Limit of the 130nm lithography process?

[Albuquerque]

Ok, so something else to play with... Over in the Intel forum, we're having a good discussion on the likely demise of the 0.13micron process, and I thought this would be a better topic for the guri (plural of guru? hehe) to discuss.

Some background:

The previous 180nm process started with the P3 550's and continued all the way through the P3 1.4ghz parts. After the 1.4ghz point, these processors were basically at their upper limit. We already know what happened with the 1.13's ;)

But then the 180nm process was used in the new core design of the P4 Willys, which reached all the way to 2ghz. We already know that core design has a lot to do with maximum clock speed, so obviously the 180nm process was helped by the much deeper pipelines.

Ok, so here we are at 130nm. This process has stretched from the 1.6A end all the way to the current 3.06B. The newest C1 MVID chips are hitting 3.5+ghz on stock voltage, so obviously the process and the core both have some life left...

The question is, where do you think it will stop? I'm of the opinion that Intel could very likely keep the 130nm process as far as the 3.6ghz mark with the new 800mhz FSB parts, even perhaps to the 3.8ghz or even a slight chance of the 4.0ghz mark.

What's your opinion? :)

[doc]

This will not answer your question, but will give you some more food for thinking. Size is not only thing that has changed in the last 10 years, the materials that go into a CPU have changed as well and the deposition of these materials is getting better as well. If one were to shrink a 486 down to current size, it would not run at the speeds we currently have, if it did run that fast it would not do it very long. The oxides and metalizations used in the old 486s was not as good as what is used now. plus there has been changes in the design of CPUs and layouts. All these things put together get you more speed, but size is the easy one to see.

[Da Big Red]

i'm actually doing a bachelor of physics in Nanotechnology at the moment and the theorical limit on photolithography is 90nm or so our tutors have told us

[Albuquerque]

Well, I'm not exactly looking for the limit of the actual photolithography process itself, just the speed aspect of it ;) However, I was under the impression that IBM and Intel both had shown some form of SOI lithography process at the 60nm level... I'll have to go Google it up and see if I can find a related link to that particular topic.

But back to the speed thing...

I hadn't considered the actual materials used, and you are right about the "optical shrunk" 486. Haha, but can you imagine that? :D I think it would be hilarious if they could make a few of those just for giggles -- a 0.13 micron 486DX10/500mhz using our current materials and processes. Slap that puppy on a 486 board with a much lower voltage regulator and watch it fly ;)

Sad thing is, the poor thing probably wouldn't get that fast only because the motherboard would be so god-awfully tuned for it. Beh.

Well anyway, back to the topic at hand. I suppose our current 130nm process could be further enhanced with SOI, copper interconnect and other such technologies which would allow it higher speeds at lower voltages yet again. The only restraint here is that, Intel has no reason to spend such money on a process that's going away.

It seems to me that the current 130nm process is hitting the 3.5ghz mark on air cooling and stock voltage, which means I'm pretty sure that Intel could build P4 3.2 and 3.6ghz chips on their current processes only with minor tweaking to the core design.

[Enigma422]

Actually Albuquerque, SOI and lithography are two different things. SOI is used to reduce the capacitance of the transistors and allows for faster switching times, thus faster chips. SOI is an imbedded layer of silicon dioxide in the silicon substrate. This is usually done before any device processing begins and is done through several methods. One method that comes to mind is growing an epitaxial layer of silicon on top of silicon dioxide. Another method is to grow a layer of silicon dioxide on the silicon wafer and attach another layer of silicon on top of the substrate.

On the other hand lithography is the process of imaging the different layers of a device onto a wafer. This is part of device fabrication.

Now on to how far the 130nm process can go. It all depends on if a company wants to push smaller features or recycle their current process. The 130nm process is far from its limits, but need enhancements to increase speed. One thing currently being researched is strained silicon. By growing a layer of germanium on top of a silicon substrate and then growing another layer of epitaxial silicon on top of that you begin to strain the lattice of the top layer of silicon. This is because the latice constant of germanium is larger than the lattice constant of silicon. The strained silicon lattice allows for higher carrier mobility, thus faster switching, and faster chips. This modification can help in extending the life of the 90nm process.

But then again, Intel in all their wisdom has integrated the 90nm process with strained silicon for their upcoming Prescott core.

Note this is talking about the entire 130nm process as a whole. 130nm lithography process has no limits, it states that the wavelength of the steppers used for lithography is 130nm which is in the DUV range.

Ba Big Red, photolithography is dead, everything is done in DUV and currently research is being done in EUV lithography.

[Paradoxmaker]

Working in a semiconductor fab I get to see a lot of trade magizines and I remember articles on shrinking the die size to the limits of silicon by using X-rays as exposure tools instead of EUVL(extreme ultra violet lithography). Looking at these links it seems some fabs are producing small scale chips for memory in the 13 nanometer range now. Link 1 (http://www.vhti.org/Companies/salxray.htm)
Link 2 (http://www.xraylith.wisc.edu/overview/overview.shtml) Link 3 (http://courses.nus.edu.sg/course/phyweets/Projects99/Atom/xray.html)
According to these articles one of the main drawbacks to this process is the cost of making masks range in the $12,000-$15,000 each for 4" masks. That is quite a high investment. I've been told before the masks used in our fab are in the $150-$200 range each for a 4" wafer mask. We however do not come close to the tolarances that processer makers use. Our photo process has a 1mil range tolarence for mask alignment on the wafer.

[Albuquerque]

Hey, good stuff! Thanks for the info! My brain is now every so slightly larger now ;) I didn't know EUV was getting that close, kinda cool to hear.

[Axeman098]

Intel is aggresively pushing EUV pilot production by 2007-2008 and mass production by 2009. While the current lithographic wavelength is 193nm, the hope is the pare it down to 157nm within two years. By contrast, EUV would plunge the wavelength to 13nm. This could mean processors with 1 Billion or more transistors by the end of the decade.

Okay...I gotta stop now....head is hurting .... too...much... information...must...reboot!

[futura2001]

Nah, spintronics are going to be where its at...
Back the discussion though:
Well, since there are chips that have run at 4.2+ ghz, I would say that if required Intel could make a 4.1-2 ghz processor, but as AMD and Intel are quickly realizing, there is a very small market for that. The chips would be for a very, very small niche market, and then you have to account for cooling. Which would be somewhere along the lines of a Geforce FX Ultra cooler.
So, maybe 4 at tops, but probably closer to 3.6-8

[Axeman098]

Keep in mind that a GOOD while back, Tom's Hardware.com came out with an article about a 3.6Ghz P4 Prototype that Intel brought to the lab. They booted the chip on a P4T533 with a modified BIOS. This was LONG before the 3.0Ghz was even released. I wouldn't be suprised if the lucky few Intel Alpha geeks are already playing with 4+Ghz engineering samples somewhere laughing at the dweebs that boast about their "smokin" 3.06Ghz rigs. They're probably playin with stuff RIGHT NOW that you and I can only dream about. Research samples are almost surely a considerable margin ahead of current market "state-of-the-art". And I bet somewhere someone is playing with a "Tejas" sample as we read this....lucky suckers!!!

[Finken]

Originally posted by Axeman098
Keep in mind that a GOOD while back, Tom's Hardware.com came out with an article about a 3.6Ghz P4 Prototype that Intel brought to the lab. They booted the chip on a P4T533 with a modified BIOS. This was LONG before the 3.0Ghz was even released. I wouldn't be suprised if the lucky few Intel Alpha geeks are already playing with 4+Ghz engineering samples somewhere laughing at the dweebs that boast about their "smokin" 3.06Ghz rigs. They're probably playin with stuff RIGHT NOW that you and I can only dream about. Research samples are almost surely a considerable margin ahead of current market "state-of-the-art". And I bet somewhere someone is playing with a "Tejas" sample as we read this....lucky suckers!!!

Haha, makes you wanna eat your pants, doesn't it? :)

I remember some time ago in a thread called something like: My plan to hit 4GHz on air... it's still active btw
I posted that it should be possible before prescott.

I would bet that they can get 3.6-8 out of the 130nm.
Maybe actually closer to 3.8 considering that there are several cases of 3.5 today...

Btw, you can combine strained silicon and SOI right?
90nm with both of those techs should open up for some bad clockspeeds :eek:

[BigRed]

Originally posted by Albuquerque
The previous 180nm process started with the P3 550's and continued all the way through the P3 1.4ghz parts. After the 1.4ghz point, these processors were basically at their upper limit. We already know what happened with the 1.13's ;)

But then the 180nm process was used in the new core design of the P4 Willys, which reached all the way to 2ghz. We already know that core design has a lot to do with maximum clock speed, so obviously the 180nm process was helped by the much deeper pipelines.

180nm stopped with the 1.13 p3 (failure) and there were some 1.1ghz celerons that were made on that process.
But the 1.1, 1.2, 1.26 (512k cache), 1.3, 1.4, and 1.5 (yes a few were made befroe intel axed the p3) p3's were on a 130nm process. Along with the 1.0a, 1.1a, 1.2, 1.3, and 1.4ghz celerons, they were also on the 130nm process.

Opps just realized the thread is a month old, im bored though :p