Intel Completes Next Generation, 32nm Process Development Phase

Toggle sidebar Toggle sidebar
I

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Intel Completes Next Generation, 32nm Process Development Phase

The Intel 32nm paper and presentation describe a logic technology that incorporates second-generation high-k + metal gate technology, 193nm immersion lithography for critical patterning layers and enhanced transistor strain techniques. These features enhance the performance and energy efficiency of Intel processors. Intel's manufacturing process has the highest transistor performance and the highest transistor density of any reported 32nm technology in the industry.

http://www.intel.com/pressroom..._releasepri_20081209_r
Click to expand...

I think most people (who are involved in this industry) expected a deluge of 32nm articles and informations from Intel at the upcoming IEDM.

What stands out to me is that Intel PR is actually pre-announcing the impending announcement. Seems like a shift in their PR handling of process technology announcements. Or maybe I just never paid as close attention in the past as I am now.

At any rate it will be interesting to see how much of a performance scaling Intel was able to coax out of their HK/MG planar CMOS. (e.g. short-channel effects in addition to scaling Gox with these HK integration approaches)

Will it still be disposable (replacement) gate integration? I'm really curious to know.
 
D

darkrisen2003

Senior member
Sep 13, 2004
382
0
76
Hello IDC. Hopefully I can ask this without offending you in any way but about the 32nm proccessors they are working on. What are the chances that they will work on the existing core i7 motherboards being produced? I have looked at some of the articles about the proccessors and havent seen the question brought up. I know you are very articulate and have a way with words but I would like it spelled out if you happen to know.

I remember when 775 first changed to the core 2 and most people were upset that they had to change to a new MB which is why the question is being asked. I planned on building a new machine sometime around febuary and dont want to get burned again like I did in the begining of 06. I have hated my build since I first made it and havent been able to save any cash to do anything about it till now. The motherboard I got and the moniter and video card have all 3 been such dissapointments to me.
 
I

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Originally posted by: darkrisen2003
Hello IDC. Hopefully I can ask this without offending you in any way but about the 32nm proccessors they are working on. What are the chances that they will work on the existing core i7 motherboards being produced? I have looked at some of the articles about the proccessors and havent seen the question brought up. I know you are very articulate and have a way with words but I would like it spelled out if you happen to know.
Click to expand...

Ha ha, hey there are no sacred cows here on these forums :)

Yes hopefully I can answer in a way that is understandable ;)

Let me be upfront and say I don't know anything specific whatsoever about Intel's 32nm processor sockets. I can give you my speculation based on experience though.

Originally posted by: darkrisen2003
I remember when 775 first changed to the core 2 and most people were upset that they had to change to a new MB which is why the question is being asked. I planned on building a new machine sometime around febuary and dont want to get burned again like I did in the begining of 06. I have hated my build since I first made it and havent been able to save any cash to do anything about it till now. The motherboard I got and the moniter and video card have all 3 been such dissapointments to me.

-------------------------
Antec 900
Asus Striker Extreme
Core 2D 6600
Thermalright Ultra 120 Extreme With 120 mm Antec Blue Led Fan
2 gigs Corsair Value Ram DDR2 667
WD 250 Gig Hard Drive x 2
Lite on Sata Dvd burner with Lightscribe
Geforce 8600 GTS 256 Meg ( Sucks )
Sceptre 22 inch HD Widescreen Monitor
Click to expand...

I see you have the original ROG striker extreme. I too had that board and all I can say is no surprise you have buyer's remorse. At the time there really weren't many choices, but man-o-man did that board suck compared to today's crop of P35/X38/P45/X48 LGA775 boards.

The problem with the early 775 boards was that the signal integrity challenges posed by the quadcore chips was not comprehended all that well by the board engineers when the boards were initially being developed.

The 680i-based mobo's did not leave enough margin in the trace layout to deal with the signal-quality when higher FSB speeds were required for 45nm yorkfield.

So if we assume that Intel will make Westmere (32nm nehalem) socket compatible with current LGA1366 implementation then the question we ask ourselves is will westmere pose some manner of signaling and/or voltage regulation challenge that will invalidate current designs in a way that the Kentsfield->Yorkfield transition did?

My answer to this question is no, it won't. We should have known that yorkfield was going to be problematic to kentsfield LGA775 era mobo's because those mobo's overclocked like ass. My striker extreme could barely manage a stable FSB increase to 300MHz from the stock of 266MHz. It was obvious (in hindsight) that there was simply no-way that board was going to run a yorkie stable at 333MHz FSB.

So I say look to overclocking limitations in the current crop of i7 mobo's to determine the liklihood of an incompatibility with future 32nm processors. We see the baseclock on some of these i7 920 overclocks reaching 200MHz, up from the 133MHz stock clock.

So long as Intel keeps westmere baseclock to something under 200MHz then we have no reason at this point in time to expect westmere to not function in current LGA1366 mobos.

But again to reiterate I have no specific info on this, just going by logic. Lots of variables can cause the chip itself to not function in current mobo's. The Vcore may be too low for the voltage regulators to function (although I doubt this will happen), or current draw may be too high for mobo's to survive (as happened with the 140W TDP AMD 9950's on some older AM2 mobo's).

Personally, having lived thru the first round of quadcore CPU's (B3 stepping QX6700 here) and first round of 680i mobo's (striker extreme) I will never ever ever buy first iteration computer hardware ever again. Wait till the second retail stepping of i7 hits the stores before buying i7. Wait for second gen LGA1366 mobo's before buying a mobo. Etc.

Save yourself the headache of buying someone else's prototype engineering effort, wait for the product that the engineers wanted to build to finally make it to the store shelves without marketing pushing to get something out sooner despite the bugs.
 
I

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Update: EETimes has an article on the announcement and includes some of the 32nm design rules (presumably from the IEDM abstract?):

Intel to extend high-k lead at IEDM

The transistors feature dual band-edge workfunction metal gates and high-k gate dielectrics with an equivalent oxide thickness (EOT) of 9-nm or 9 angstroms. In comparison, the company's 45-nm high-k designs have an EOT of 10-nm. The 32-nm version enables Intel to reduce transistor variability,'' Bohr said.

At 32-nm, Intel's transistor gate pitch is 112.5-nm. Intel's 32-nm logic technology provides about 70 percent linear feature size scaling and 50 percent area scaling, as compared to the company's 45-nm process. In addition, the process enables the highest drive currents reported to date for 32-nm technology.

http://www.eetimes.com/news/se...l;?articleID=212300580
Click to expand...

Planar CMOS is recycled as expected (2-node minimum is industry standard for re-using existing technology on future nodes) but with the standard smaller dimensions and 4th gen strain techniques to get more performance out of the smaller xtors.

Meanwhile, the company is also working on its 22-nm process. The technology will make use of 193-nm immersion scanners with either double-patterning or computational lithography techniques, he said. For 22-nm, EUV ''probably won't be ready,'' he said.
Click to expand...

EUV delayed another node, absolutely no one in the industry is surprised by this.
 
D

darkrisen2003

Senior member
Sep 13, 2004
382
0
76
I appreciate you understanding what I meant. I was afraid you would take it the wrong way. That does sound about right though which sucks for me. I really want to replace this machine as soon as I have the money. Well since those proccessors wont be out untill closer to the end of 09 im not sure what I should do. I was thinking about buying the pieces one at a time as the money frees up. Starting with an antec 1200 since I liked the 900 so much I figured the 1200 would be a nice larger version of what I have now. Keep hearing that there is about to be a new 295 gtx nvidia card soon so was thinking about jumping on that provided the price isnt sky high. This time going for a nicer moniter that has an HD port that doesnt burn out and has hdmi ( thinking about 24-28 inch for the higher res. ) Asus burned me with the striker and if you check there forums they have been dropping the ball on alot of boards within the past year or 2 so I was thinking about switching to EVGA since everyone raves about there support and there video cards are great anyways so why not. If the ram is cheap enough when I have the cash I would like to get whatever is max for the board since I will be using windows vista ultimate 64 bit and im a gamer so ram helps no matter what some try to claim. IDK would you have any suggestions? Everyone has there brands and I completely understand that but im open for suggestions. My only brands I stick with are intel for CPU, corsair for ram, and evga for video card. Other than that im all ears.
 
M

Martimus

Diamond Member
Apr 24, 2007
4,488
153
106
Originally posted by: Idontcare
So long as Intel keeps westmere baseclock to something under 200MHz then we have no reason at this point in time to expect westmere to not function in current LGA1366 mobos.
Click to expand...

Based on How AMD has handled their memory clock without the FSB, I doubt Intel will change theirs from 133MHz. It has already been shown that it isn't needed in their competitors chip. Of course Intel may do it anyway to sell more chipsets, unless AMD pulls a miracle and starts stealing marketshare before Westmere.

I know that I was excited about the Phenom II results, but seeing initial leaked benches it looks more like their top chip will be ~ equal to a Q9450 in performance. I was hoping for better than a Q9550. If that turns out true I will likely build my new system from a Q9550 in the spring, unless I decide to buy a 940 BE and overclock it. Luckily it looks like Westmere will be out the following spring, so I have that to look forward to.

I do have a question. This press release doesn't seem to actually be anouncing anything, other than what the 32nm process will entail. Do they usually anounce that they are switching production to a new process, and if so is this supposed to be that announcement?
 
I

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Originally posted by: Martimus
I do have a question. This press release doesn't seem to actually be anouncing anything, other than what the 32nm process will entail. Do they usually anounce that they are switching production to a new process, and if so is this supposed to be that announcement?
Click to expand...

It is atypical PR for Intel. It is typical PR for the foundries (TSMC, UMC, IBM, etc).

So for some strategic reason Intel has shifted their PR approach to publicizing process technology milestones.

What they are announcing is that basically the major integration features are complete, toolsets are locked down, etc, and now they embark on the remaining milestones (yield and reliability) on they way towards product release next fall.

Yield and reliability improvements involve tweaks and changes to integration too, so it's not like they could start production today, but the biggies are complete such as hitting their spice models and xtor parametrics as internally committed a couple years ago. (i.e. silicon now functions to the models committed for design teams when 32nm groundwork was first undertaken in late 2005)

They may be doing this in anticipation of the likelihood that their primary competitor (AMD) will shift more towards this method of publicizing process technology advancements by way of TFC being a foundry and they TFC needing to do it in order to compete with TSMC and UMC mind-share.

For us end-users it doesn't mean much, just confirmation that Intel is on-track and encountering no major roadblocks in their 32nm timeline. For shareholders I'm sure it is intended to bring confidence to margin erosion concerns in a recessionary environment. For corporate customers no doubt it speaks to building more confidence in Intel's 32nm roadmap and increases the negative sentiment towards AMD's (x86) and IBM's (Power6) 32nm technology viability. Sort of a "get on our wagon now or be left behind when your competition decides to" mindshare mentality.
 
I

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Sanjay Natarajan, Intel?s 32 nm program manager, will present a late paper on Intel?s 32 nm process on Wednesday at IEDM, which is being held this year in San Francisco. The gate pitch was reduced to 112.5 nm, which compares with 160 nm for the previous 45 nm generation.

?That is pretty close to 70% linear scaling,? said Senior Fellow Mark Bohr, who is in charge of the 22 nm ?pathfinding? research program at Intel?s Hillsboro, Ore.-based Logic Technology Development Group. ?We achieve 50% area scaling, keeping us on a two-year cadence. The drive current increases 14% compared with 45 nm, which when combined with gate capacitance reduction gives us a 22% performance improvement.?

http://www.semiconductor.net/a...436.html?desc=topstory
Click to expand...

This link even has some Idsat vs. Ioff graph data for both nmos and pmos to support the 14% Idrive increase statement. Operating voltage is 1V. Nice.
 
P

palladium

Senior member
Dec 24, 2007
539
2
81
Personally, having lived thru the first round of quadcore CPU's (B3 stepping QX6700 here) and first round of 680i mobo's (striker extreme) I will never ever ever buy first iteration computer hardware ever again. Wait till the second retail stepping of i7 hits the stores before buying i7. Wait for second gen LGA1366 mobo's before buying a mobo. Etc.
Click to expand...

That is so true, I wish I could do that but I think stretching my P4 to H2 2009 would be too demanding.

The company is on track for production readiness of this future generation using even more energy-efficient, denser and higher performing transistors in the fourth quarter of 2009.
Click to expand...

So Larrabee might be 32nm afterall....
 
I

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Originally posted by: palladium
So Larrabee might be 32nm afterall....
Click to expand...

Well I have no doubt that a 32nm iteration of Larrabee will eventually be created and sold...but the initial tape-out and samples of Larrabee are 45nm.

The transition from 45nm to 32nm for Larrabee will entirely depend on the performance, cost, and power consumption results of the 45nm iteration. Merced was handled no differently.
 
I

Idontcare

Elite Member
Oct 10, 1999
21,110
59
91
Originally posted by: uclaLabrat
How are nodes decided? Why 22nm instead of 20nm? or 16nm?
Click to expand...

They are labels meant to denote a succession that things in one node are generally smaller than the similiar things in a prior (larger number) technology node.

Some feature's dimensions are related to the label, but it is a loosely held rule and rarely applies in real life.

The ITRS set guidelines for expected dimensions of a given node (such as metal pitch, gate pitch, etc) but even they have given up on having the label of "node" mean much of anything numerical anymore.

The labels are intended to reflect the ~0.7x linear scaling that many features receive in going from one node to the next.

Most features (metal pitch, gate pitch, contact pitch, etc) in an IC will shrink by ~0.7x in going from 45nm to 32nm "node".

Likewise they will shrink an additional 0.7x in going from 32nm to 22nm "node".

22/32 ~= 0.7

There really is no value added by making the labels match any more closely to the physical dimension within the chip. It serves its purpose already.
 
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom