backtop

Print E-mail del.icio.us 25 comment(s) - last by masher2.. on Dec 9 at 12:48 PM
45nm has yet to be upon us and the major players are already looking

With the main buzz about the next semiconductor process shift being to 45nm, Intel and TSMC are already gearing up the development of the 32nm node, reports DigiTimes. The article quotes Intel's director of technology strategy Paolo A. Gargini as saying that its development of the 32nm process is “in good shape.”

 

Gargini also said that Intel co-founder Gordon E. Moore’s Law is expected to hold true for the next 10 to 15 years. Immediate challenges to the Law, however, are expected to crop up as manufacturing moves towards 22nm. International Technology Roadmap for Semiconductors predicts that chips built on the 22nm process may not appear until 2015.

 

Furthermore, TSMC is supposedly running tests with the 32nm process, and CEO Rick Tsai revealed that the foundry has a team devoted specifically on developing 32nm technology.

 

To make transitions to more advanced processes possible, semiconductor companies have relied on new technologies such as three-dimensional packaging and multi-gate finFETs. Intel has also hinted it will start using tri-gate transistors on its 32 or 22nm products.

 

Looking at the immediate future, both Intel and TSMC confirmed that the 45nm technology it is developing is mature and that volume production will begin in the second half of 2007.

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
*groan* Moore's "law"
By MonkeyPaw on 12/6/06, Rating: 0
RE: *groan* Moore's "law"
By wien on 12/6/2006 8:00:18 PM , Rating: 2
Moore never said anything about processing power when he made his prediction. It was about chip "complexity", which is usually understood as transistor count per chip.


RE: *groan* Moore's "law"
By slashbinslashbash on 12/6/2006 8:07:28 PM , Rating: 4
Moore's Law (as originally stated by Moore, although he did not call "Moore's Law") has ALWAYS been about the number of transistors you can fit on a chip, given a certain monetary budget.

http://en.wikipedia.org/wiki/Moore%27s_law

That's gotten translated by journalists and other laymen into "doubling performance" and not doubling transistors. "Doubling performance every 18 months" has ALWAYS been a misinterpretation of the idea.

For highly-parallel applications like GPU's, doubling the number of transistors can indeed come very close to doubling the performance. Although we are increasingly hitting a wall in terms of what desktop processors can accomplish with today's software, today's software will be replaced eventually.


RE: *groan* Moore's "law"
By DallasTexas on 12/6/06, Rating: -1
RE: *groan* Moore's "law"
By rushnrockt on 12/6/2006 9:19:33 PM , Rating: 4
Seems like his 18th will be 10 years before yours.


RE: *groan* Moore's "law"
By MonkeyPaw on 12/6/2006 9:46:29 PM , Rating: 3
lol, did I kick Moore to the curb in my statement above? I don't think so. I was venting my disgust of "Moore's law." Moore himself never called his own statement a "law" but more of a prediction or an indication of the future. I don't doubt that the co-founder of Intel is a smart guy, but seriously, what is so magical about doubling transistors? Did you know that when Intel went from Pentium D to Core 2, the total transistor count went down (376m to 291m)? My point? Transistor count is not a reflection of a CPU's performance (as the Athlon X2 also has fewer transistors than Pentium D and outperforms it). In the most recent Intel desktop transition (Pentium D to Core 2 Duo), a decrease in transistor count actually resulted in a substantial increase in performance. Why should we care if, when, or how often the transistor count doubles, when other innovations (new instruction sets, improved branch predictors, virtualization, etc) are far more significant? Moore's law tells us nothing, as a doubling of transistors could either mean the next Pentium D, or the next Kentsfield? You can barely compare the 2, IMO.


RE: *groan* Moore's "law"
By masher2 (blog) on 12/6/2006 11:53:41 PM , Rating: 2
> "but seriously, what is so magical about doubling transistors?..."

Moore's original statement was simply an observation of basic geometry. Since transistor counts are a function of the AREA of a chip, a halving of the feature size results not in a doubling of the features, but a quadrupling. Therefore, if the semiconductor industry moves to new processing nodes in linear time, transistor counts will grow geometrically.

"Moore's Law" is heavily used by Intel as a marketing tool. However, its flat out wrong. Why? Because we don't move to new processing nodes in linear time. Moore originally preducted a doubling every 12 months...as new nodes were coming about that fast then. It was then scaled back to 18 months, then to two years, and now, for the next two nodes, appears will be more like 3 years.

Sure the node progression appears linear over the short term. But thats just a basic physics...any process appears linear if you consider a short enough time interval. In reality, litho nodes will continue to be spaced every-further apart. And so, while transistor counts will still keep doubling...they'll do so slower and slower.


RE: *groan* Moore's "law"
By DallasTexas on 12/7/06, Rating: -1
RE: *groan* Moore's "law"
By iodinegalaxy on 12/9/2006 3:42:30 AM , Rating: 2
Even though transistor numbers downscaled from Pentium D to Core 2 Duo, a die area also decreased, thus raising up density.


RE: *groan* Moore's "law"
By masher2 (blog) on 12/9/2006 12:48:53 PM , Rating: 1
Device density is primarily determined by the process node. Two cpus generated on a 65nm process are going to have roughly the same transistor density, regardless of die size or physical counts. And in fact, Moore's famous law is really just a statement about process nodes, as mentioned elsewhere in this thread.


How long until...
By Le Québécois on 12/6/2006 6:22:53 PM , Rating: 2
They won't be able to reduce the size anymore?

Not that I have anythings against "pushing the envelope" when I come to technology, especially when it come to electronic since it's my field of study but I think that at some point they won't have any choice to completly change from electronic to photonic computing or even beyond with quantum optics.

But how long will that take?

It's an actual question so feel free to answer it.




RE: How long until...
By stromgald on 12/6/2006 6:29:26 PM , Rating: 1
I think they're already developing photonic computing . . . its just not ready yet, and will probably still be clunky and expensive when we hit the wall with electronic computing. From the article's estimates, 22nm will come around 2015. That's pretty far off considering the short time to go from 65nm to 45nm. I'm thinking the reason for that is because we're starting to hit a wall in developing smaller processes. One level lower than 22nm might be as far as its feasible to go.


RE: How long until...
By Le Québécois on 12/6/2006 8:43:35 PM , Rating: 2
quote:
I think they're already developing photonic computing . . .


Oh they are doing that and that scare me because this is NOT my field of study and I don't want to become obsolete :'( .

Now seriously. This is big and when this time comes, computer and electronic will become a whole mess for the time needed to complete the changes.

But until then, this is a pretty good news.


RE: How long until...
By theapparition on 12/7/2006 7:22:53 AM , Rating: 2
Just FYI,
There is a theoretical limit to the speed of any processor. For electrons, it is the die size divided by the speed of electrons in silicon. For photonic computing, it is the die divided by the speed of light. These numbers are completely theoretical, mind you, and we can only approach them, but it does give an idea as to the maximum speed in a current technology. In other words, you can't put data in one side of a chip and get it out any faster than the speed of light on the other.

When I ran the numbers myself, I was shocked at how close we were to those limits. For a .375" die, the maximum speed in photonic computing could only be 31.5GHz. Considering we are even a small fraction of that speed now is a remarkable feat.

That is why it is so important to embrace parallel computing. It is the only way to scale performance without speed limitations (not surprising the direction Blue and Green are going).

What is interesting is going to be the programming in the following years. For instance, most mp code is very inefficient on single processor systems, and vice-versa.


RE: How long until...
By masher2 (blog) on 12/7/2006 9:26:12 AM , Rating: 2
> "Just FYI, There is a theoretical limit to the speed of any processor. For electrons, it is the die size divided by the speed of electrons in silicon..."

This is true only if one assumes a monolithic synchronous clock. If each portion of a die is independently clocked, the limitation disappears. It's rather difficult to do this with a single core, but with multiple cores on a single die, its not a problem. Thus, clock rates will still continue to rise and one day exceed the "theoretical" limit.


RE: How long until...
By coldpower27 on 12/6/2006 7:37:27 PM , Rating: 2
There is an absolute limit as the size of the silicon atom itself at around 0.3 nm in diameter, if I am not mistaken.

Though it's hard to say what the feasible limit of silicon semiconductors will be.

We have roadmaps up to about 22nm node for the moment and if it follows the 2 year transition rate we have been experiencing for the past few generations, it should arrive theoretically for Intel in late 2011, so if the article is predicting 2015 before we see any products produced on the 22nm node, then something needs to happen that makes the development of this node more difficult then any node prior if there is an additional 3.5 years extra before products appear.


RE: How long until...
By rushnrockt on 12/6/2006 9:21:01 PM , Rating: 4
The silicon atom is not the limitation of the size of the circuit, electron tunneling is.


RE: How long until...
By tuteja1986 on 12/6/06, Rating: 0
This is all great but
By Dfere on 12/7/2006 8:16:18 AM , Rating: 2
The real news here is what will happen if we indeed do hit a wall. Historically, Intel's big advantage has been in company resources and to be the first to shrink die size, then benefit from it. AMD, to date, has been able to come up with a design change to keep ahead of Intel. Until Core 2. What happens if Intel hits a wall and AMD is able to compete with less of a disadvantage in manufacturing process? Very interesting indeed! I actually think this will push Intel to innovate more quickly.




RE: This is all great but
By iNGEN on 12/7/2006 2:30:12 PM , Rating: 2
Necessity is the mother of invention. When there is no longer a competitive advantage in process shrink, the eggheads will come up with something new. I actually look forward to the wall.


FutureNews!
By therealnickdanger on 12/6/2006 5:59:28 PM , Rating: 2
With news like this, DT should change its name to FutureNews, and then give us the 11 o'clock news at 6... ;-)

j/k, nothin' but love for DT.

I'm just getting used to