Moore's Law Limits Pushed Back Again 334
quackking writes "Since the weather in Rochester stinks, people spend a lot of time indoors making cool stuff. And at Rochester Institute of Technology, they have figured out how to make silicon chips with 38 nanometer rules..this is an order of magnitude better than what is standard at present. The process is called liquid-immersion nanolithography, a cool idea - starting with the commonly-observed phenomenon that things look bigger under water - they submerge the silicon wafer."
Power use(rs) (Score:5, Interesting)
Betcha the top-of-the-range chips that Intel and AMD make will *still* manage to consume ~100W of power
Talking of power users, it could make for some seriously large on-die L2 (or even L1) cache though. Since the fetch-from-memory is like hitting a brick wall (for a CPU), the more the better - look at how the P4EE performs compared to the non-EE version...
I guess it could also be used for lots of on-chip cores. 16 CPU's per die would be nice, although they'd have to have a large die for all the memory traces going to the motherboard. Even AMD's hyper-transport might struggle with that
Simon.
Re:Power use(rs) (Score:5, Informative)
Because leackage is only increasing with further process shrinks. And its increasing A LOT.
With 0.35 um processes, leackage power was in the
More logic transitors will only result in even more waste, at least with tratitional design rules...
Re:Power use(rs) (Score:3, Informative)
Re:Power use(rs) (Score:5, Funny)
Which means they need faster processors to design their cooling systems. Oh, the Humanity!!
Re:Power use(rs) (Score:3, Insightful)
100W isn't that much power or heat to dissipate.
It's useless to complain that improving computing power is increasing power consumption.
It's more useful to complain that battery capacity is not increasing, or that computers come with inadequate cooling systems.
Re:Power use(rs) (Score:5, Insightful)
Re:Power use(rs) (Score:4, Funny)
As a side note, I have to admit I actually feel pride for being a student who goes to RIT.
Re:Power use(rs) (Score:3, Interesting)
The current top of the line practical process is the 90nm process which Intel and IBM use, and AMD is soon moving to. 65nm is already in full swing as far as design and testing go(large-scale introduction in 2006, maybe 2007), but I believe 65nm is the smallest process currently designed yet; everything else past that(up until now) was theoretical.
Re:Power use(rs) (Score:4, Interesting)
Since the fetch-from-memory is like hitting a brick wall
DRAM should be eliminated then. We should all move to SRAM (the kind of RAM used for cache). It is expensive right now, but if produced in large quantities, it will become just as cheap as DRAM currently is.
Why don't they just use a car mirror? (Score:5, Funny)
"objects may be larger than they appear."
I have them all over my bedroom.
Re:Why don't they just use a car mirror? (Score:5, Funny)
Cheers.
the traditional method is.... (Score:5, Funny)
Re:the traditional method is.... (Score:2, Funny)
"Candy is dandy, but. .
Well, you obviously know the rest.
KFG
Re:Why don't they just use a car mirror? (Score:3, Funny)
LK
Cool Idea? (Score:5, Interesting)
As these chips get smaller and smaller, 'cool' is the only thing that isn't going to be synonymous with these chips.
Any idea how they are going to deal with stability and cooling of these new chips. New computers already use some pretty crazy cooling systems.
Will watercooled systems become the norm?
Re:Cool Idea? (Score:2, Insightful)
Re:Cool Idea? (Score:5, Insightful)
1) Cost
2) Difficulty of setup
3) Public knowledge
Cost will naturally come down as usage increases. Setup, on the other hand, is still a rather difficult process involving thermal paste, clamps, and lots of water. Most people won't accept it until there's a big funnel marked "ADD WATER HERE" - compounding the problem of non-acceptance is tha fact that nobody knows the cooling we've seen. Nobody's even thought about it. Simply, it's a rare event for a stock PC to spontaneously overheat. The major PC manufacturers tend to prevent that from happening, and can't sell PC's that aren't just "plug and play" in the hopes that customers won't be driven away by complexity of setup.
So, common watercooling will be a fair ways away, as it still needs to be perfected. I suspect that one of the first major barriers will be modding a case to heat a tank of tropical fish. Overclockers may want to use this technique to simulate conditions outside a geothermal vent and run their own curious creature farms.
Re:Cool Idea? (Score:3, Interesting)
Would negate the need for water cooling at least for a while.
Vapo Chill more likely (Score:4, Interesting)
Re:Cool Idea? (Score:3, Insightful)
Re:Cool Idea? (Score:5, Informative)
Re:Cool Idea? (Score:2)
They where also quite expensive. Even today, a good water based cooling for your CPU is costly. In any case far more costly (and cumbersome) than most CPU coolers.
It's called BTX. (Score:2, Informative)
Re:It's called BTX. (Score:2)
All BTX does is cooling the CPU first, then letting the warmed up air circulate over the other components inside the case, shortening their lifespan.
Here's an interesting thread: http://www.hardforum.com/showthread.php?s=&thread
Re:Cool Idea? (Score:5, Interesting)
It's very new news, and not a lot of people realize the impact yet. Cooling is the problem, not scaling. This will allow performance and levels of integration several orders of magnitude greater than otherwise possible.
Good article too (linked above). Check it out. The printed version off EETimes has some great explanatory graphics too, but those don't seem to be on the web version.
Re:Cool Idea? (Score:2, Insightful)
The heat problem is due to raising the clock at the same time and/or keeping the size the same but adding more transisitors (e.g. P4).
Re:Cool Idea? (Score:2, Interesting)
That is only partly true. By cramming more onto a smaller space you increase the problems associated with heat. While the overall heat per transistor has increased the suseptibility of the transistors to heat induced instability is worse and the heat dissapation is worse because there is less area over which to dissapate heat.
Also the distance really has little to do with the voltage, it is more related to the resistance and current. The lower voltages are related to other improvements.
The power savings
Re:Cool Idea? (Score:2)
So Small (Score:5, Funny)
Re:So Small (Score:3, Funny)
Re:So Small (Score:3, Funny)
or
Opteron Wow!, now with Olestra (some leakage may occur)
Re:So Small (Score:2, Funny)
Why not "Chiplets", which is what I have taken to calling the numerous small broken bits of potato chips in the bottom of the potato chip bag.
That explains (Score:5, Funny)
I knew there was a perfectly rational explanation for why love in the swimming pool always seems more passionate...
Re:That explains (Score:3, Insightful)
This will never catch on. (Score:5, Funny)
Re:This will never catch on. (Score:5, Funny)
Re:This will never catch on. (Score:2)
The old joke around the lab with hardware that was dying was "Oh, have you checked to make sure it floats yet?"
Guy knows his stuff... (Score:5, Interesting)
Just google for "Emil Piscani". Not quite what I wanted, but interesting enough...
High index of refraction is this trick. (Score:2)
This trick is to use the high index of refraciton of the liquid to slow the light. Slower light, shorter wavelength for a given frequency, smaller features before it dies.
Vacuum and zero gravity may have other useful effects but unless they enable something new they seem unlikely to affect feature size by the
Stupid question... (Score:5, Insightful)
But 38nm a "magnitude better" then current processes?
intel is shipping 95nm prescotts (even if the process sucks (see leakage power), but that has nothing to do with the lithography per se), and 65nm is in developement, with sample chips demeonstrated.
So its more like a "binary order of magnitude" better than current processes..
Re:Stupid question... (Score:5, Informative)
Re:Stupid question... (Score:2)
Re:Stupid question... (Score:2)
Seemed reasonable at the time, given the digital/binary logic context...
Re:Stupid question... (Score:2, Funny)
Everybody knows that you can increase the fidelity of your lithographic process by increasing the numeric aperture of the optical system.
Why do I feel singled out here?
Re:Stupid question... (Score:3, Insightful)
Re:Stupid question... (Score:2)
100,100 nanometers? Ouch!
Re:Stupid question... (Score:5, Funny)
Re: Stupid Question... (Score:2, Funny)
Having read your post, I would say you probably are on topic, but I can't say for sure since my brain just exploded.
bigger? (Score:5, Funny)
wonder what would happen to CowboyNeal when he's placed under water....or the national debt....
Useful Links (Score:5, Informative)
Main Site: http://www.rit.edu
Kate Gleason College of Engineering: http://www.rit.edu/~630www/index.htm
Microelectronic Engineering Department: http://www.microe.rit.edu/
Optical Lithography Research: http://www.rit.edu/~635dept5/
While RIT-bashing is one of the most popular activites around here, RIT isn't such a bad place to be at, even if you're a liberal arts major (which I am). They do some pretty neat stuff around here!
Re:Useful Links (Score:2)
Re:Useful Links (Score:2)
The administration has a tendency to do stupid things without the students being consulted, or if they are consulted, against the majority opinion:
1997: Changed residence halls to alcohol-free zones, stopped selling alcohol in the resident grocery.
2000: Signed a 10 year exclusivity contract with Pepsi, banning Coke and other smaller soda companies (Jolt!) from being sold by any food service on campus.
2000: Began accepting more students than they could house, putting many students in temp
Re:Useful Links (Score:2)
Amusingly enough, I saw an article in the Democrat and Chronicle(the same newspaper mentioned in the story) that said that the populace of Rochester feels much the same way about the city itself.
Oil better than water better than air (Score:5, Informative)
Better article to explain immersion lithography (Score:3, Informative)
Still, the work of the researchers should eventually get us there.
Carbon disulfide & Methylene iodide even bette (Score:5, Informative)
Good Idea/ Bad Idea (Score:5, Interesting)
What I mean is developing new materials to create processors, or reinventing current methods to save space and power, rather than finding new ways for the status quo to stick around longer. The September 2003 edition of Wired Magazine had a cover story about creating flawless synthetic diamonds, and the possible uses for them as processor components. It turns out that a diamond semiconducter has been developed.
It has met with major disapproval from both the diamond cartels (ie DeBeers, as synthetic diamonds have the potential to damage their business) as well as semiconducter manufacturers (since they have so much invested in silicon).
It's possible that both could work together, as the diamond semiconducters are in their relative infancy, and this could provide an interim solution. Or, the big manufacturers could try and drag Moore's Law on as long as possible with silicon.
Re:Good Idea/ Bad Idea (Score:2, Interesting)
OT Question about DeBeers (Score:2)
what about optical chips? (Score:5, Informative)
Sorry for the basic questions, but I'm just a programmer
Peace & Blessings,
bmac
Re:what about optical chips? (Score:2)
No. Not at all.
1.) It will be a LOOOOOOOOOOOONG time before all chips are optical. It will take a while for the optical processors to be developed so that they are as fast as whatever the current batch of processors is, and as cheap. Even then, it's difficult to imagine all the chip makers scrapping all their factories and going all optical.
2.) I imagine there'll be some big r
Re:what about optical chips? (Score:5, Informative)
Is it good news? (Score:4, Interesting)
Re:Is it good news? (Score:2, Insightful)
All very nice (Score:3, Insightful)
Re:All very nice (Score:2)
Immersion Litho Tools (Score:2, Informative)
ASML [asml.com]
Canon [canon.com]
Nikon [nikon.co.jp]
One Technology Drives Another (Score:3, Interesting)
I can see the guys who design chip coolers getting excited about this, cause lets face it - This is their porn.
Things look bigger... (Score:4, Funny)
Wrong Industry Focus (Score:2)
The PCI bus has come of age but it makes no sense to have all this speed and power if we cant stuff or retreive it as we wish from a 150M/s device.
Re:Wrong Industry Focus (Score:3, Insightful)
Optics (Score:5, Insightful)
An approach to reducing the wavelength is to simply go to higher frequency, but that poses some tough challenges (x-rays don't behave like visible light). What these researchers have done is to change the environment such that the wavelength is smaller (light is 25% slower in water, and because frequency doesn't change, wavelength gets shorter). Anyway, 25% is hardly "an order of magnitude".
Re:Optics (Score:2)
Eg if you want to etch a cross:
|
--|--
|
You make a litography mask somewhat like this:
|||
==.==
|||
But the price of a mask set with this tech is very high
Lobster Tank (Score:2)
Doubtful, lobsters hate litigation.
This description annoys me (Score:4, Insightful)
The technique mentioned seems to be forming a liquid lens. So my question is, why is this any different from a conventional lens. Lenses are used all the time in lithography.
Does it help with diffraction limited optics or is there some other reason?
They're talking about increasing the imaging limits even at current wavelengths by effectively changing the lensing system, so it has to be something to do with reducing diffraction IMHO.
So... anyone here care to offer a reasonable technical description of why this is?
Does a single refractive interface with the wafer submerged and therefore inside the lens help reduce diffraction?
Re:This description annoys me (Score:3, Informative)
It annoys me too, but for a different reason. The fact that things look bigger in water is completely irrelevant to the subject at hand - that is just an optical trick to our eyes (which are outside the water) when the water is contained in a round container (a flat container does not exhibit this effect!) and would make no difference to a lithography system.
Contrary to the parent comment's
Re:This description annoys me (Score:2)
Does this make sense to anyone?
Re:This description annoys me (Score:2)
Re:This description annoys me (Score:2)
What I mean is obvious and I know what refraction is I don't need a description of it, it's the fundamental way lenses work, I also know what diffraction is and the difference between refraction and diffraction. In imaging there's a trade of between refraction and diffraction where the area of confusion becomes diffraction limited especially for smaller appertures.
Thanks for the rest of your post, however the lack of any mention of a
Re:This description annoys me (Score:2)
Thanks for your excellent explanation. I finally understand I think. So the focus is diffraction limited and by placing the focusing apperture in liquid they reduce the diffraction limited circle of confusion on the target. Excellent, thanks, now why didn't they just say that?!
Doctor Who... (Score:4, Funny)
This seems supiciously like the Doctor's explanation as to why the inside of the Tardis is bigger than the outside: Take two cubes, one bigger than the other. Now hold the bigger one at an arms length just far enough so that it looks like it can fit in the small one. That's how its done!
Leela: "That's nonesense"
Doctor: "Nonesense? That's transdimensional engineering, a key time lord discovery."
ALRIGHT! (Score:2, Interesting)
Moore's law says we need a new direction (Score:5, Insightful)
We need a new direction. Moore's law is still in effect, but it doesn't dictate die sizes, only speed and cost. The most obvious alternate road is parallel processing. Multiple chips in other words. We're already doing this. Outside of the PC world this is old hat. We think we're all 1337 because we have a four-way Xeon server, but the non-PC world just yawns at this.
My prediction: PCs will have 64 processors, each of which will be cooler than today's 3GHz+ p4, but will provide a magnitude more processing power. Software (or compilers) will have to designed for this new architecture, but it's the only way we're going to see a PC capable of running Longhorn or Linux 2.8 that doesn't take 500 watts.
p.s In the meantime, software follows Moore's Anti-law, which states software will waste all additional resources provided by Moore's Law. If only software would keep up with hardware I would be ecstatic. When WP5.1 did 98% of what WordXP does today, but did it on a 640K 16-bit processor, it's hard to say software is improving in any area but the GUI.
Re:Moore's law says we need a new direction (Score:2, Insightful)
Re:Moore's law says we need a new direction (Score:3, Informative)
Unfortunately for you it's not the de-integration you predict, rather more still more integration with a very clever, very scaleable, amazingly efficient, built-in cooling system.
Re:Moore's law (Score:3, Insightful)
I
Not so revolutionary (Score:4, Insightful)
Heat sink (Score:4, Insightful)
Sorry, but the heatsink is part of the CPU, IMO. The CPU cannot run without the heat sink, so it has to be considered a single unit.
Given that, I'd say that CPU's are increasing in size, not decreasing. My 486-66 may have been 100 times slower than today's P4-3GHZ, but it was also 50 times smaller since it needed no heatsink. Thus, I'd say we only increased size/speed ratio by 50%.
I thought PowerPC G5 chips were made this way (Score:3, Informative)
Some more (better/worse) links on the topic (Score:2, Informative)
Re:Some more (better/worse) links on the topic (Score:2, Interesting)
http://biz.yahoo.com/prnews/040218/nyw086_1.html [yahoo.com]
http://216.239.41.104/search?q=cache:Lurb_nMoW6YJ: www.siliconstrategies.com/story/OEG20030227S0068+i mmersion+lithography&hl=en&lr=lang_en&ie=UTF-8 [216.239.41.104]
http://www.future-fab.com/document.asp?d_id=189 [future-fab.com]
Don't forget George Costanza's Law (Score:3, Funny)
Shrinkage (Score:3, Funny)
Teraflops from 1 CPU with no heat (Score:3, Interesting)
A snip from the article:
But the greatest potential for CVD diamond lies in computing. If diamond is ever to be a practical material for semiconducting, it will need to be affordably grown in large wafers. (The silicon wafers Intel uses, for example, are 1 foot in diameter.) CVD growth is limited only by the size of the seed placed in the Apollo machine. Starting with a square, waferlike fragment, the Linares process will grow the diamond into a prismatic shape, with the top slightly wider than the base. For the past seven years - since Robert Linares first discovered the sweet spot - Apollo has been growing increasingly larger seeds by chopping off the top layer of growth and using that as the starting point for the next batch. At the moment, the company is producing 10-millimeter wafers but predicts it will reach an inch square by year's end and 4 inches in five years. The price per carat: about $5
I don't know what we can look forward to after that =P
The real scoop on immersion lithography (Score:5, Informative)
All optical systems have a certain "numerical aperture" or NA, which is equal to the index of refraction of the immersion medium times the sine of the half angle between the lens and the imaging plane, ie: NA = n * sin(theta/2). In traditional lithography systems, the immersion medium is air with an index of ~ 1.0, so the theoretically maximum NA is 1.0, since the half angle cannot be greater than 90 degrees.
NA is basically a measure of how many diffracted rays make it through the lens. When light passes through the mask, the light diffracts, i.e. spreads out. Light that spreads out the most contains the highest-resolution information. The lens's job is to collect as many rays as possible and focus them to form an image. Of course, not all of the rays are collected, so the image is degraded somewhat. Lens designers try as hard as possible to collect as many orders as they can (i.e. increase NA) to give the highest possible resolution.
The resolution of the system is actually proportional to the wavelength divided by NA, so there are two approaches to making smaller printed features. First, the wavelength can be made smaller. This has been done over the years, and has been moved from 436nm to 365nm to 248nm and now to 193nm. An attempt was made to move to 157nm, but the materials challenges proved to be too difficult. (These wavelengths, by the way, are either peaks of the mercury spectrum, or various excimer laser wavelengths). Second, the NA can be made larger. This has also been done, with each generation of imaging systems having higher NA. NA has gone from 0.3 or so to 0.75 or 0.80. Every time a move in wavelength or NA has occured, a tremendous amount of research and development has been needed. Also, the imaging systems have become increasingly complex. A state-of-the-art 193nm "scanner" now runs for around $15 million.
Immersion lithography works because you can increase the NA above 1. Water has an index of refraction greater than 1.0 (it's 1.333 for visible light, not sure for 193nm). Of course, this is all math. What's really going on is that rays that are diffracted at such a large angle that would normally be totally internally reflected inside the lens, can now be transmitted. As I said, the more diffracted rays that make it through, the higher the resolution you can achieve.
Although I doubt water immersion will be good enough for the 38nm node, other immersion liquids with a higher index of refraction will increase the NA further still, and push the resolution even higher. It is thus very likely that optical lithography, whose death has been predicted forever, will continue to be the dominant technology in making microchips.
Re:Evidently, I picked the right college..., (Score:2, Funny)
It's snowing today, the week before Easter (Score:2, Funny)
Re:lenses (Score:2)
There's more to it than that.
The index of refraction for water is greater than that of air. So in effect, they can create "more magnification" than with air alone and standard lenses.
Re:lenses (Score:2)
Re:But which way is Moore (Score:3, Insightful)
Before we get to that point we'll have the problem of finding a new dielectric; The silicon dioxide gates are currently a few molecules thick and can barely hold back a few volts. It'll take either a new dielectric material, or a far colder chip (because to keep working with SiO2 means lower voltage, and therefore a lower temperature to keep static accpetable).
Personally I place my bet with optic