologies every 18 to 24 months. In the recent years the cadence changed a bit since development of new manufacturing processes and building new fabs became extremely expensive, but Intel Corp. keeps introducing new fabrication processes every two years and new product families every year proving the financial viability of Moore's law. The world's biggest chipmaker believes that the new process technologies enable it to integrate more functionality into chips while keeping their prices relatively flat. However, Intel is among a few companies who produce so large amounts of chips that it can cover development costs.

AMD believes that it should reconsider its typical strategies and tactics. In particular, it needs to innovate within existing and proven process technologies and not wait till better manufacturing processors become available and mature.

"We have some of the best products now! Look at the Brazos chip, over 20 million sold, we think we have taken share in the emerging markets. [...] That's the 40nm process. It is small. It is a beautiful product, it has got great margin. That's the future! The ability to deliver, to execute," stressed Mr. Read.

The new chief executive of AMD implied that high process technology development costs eventually translates into high product prices. In case of economic downturn, it will be very hard to remain profitable if demand collapses. On the other hand, if transition to newer fabrication process takes a longer time, it will become easier from financial point of view.

"Huge set of investments [into new process technologies and their ramp up] is almost like a chain around your neck. You've got to sell gigantic price processors particularly to commercial segment when commercial is nearing the end of the corporate refresh. You've got consumerization spreading around the planet. Why are you going down that path? Sure, there are advantages in moving down the node [...], there is a lot of profit and opportunity for us in terms of our business [with the currently available process technologies]. Why moving down the node too fast?" asked Mr. Read rhetorically.

Rory Read believes that the market of semiconductors will need to change drastically from what it is today. Even though AMD will inevitably adopt both 20nm and 14nm fabrication processes in the coming years, it will change its adoption approach and expects companies like Globalfoundries to change their pricing models so to make transitions to thinner processes viable for fabless designer of chips.

"Just go look at the cost of wafers as you move down those technologies, they are not going down, they are going up! If the yield does not go up, how do you get your return? You have to charge bigger prices. We will get there, we will move down [but ultimately there got to be different pricing model]," said the chief executive officer of AMD.

 

[cbn]

Some more information from this paper (PDF) <----Thanks to the Anandtech member who originally posted this source of information in another thread.

V. SUMMARY AND IMPLICATIONS
The 30-year-long trend in microelectronics has been to increase
both speed and density by scaling of device components
(e.g., CMOS switch). However, this trend will end as
we approach the energy barrier due to limits of heat removal
capacity. For nanoelectronics, this result implies that an increase
in device density will require a sacrifice, due to power
consideration, in operational speed, and vice versa. Thus, it
appears that we are entering a regime where tradeoffs are required
between speed and density, quite in contrast to the
traditional simultaneous benefits in speed and density from
conventional scaling.

 

[Lonbjerg]

*Intel's 3D Tri-Gate transistors chuckles in the corner*

The way I read it is that they declare that Intel has run to far off to be catched and AMD's options for process technologies is less than steallar compare to Intel's massive FAB and process advantage...

 

[Arkaign]

I don't think it can adequately be expressed how screwed AMD was over the decisions to invest heavily in ATI at the top of the market and lose all of their fabs. It's been a steady decline for them ever since, and it makes sense :

Every computer needs a CPU, and competitive ones command good ASPs.
Not every computer needs a dedicated GPU (90&#37;+ probably have integrated when you look at the entire market)

Intel could have bought ATI or Nvidia with the pocket change in their petty cash drawer, but didn't. What did they know that AMD didn't?

 

[tommo123]

how about a trade? AMD sell ATi to intel in exchange for being allowed to use intels fabs to make their CPUs?

 

[Lonbjerg]

how about a trade? AMD sell ATi to intel in exchange for being allowed to use intels fabs to make their CPUs?

1. Intel have no interest in "ATi".
2. Intel has no interest in letting AMD use their FAB's.

 

[videogames101]

This is just another way for AMD to say they can't compete.

 

[Arachnotronic]

This is good. AMD doesn't really have a chance against Intel for the very cutting edge (who does?), so it's good that this CEO wants to play to their strengths rather than pursue the cutting edge process tech.

This Rory Read actually seems like a smart, level headed guy who knows what he's working with and ultimately knows how he's going to maximize profit with the tools at han. It's refreshing to see that after guys like Hector Ruiz (a slimebag/complete idiot) and Dirk Meyer.

 

[know of fence]

AMD believes that it should reconsider its typical strategies and tactics. In particular, it needs to innovate within existing and proven process technologies and not wait till better manufacturing processors become available and mature.

That basically means a strategy like Tick-Tock, maybe even Tick-Tock-Tock. A slowdown (and total limit) of miniaturization is inevitable and we've seen the benefits of iterative strategies already (Nehalem vs. Sandy Bridge).

 

[cotak13]

Do you guys go out and buy via processors in any large numbers? No. Why?

Read is saying "we can't afford to keep up with intel, so we will now try to sell you stuff that's slower cheaper and hope you guys buy in".

We just got a brazo laptop for someone as a gift and I have to say its pretty good for what it is. But only because we found a great refurb deal. At full price I would have wonder why we didnt get a faster laptop. It's truly amazing how brazo hits the good enough performance level. But at market price it doesn't make so much sense.

And the real issue with the APU strategy is where are your proprietary controls? What's stopping intel from eating your lunch with a faster chip of similar configuration? People would cite the long history of bad graphics from intel. However I would remind people that intel hasn't been too bad for SB. And IB should get better. At what point is graphics a null factor?

Of course amd would like their gpgpu technology be adopted and make that their stick to best intel with. But the issue there is gpgpu for gp compute is not gaining much traction. Whether it will prove to be anything of importance for daily computing needs its still a big question. Generally I think not because it's not a cure all for increasing performance for any compute problem. However, it could be surprising and make our media world much richer.

Overall my take is that Read is setting expectation on what amd will do. I wouldn't read, lol, too much into whether it's a good stetegy or not.

Btw sorry if there are iPad typos.

 

[ncalipari]

expect huge price increase from Intel side.


I think this is another milestone in the death of x86.

 

[soccerballtux]

now now, AMD, just because Bulldozer can't hit 3416967080591056 ghz does not mean process improvements with new processes are disappearing.....

 

[aphelion02]

Loser talk.

 

[Fox5]

Do you guys go out and buy via processors in any large numbers? No. Why?

Read is saying "we can't afford to keep up with intel, so we will now try to sell you stuff that's slower cheaper and hope you guys buy in".

We just got a brazo laptop for someone as a gift and I have to say its pretty good for what it is. But only because we found a great refurb deal. At full price I would have wonder why we didnt get a faster laptop. It's truly amazing how brazo hits the good enough performance level. But at market price it doesn't make so much sense.

And the real issue with the APU strategy is where are your proprietary controls? What's stopping intel from eating your lunch with a faster chip of similar configuration? People would cite the long history of bad graphics from intel. However I would remind people that intel hasn't been too bad for SB. And IB should get better. At what point is graphics a null factor?

Of course amd would like their gpgpu technology be adopted and make that their stick to best intel with. But the issue there is gpgpu for gp compute is not gaining much traction. Whether it will prove to be anything of importance for daily computing needs its still a big question. Generally I think not because it's not a cure all for increasing performance for any compute problem. However, it could be surprising and make our media world much richer.

Overall my take is that Read is setting expectation on what amd will do. I wouldn't read, lol, too much into whether it's a good stetegy or not.

Btw sorry if there are iPad typos.

Right now, AMD has a price advantage with Brazos over core i3, but it's probably only in the tens of dollars, not really a concern for the first world. Brazos is about half the die size of the core i3-2100...and about 1/5th the performance. And intel only needs another die shrink or two advantage on AMD in order to close that gap.

At the high end, Bulldozer turned out much bigger than it should have due to the 32nm shrink not shrinking very much at all. GF 45nm was about equal to Intel 65nm, now their 32nm is barely an improvement. At the moment, AMD is less screwed on the low end than the high end due to Atom sucking, and that probably won't change because intel is positioning atom for smart phones and brazos is aiming for cheap computers and ultraportables. But once Intel hits 22nm or below and AMD is still stuck at 40nm TSMC or 32nm GF, intel's ulv parts can start competing with AMD on price.

 

[pelov]

You guys do know he's right... right? He's not talking about 22nm, he's talking below Intel's 14nm. Though tri-gate helps, it doesn't solve the issue: you can't shrink forever, Physics doesn't allow it. More importantly, the reliance on shrinking in order to increase your performance gains from generation to generation is also misguided. There will inevitably be a roadblock in a few years time, so might as well start planning ahead of time.

This has little-to-nothing to do with Intel. In fact, I can guarantee you that the management at Intel is thinking the same thing. Fab/architecture advantage aside, they too will hit the same roadblock, even with nifty little tri-gate. Unless quantum computing can make massive strides in the next few years, Moore's law will give way to the laws of physics.

 

[Lonbjerg]

You guys do know he's right... right? He's not talking about 22nm, he's talking below Intel's 14nm. Though tri-gate helps, it doesn't solve the issue: you can't shrink forever, Physics doesn't allow it. More importantly, the reliance on shrinking in order to increase your performance gains from generation to generation is also misguided. There will inevitably be a roadblock in a few years time, so might as well start planning ahead of time.

This has little-to-nothing to do with Intel. In fact, I can guarantee you that the management at Intel is thinking the same thing. Fab/architecture advantage aside, they too will hit the same roadblock, even with nifty little tri-gate. Unless quantum computing can make massive strides in the next few years, Moore's law will give way to the laws of physics.

You are aware of fullnodes and half nodes right?

Fullnodes:
90 nm
65 nm
45 nm
32 nm
22 nm
16 nm
11 nm

Halfnodes:
80 nm
55 nm
40 nm
28 nm
20 nm
14 nm
10 nm

20nm is a halfnode of 22nm...so your answer is wrong as he states (it's in the freaking OP):

"Now, let's talk about 20nm and 14nm."

 

[jsedlak]

You are aware of fullnodes and half nodes right?

Fullnodes:
90 nm
65 nm
45 nm
32 nm
22 nm
16 nm
11 nm

Halfnodes:
80 nm
55 nm
40 nm
28 nm
20 nm
14 nm
10 nm

20nm is a halfnode of 22nm...so your answer is wrong as he states (it's in the freaking OP):

What?

 

[pelov]

Yes and I can't wait for my new 7xxx GPU

<10nm is very near. Intel is already fiddling with 14nm. You interpreted his statement as "what happens next year" whereas his intention was "folks, we have to plan ahead." 22nm and 14nm will be more costly for AMD than Intel so he's shifting strategy now rather than later (when you'll be forced to anyway cause of the whole physics thing).

 

[Lonbjerg]

What?

http://en.wikipedia.org/wiki/Die_shrink#Half-shrink

 

[cbn]

Right now, AMD has a price advantage with Brazos over core i3, but it's probably only in the tens of dollars, not really a concern for the first world.

http://www.investorvillage.com/mbthread.asp?mb=476&tid=10759150&showall=1

Yep, the price is rather close according to this link (originally posted by another Anandtech member).

Ontario 2C Mobile is $9 less than Intel's Value Mobile<----Apparently this is Intel's tier above atom.

Of course, for now AMD can compete with Intel's "value mobile" on battery life but what happens if the following prediction you made comes true:

But once Intel hits 22nm or below and AMD is still stuck at 40nm TSMC or 32nm GF, intel's ulv parts can start competing with AMD on price.

How will the device makers "product differentiation scheme" factor in to help AMD in that scenario? If the resulting product differentiation is forced in the direction of cheaper batteries I think even the "battery life" metric gap could be closed and AMD's Brazos line-up risks going the way of VIA Nano.

 

[paperfist]

I don't think it can adequately be expressed how screwed AMD was over the decisions to invest heavily in ATI at the top of the market and lose all of their fabs. It's been a steady decline for them ever since, and it makes sense :

Every computer needs a CPU, and competitive ones command good ASPs.
Not every computer needs a dedicated GPU (90%+ probably have integrated when you look at the entire market)

Intel could have bought ATI or Nvidia with the pocket change in their petty cash drawer, but didn't. What did they know that AMD didn't?

Intel can buy anyone, but will the government let them?

Apple, Mobile, etc could buy anyone. That doesn't mean they know something someone else doesn't...

 

[Borealis7]

time for AMD to bet All-In and go for Graphene+nanowire Interconnects!
Its crazy, but it might just work.

 

[Arachnotronic]

time for AMD to bet All-In and go for Graphene+nanowire Interconnects!
Its crazy, but it might just work.

Yeah, they've got tons of R&D money and fabs to do that.

 

[Borealis7]

did i mention it was crazy?

 

[cbn]

expect huge price increase from Intel side.

I think it will depend on how much Intel can increase performance on 14nm and 10nm. (See post #2).

(Based on the very little I know) I can still see future nodes lowering power consumption by at least 25&#37; per node jump, but I am not so sure single threaded performance will continue to get the same bumps (they got in the past) without heat density workarounds by Intel?

I think this is another milestone in the death of x86.

In what way to do mean this?

By ARM competing with AMD?

ARM still has plenty of "heat density" room to scale up. So even if ARM's node progression also slows down they can still move up the food chain by designing larger CPU cores on the same node (provided a growth pathway beyond smartphones is available to them.)

Therefore we just have to wonder what plan Intel and AMD have in mind?