IBM's Snowflake Microchips

Phantom of the Opera writes "The BBC reports that using self-assembled polymers and copying natural patterns, IBM hopes to have microchips that are 30% faster and consume 15% less energy. The secret? Adding a little nothing in all the right places."

Comments

Boo

[fatduck]

What happened to the provocative, editorializing troll-summaries? How am I supposed to start a heated argument based purely on speculation? You give me what, two sentences, like you want me to read TFA? Well, fuck you. Self-assembling polymers? Copying natural patterns? Who makes these things, IBM, or CYBERDYNE? What if these get into the hands of our children? Will the next school shooting be 30% faster and 15% more efficient?

Re:Boo

[pipingguy]

[ IBM ] [ Hardware ] [ Technology ]

fatduck writes:
"What happened to the provocative, editorializing troll-summaries? How am I supposed to start a heated argument based purely on speculation? You give me what, two sentences, like you want me to read TFA? Well, fuck you. Self-assembling polymers? Copying natural patterns? Who makes these things, IBM, or CYBERDYNE? What if these get into the hands of our children? Will the next school shooting be 30% faster and 15% more efficient?"

I wonder what the Slashdot community will say about this?

> [+] hardware, ibm, technology (tagging beta)

Re:Boo

[RealGrouchy]

Yeah, it's a little editorial tactic called "putting a little nothing in all the right places".

- RG>

Re:Boo

[someone1234]

I guess you miss the 'snowflake in hell' notes?

Re:Boo

[Almir]

the article should have been named: 'should apple buy ibm? macs are already unique and beautiful snowflakes, now running 30% more uniquely and beautifuly!' but amd can still make the fans. to keep cool, you know?

Re:Boo

[Gilmoure]

should apple buy ibm If Apple bought IBM, there would suddenly be enough neckties and suit jackets to power an Orion. Would have to make sure they weren't stored in one local or they might initiate fusion. Bleah!

Ooh, but then there'd the great turtleneck drought of '07. Would be boring my great grandchildren to tears with that tail, I would.

Re:Boo

[BitchKapoor]

So a 35% chip speed increaes by 2009? Something tells me chips will be a hell of a lot faster than that by then anyway.

Did you ever stop to think, how are they going to actually get a lot faster? This technology is part of the story behind how chips are going to get faster. It will be combined with other technologies so that we can keep up with the demand for increasing performance. Chips don't just get faster on their own, there's a whole industry working on it.

Nature's Little Inventor

[MankyD]

I know it comes as a surprise to no one that Mother Nature has some truly incredible engineering at work. I still, however, find it fascinating and amazing when examples like this come to light; I feel we will continue to see a lot more discoveries like this for the foreseeable future

I have two questions for Slashdot: Are there any other unique examples of learning from nature that you'd like to bring to light? And on a different note, do you think nature has perfected certain tasks and that its engineering can not be surprassed (at least in some areas), or are there things that even nature hasn't perfected?

Re:Nature's Little Inventor

[Anonymous Coward]

Me family crush by rolling boulder. Me invent wheel.

Re:Nature's Little Inventor

[Shihar]

I think it is safe to say that 'nature' has not 'perfected' anything. Certainly there is a lot out there to be inspired by, but when it comes down to it nature performs a lot of guess and check to solve problems (although, even the use of the word problems is debatable, nature isn't a think that has problems).

Personally, I think that you can step back and view all of 'nature' and include humanity. If you do, I think you will come to the surprising conclusion that humans are just another step on the path. I am not saying the path leads anywhere, but you see a sort of progression going on.

Picture the universe how it was. It used to just be a mess of boring old atom parts. The parts formed up into atoms, and the atoms started forming up into molecules. Now, there are some molecules out there that form pretty easily. Hydrogen merrily grabs other hydrogens, carbon loves a pair of oxygen, so and and so forth. We are still talking about a pretty simple universe. At some point more interesting things started to happen. These molecules started to form into more complex molecules. Long complex strands of organic molecules started popping up (among other things). The universe is starting to get a little more diverse at this point. On Earth, at some point, these organic molecules started to show some really crazy behavior. They started self assembling into even more complex structures and forms. We have a sort of non-biotic evolution going on that slowly leads to more complex molecules and systems of molecules. At some point, we get the first bits and pieces of life.

Once life shows up, things really kick into over drive. This slow multi-billion year process that got us basic organic molecules explodes as pieces of the universe come together to form the truly complex chemical system that makes up life. Evolution takes over and life begins to change rapidly. We are still talking about single celled organisms. At some point in a not-too-distant-past (well, on a cosmic scale) life started to get really complex, really quickly, as multi-celled organisms burst onto the scene. At some point, in just a blink of an eye on the cosmic scale, humans popped up from the evolution of multi-celled life.

With the introduction of humans, this natural evolution towards complexity dramatically speeds up. Non-biotic evolution was slow. Biotic evolution was faster, but still took millions of years. Intelligence though... that was fast. Where evolution found it was by tedious chance, intelligence could find its way through rapid (although messy) computation. Throw in language and writing which allows easier data retention, and intelligence gets even faster.

There is a theme to this. Greater complexity, faster and faster. Personally, I think that we are on the cusp of the next great revolution in this universe. In the same way the universe moving from random molecules bumping around to evolution, and moving from evolution to intelligence was a dramatic change, I think we are on the cusp of the next revolution. The next revolution is of course strong AI, which can create ever accelerating growth in intelligence. I am not saying it is good or bad, just that it is next. I think to separate intelligence and (eventually) AI from evolution and molecules randomly bouncing off each other misses a larger trend. It isn't human Vs inhuman, it is the universe rapidly finding better ways to create more complex systems, and create them faster and faster.

So, to go back to the original poster... I don't think you really can separate the works of man from the works of nature. A computer is a work of nature at its finest. The fact that a computer is wrought from human intelligence doesn't make it any less an awesome work of nature then a monkey that got to be the way it is through evolution, or a molecule that got to be the way it is through a chemical reaction. The greatest work of nature (from my perspective of course, it isn't like 'nature' has a goal in mind) is human intelligence. That said, I doubt that intelligence is the last step on the path. I think the greatest works of nature are yet to come.

Re:Nature's Little Inventor

[lixee]

And on a different note, do you think nature has perfected certain tasks and that its engineering can not be surprassed (at least in some areas), or are there things that even nature hasn't perfected?

My, yes. The human body! Good luck getting anywhere close to that.

Re:Nature's Little Inventor

[ushering05401]

While manufacturing processes that enable chipmakers to accomplish this type of feat are innovation I wonder if the designs themselves are going to be considered original.

There wouldn't be any issue with proving prior art.

Regards.

Re:Nature's Little Inventor

[FooAtWFU]

On a cosmic timescale, I can agree with you about Strong AI. But, having spent the past semester (exam on Wednesday!) learning about the state of AI, I can indeed assure you that you will need to wait quite a while before you start to see anything really Strong AIish coming to pass. Maybe towards the end of my lifetime you'll start to see something decent, but I'm guessing not. And even if they do have something by then, consider that NIs (natural intelligences) take many years (decades!) to get to the point that they do. Some of that can probably be skipped for the artificial version, but I've got little reason to doubt training the things will be Fast at all.

Re:Nature's Little Inventor

[xenocide2]

The thing about AI is that what it is keeps changing. Twenty years ago, Expert Systems was AI. Chess programs used to be considered AI, and some people still do, but generally the pattern seems to be that AI is about unsolved problems. Once we solve them, it's no longer intelligent. Some of this I can understand, simply because as we discover how things work they lose some mystery and intelligence is one of the most mysterious properties we assign to humanity. But very little of the early stuff tried to do learning. As if once we knew how to win at chess, we could take that and apply it to the rest of the universe, treating it like a bigger instance of a chess game.

Re:Nature's Little Inventor

[rm999]

You are talking about technological singularity
http://en.wikipedia.org/wiki/Technological_singula rity

I think it is a very interesting theory - basically it states that up to now we have had exponential growth in significant "paradigm" shifts. While the first human technologies took 1000s of years to develop, today there is some revolution in technology every few years.

Singularity claims that when AIs become smart enough to take over the human's role in designing technology (and themselves), we will experience something faster than exponential growth.

This is a very similar pattern to what happened with ICs. The first computers were meticulously hand-built by people. Within a few years, computer aided design helped humans build faster, more impressive computers in a fraction of the time. Chip synthesis tools now allow us to design chips with a few 100 lines of code. While chips have remained "only" exponential in growth, this is more of a limitation of the chemical and physical properties than the design limitations.

Re:Nature's Little Inventor

[fractoid]

To bastardize a perfectly good term, 'hard AI' is the "Intelligence of the Gaps". It's the 'hard stuff' we need to figure out in order to boil currently human-requiring tasks down into programmatic algorithms. The chess argument is actually true - you *can* apply a search-based AI to the universe, if you're a theoretical computer scientist with an infinite Turing machine. Of course, we don't, which is where things get interesting.

The problem, as you say, is that as soon as we can code an algorithm to solve a task, that task is no longer considered to require intelligence. I blame the fact that we don't actually know wtf 'intelligence' is. All attempts to define it that I've seen are circular. Eventually, I don't doubt we'll have algorithms to solve any task a human could reasonably be expected to solve (so I guess I'm a strong AI proponent); will we then be forced to define ourselves as unintelligent? Or will we be forced to conclude that our code is intelligent? Most likely, it seems to me, is that we'll simply end up defining intelligence as "what humans do to solve problems". Kind of like what we had to do with swimming when submarines were invented.

Re:Nature's Little Inventor

[alphamugwump]

The big problem with humans is we're so low bandwidth. We can type at, what, 50 WPM, and talk maybe twice that? If you gave an AI a reasonable connection to the internet, I thing you could train it a lot faster.

If there is going to be a singularity (I don't really see why there wouldn't) strong AI wouldn't be the really world-changing thing. Humanoid robots are way, way, old school, and not really that interesting. The interesting thing would be what would happen when humanity is networked together with a high-speed connection, without a bottleneck at the eyeballs.

Re:Nature's Little Inventor

[kripkenstein]

>> do you think nature has perfected certain tasks and that its engineering can not be surprassed [sic]?

My, yes. The human body!

The human body? Well, most any organism would be just as good an example. Engineering is no where near the efficiency and capabilities of, say, a housecat (in both perception and action).

Re:Nature's Little Inventor

[zevans]

My, yes. The human body! Good luck getting anywhere close to that. Oh, this old chestnut.

The human body is a poor solution for many of the tasks that it performs. Ask anyone with cruciate ligament injuries.

If you spent a billion engineer-years on designing something that could hunt, climb trees, AND swim, you'd come up with something very different to the current design.

Re:Nature's Little Inventor

[dreamchaser]

I know girls who type at 100wpm and talk at what seems to be a hundred times that rate ;)

On a more serious note, if we ever do get a high bandwidth brain-computer link, I wonder what that will do for brain development and education? While it would be nice to have the ability to download new skills (a la 'The Matrix'), I somehow think there are other limiting factors (I won't get into the need to train the rest of the nervous and muscular systems before one can say "Whoa, I know Kung Fu").

Good thinking fodder though. I'll plod my low bandwith way through it some more after I do some work.

Re:Nature's Little Inventor

[dpilot]

>I know girls who type at 100wpm and talk at what seems to be a hundred times that rate ;)

But how fast are they moving information, as opposed to filling air with sound?

Like, you know I wouldn't even, like call some of those, like, extra like, sounds somthing like "error correction", you know, but just like, like more like extraneous junk, like you know, like junk DNA, you know. Unless of course junk DNA turns out to actually be useful, once we learn more about it.

Re:Nature's Little Inventor

[hab136]

My, yes. The human body! Good luck getting anywhere close to that.

Tonsils, appendix, wisdom teeth - useless things that often need to be removed because they cause trouble
Only redundant in some essential systems (kidneys, eyes, lungs) but not others (heart, liver)
Limited storage space - the reason the brain is so folded over is because it's run out of room in the skull
Terrible vision compared to cats, eagles
No ability to hibernate (bears) or regrow limbs (starfish, some lizards)

The human body is far from perfect.

Re:Nature's Little Inventor

[marcosdumay]

Most people really don't understand exponential* growth... Take a hint: It starts slow, very slow.

Not saying we'll see AI in our lifetime (altough lots of people really think we will), but it may be just around the corner, and still your analisys wouldn't see that.

* Technology is growing at exponential or highter than exponential rate everywhere, even on IA. An highter than exponential rate is different, because it starts slower.

Re:Nature's Little Inventor

[marcosdumay]

"While chips have remained "only" exponential in growth, this is more of a limitation of the chemical and physical properties than the design limitations."

Moore's law is an economical one. Companies simply don't have enough money to spped things up, and semiconductor fctories are really expensive.

Re:Nature's Little Inventor

[frogstar_robot]

The problem, as you say, is that as soon as we can code an algorithm to solve a task, that task is no longer considered to require intelligence.

If that is the case, it seems to me that the Holy Grail would be an automated way to find practical algorithms for arbitrary tasks. Humans can eventually investigate and fumble their way towards the solution to a given problem. The more humans looking at the problem the more likely it is that an individual or team will find the solution. When we can make a machine that can do that over a broad range of tasks then we'll have a strong AI to talk about.

Re:Nature's Little Inventor

[0100010001010011]

Lightning in sand -> injection molding.

I saw some stuff in college where they were studying fulgurites (the glass created when lightning hits sand) for the optimal method of injecting plastic into molds. Lighting will always take the path of least resistance. So you have a part you want to mold, they'd pack it full of sand and then hit it with artificial lightning. They'd then study the way that the lightning propagated through the sand to determine the way that all the injection lines should run.

Re:Nature's Little Inventor

[camg188]

Are there any other unique examples of learning from nature that you'd like to bring to light?

A stream of urine is a good electical conductor.

Re:Nature's Little Inventor

[default luser]

Two things we sill cannot model %100 correctly, let alone produce something better:

1. spider silk.

2. insect flight.

Re:Nature's Little Inventor

[not-enough-info]

AIs may take a long time to train, but what's to say they won't be able to copy themselves and diverge once mature? Taken a step further, you could see this as accelerated evolution.

Re:Nature's Little Inventor

[rm999]

I have been studying AI in grad school, and mostly agree. It has actually come to the point that most AI type people prefer not to use the term "artificial intelligence" at all. You would be hard pressed to find a class called "artificial intelligence" in any prestigious graduate school (beyond maybe an intro class). Instead, look for "machine learning," "neural networks" and the like. I postulate that this is because people have made much better progress on specific topics that don't really move towards anything like strong AI - it's not even a goal for most people anymore. In fact, a lot of researchers don't even know what strong AI means. It is more of a science fiction or futurist term.

I would love to see strong AI in my lifetime, but I'm not going to hold my breath, nor am I going to waste my time researching it until we know more about its required foundations.

Re:Nature's Little Inventor

[Bent Mind]

The problem, as you say, is that as soon as we can code an algorithm to solve a task, that task is no longer considered to require intelligence.

I've never really seen a definition for intelligence. Most people just give examples of what they think it means. I suppose you could apply this statement to any human state. What is love, hate, etc.?

So for my example; computers seem to have a very hard time with guess work. It seems to come naturally to humans. Humans guess at an answer, test their guess, and feed the results into a new guess. Though computers are better then us at modeling data, they stumble on the guess work. They need everything in concrete numbers. Abstract thought doesn't boil down to programmatic algorithms very well.

It's funny. I've heard so many stories about accidental discoveries. Maybe computers are just too perfect. Perhaps they need flaws and mistakes to gain intelligence.

Obligatory Simpsons reference

[The Hobo]

Ripped from snpp:

Homer, meanwhile, uses a pickaxe to make some speed holes in his own
car.

    Ned: Whatcha diddely-doin', neighbor?
Homer: Aw, putting speed holes in my car. Makes it go faster.
    Ned: Is that so? Well, gee, maybe the old Flanders-mobile could use
              some -- [a shot rings out] aah! [Ned collapses]
                [he gets up slowly] Wow! Lucky I always keep a bible close to
              my heart -- [boom!] aah! [Ned collapses]
                [he gets up] Ho ho, lucky I was wearing an extra large piece of
              the True Cross today. I think I'll go inside.
                [a shot hits Homer's pick axe]
Homer: What keeps doing that?
  Tony: I told you we should have bought more than three bullets. Let's
              just grab him!
-- Louie the henchman, not the marksman, "Homer the Clown"

Re:Obligatory Simpsons reference

[gardyloo]

Huh. Sorry for the "troll" modding. I thought it was pretty funny, and apropos.

out of nothing at all

[Anonymous Coward]

making love at of nothing at all.

wait, what

[Merc248]

something + nothing = ... more somethings

Is this sort of like the proof that 2 = 1 or what

Re:wait, what

[Phisbut]

something + nothing = ... more somethings
Is this sort of like the proof that 2 = 1 or what

Ha! Reminds me of a funny quote I read somewhere :

At the beginning, there was nothing. God said "Let there be light!". There was still nothing, but now you could see it.

Just like a snowflake.

[Anonymous Coward]

So, no two will be alike? Programming for them will be interesting.

Re:Just like a snowflake.

[hedleyroos]

Aha!

Behold Slashdot...

Two Snowflakes May Be Alike After All

Vacuum holes?

[Centurix]

Would this mean breaking one in half would make some kind of cool fizzing sound? Move over bubblewrap!

snowflakes

[jibberson]

So does this mean no two chips are alike?

just wait

[the_fat_kid]

I'm sure that it will be pumped full of all the troll fodder you hope for, and more, when it returns in six or six weeks.
this is just a "dry run" the good stuff will be in the dupe

Re:First post!

[renegadesx]

Getting sued by oil companies wont make profit. Here's how it goes

Phase 1: Create microchips
Phase 2: ???
Phase 3: Profit

The process mimics...

[mateomiguel]

The process mimics the way snowflakes are developed

Does this mean that they drop the silicon from 3 miles up and then try to catch it on their tongues?

Re:The process mimics...

[Hanners1979]

Nope, it means you create a Beowulf cluster of them, and then throw them at somebody's head while they aren't looking.

Snoflakes and Solar Panels

[Ungrounded Lightning]

So IBM has come up with a manufacturing method using self-assembling molecules to produce regular arrays of 20 nanometer objects on the surface of a silicon wafer with near-perfect yeild. (I presume, since "growth" was involved, it would be possible to use it to construct similarly-spaced objects of sizes within a factor of about 3 to 4 of the size they chose for this process.)

And yesterday we saw a slashdot article referencing work at Rice U, Los Almos Labs, and others, where 5 to 8 nanometer quantum dots on the surface of phovoltaic cells could significantly multiply the efficiency (perhaps into the 60% range) by efficiently creating multiple electron-hole pairs per incoming photon.

Seems to me the two are just ASKING to be combined into an inexpensive manufacturing process for high-efficiency solar panels.

Doubling to quadrupling the output of solar panels while keeping the cost in the current ballpark might push photovoltaic past the cost-breakeven point compared to grid power for rural and even suburban housing loads. And that could lead to enough production to bring in additional economies of scale and drive the price point farther.

This could be big.

Mimicking Nature -

[HW_Hack]

to achieve less power consumption and more speed is the new-GHZ.

Hmmm - I need to make that a bit more snappy .....

Re:Mimicking Nature -

[zevans]

to achieve less power consumption and more speed is the new-GHZ.

Hmmm - I need to make that a bit more snappy ..... Green is the new Blue?

New fab

[LarsWestergren]

In order to meet increased demands for hype, IBM also announces the construction of a new fab for hype construction. "For a long time we hoped that we might retrofit the Cell hype fab to construct Snowflake hype, it is now clear that this new processor demands a completely new fab to create the quality and quantity of hype needed. This new hype will be so overpowering every engineer and programmer will have to relearn everything or risk being left behind. We expect our Snowflake hype will outperform our closest competitor by close to 5000%. It will also be cheaper." IBM engineer Thor Larssen stated.

So...

[JonathanJooped]

White space always helps

Calm down, it's only a mask

[BooleanLobster]

FTA: "The process, called airgap, enables trillions of microscopic vacuum holes to be placed between the copper wire in chips to act as an insulator."

Copper wire is not inlaid into silicon chips, as far as I know. I don't think they meant copper wire.

This doesn't have anything to do with snowflakes or nature. Just a self-assembling polymer that can be used as a mask to etch holes in the oxide layer of the silicon chip, making the oxide a better insulator.

Re:Calm down, it's only a mask

[MrPeach]

All complex microchips have one or more metal layers. Look it up.

Re:Calm down, it's only a mask

[Chris Burke]

They mean copper wire. It isn't inlaid, it's deposited, but the result is still a copper conductor. It first saw use in microprocessors in the later nineties, and now it's used by all major cpu manufacturers instead of aluminum, which is what they used before.

Re:IBM:

[JensenDied]

Its time for the 11 O'clock news:
Global warming is showing signs of 30% growth with 15% decrease in greenhouse gas production.
-- Time Skip --
The global flooding theories are proven incorrect as the polar ice caps are turning directly into steam.
-- Time Skip --
Wolverine is found dead in the bathtub, with melted adamantium due to the heat. In other news Earth is now proclaimed hottest terrestrial planet, eat your heart out Venus.

Tweak the ratio?

[trawg]

I wonder if they can tweak the ratio so that the chips run 10% faster but consume 40% less energy. Chips are pretty fast now, but battery life is still a bit of an issue - I'd be really interested to see if they could eke a few more hours out of portable devices (for example) and I'd be happy to keep it at the same speed.

Re:You skipped a step

[somersault]

Who created the creator being though? I'm a Christian, and I do believe in God and that he created the universe etc, but it's just crazy to me that anything at all even exists, or ever did exist. Something has always existed. It's very strange.

I also don't see how people who believe in evolution etc don't see how if something evolves enough it would be a 'god' to us anyway. It seems that all these people choose to believe that we are the most advanced intelligence/power going, they want to feel in control.

Re:You skipped a step

[mikkelm]

Beauty and complexity is in the eye of the beholder.

OK Software geeks . . .

[moeinvt]

Now I am the Master!(EE)

I'll default to a "wait & see" perspective, but this has a firm basis in device physics.

One of the major speed limiting factors in microelectronics is capacitive loading. With the tiny scale of contemporary semiconductors "wire capacitance" has become the dominant delay factor. Since the wires are so close together, adjacent wires produce a parasitic capacitance effect(extra load on the circuits) similar to this.

http://en.wikipedia.org/wiki/Parallel_plate_capaci tor

(The article on parasitic cap didn't say much)

As you can see, this capacitance varies directly with the size of the wires, is inversely proportional to the distance between them(shrinking all the time with new process technologies), and directly proportional to the "dielectric constant" of the material between them.

Air has a dielectric constant of ~1.00. Silicon Dioxide, the typical insulator in semiconductors is ~3.9.

http://en.wikipedia.org/wiki/Low-k

Other glass-like materials have been experimented with, but I haven't read about many successes.

So, essentially if you could "leave out" the SiO2 insulating material, you could reduce the parasitic capacitance of the wires by a factor of 3.9. Nothing to scoff at if you can actually pull it off.

Re:OK Software geeks . . .

[Gilmoure]

I'm just bummed the chips don't look like snow flakes. That would be cool.

Re:OK Software geeks . . .

[camg188]

The do on the inside, under an electon microscope.

Re:You skipped a step

[Shihar]

How is it that there is a "natural selection" for inanimate objects? How can such beautiful, non-living structures be works of randomness? There has to be an intelligent creator-being, no?

Give God a little credit. If there is a God, he doesn't need to use magic every thirty seconds to keep the universe on the track that he wants it. A truly omnipotent god would have flicked off the big bang and walked away, trusting in its omnipotence to be able to flick off the big bang in such a way that all that it wanted would come to pass.

Think of it like this. If you are a skilled bowler, you can roll a bowling ball in such a way that you knock over all of the pins. You angle, spin, and toss the ball in such a way that it hits all the pins it needs to hit. You don't throw the bowling ball down the lane, then cry "oh shit!" and run after it to correct its course (if you are good bowler).

If there is a god, I highly doubt that it cries "oh shit!" from time to time and needs to use magic to correct the universe. If there is a god, I imagine that it can enact its will and create a universe using only laws of nature that we humans can find and understand. To insinuate that God needs to reach down every now and then and 'fix' the universe by hand is to insinuate that God is less-then omnipotent and needs to correct its mistakes.

Re:You skipped a step

[menkhaura]

Thank you for this brilliant comment on God's nature.

But what of consistancy!

[Timberwolf0122]

If intel uses snowflake technology then no 1.9999998875 chips will be the same!

/. must be out of shape today

[hcgpragt]

Let me be the first to ask: All well and good, but does it run linux?

Re:You skipped a step

[somersault]

*shrug* how the heck am I meant to know. Why are we the only universe/intelligence there is, and why shouldn't you be able to be judged just because you don't want to be? Justice doesn't work like that.

I would like to see some data of this new LowK

[cyfer2000]

Like the Young's modulus, etching properties, copper diffusion and reliability data. Any one has a real paper? Is it dendrimer based porous MSQ?

Question

[Khyber]

So, does that mean that using air as the insulator would be better, or worse? I'm not quite up to date on physics like this. Can you roll out some layman's terms?

Re:Question

[Amouth]

the air would be better as an insulator of capacidence.. not realy an insulator.. it just doesn't alow electron flow between the wires as well as the silicon so yea it could be called an insolator

Re:Question

[Ignorance Enabled]

From the article:
"Those dots are then etched away in a chemical process, forming holes, which are then capped to create a vacuum."

Unless I am reading it wrong, we're not even talking about the dielectric of air here, we're talking about a vacuum! Now a vacuum is defined as a dielectric of 1 (or maybe I should say, all dielectric values are defined based off of a vacuum). I don't know what kind of an effect it actually has though--could someone tell me if an actual vacuum (or close to it) is a special case? I'd like to see a lightning bolt in outer space (if possible) :).

As for decreasing the dielectric, what I do know is that means less leakage leads to less power use and therefore heat creation overall. The easiest way to take advantage to that is to simply increase the voltage and clock rate until you reach the previous power use. That scaling is certainly not linear (it's exponential I believe), so perhaps they are doing something else...

Re:Question

[moeinvt]

"we're not even talking about the dielectric of air here, we're talking about a vacuum!"

A vacuum has a constant of 1.00 and air is like 1.04. The extra savings is marginal when you're comparing it to '3'

"As for decreasing the dielectric, what I do know is that means less leakage"

Where do you get that from? Most of the leakage in semiconductors is from sub-threshold transistor operation, which has nothing to do with the wires. The other leakage is "gate" leakage, and to mitigate that, you want something with a very high dielectric constant.

Re:Question

[moeinvt]

If you're still reading (sorry, I was just getting back to my comments)

Well, "air" would be close to ideal from an electrical perspective. That would minimize capacitance between adjacent wires. Think of "capacitance" as a barrel that holds electrical charge. You have to fill the barrel with charge to turn a '0' into a '1'. Given that a transistor of a certain size can only pump so much charge, the larger the barrel(larger capacitance), the longer it takes to fill it up, which means that the chip runs more slowly.

Air isn't possible however because you need something with mechanical strength as well. The wires in a chip need to be done in layers, sometimes 8 or more. You do metal wiring on one layer, then fill in the spaces and cover it over with a silicide glass. You then do another layer of wiring and tunnel down through the glass(insulator) to the wires and transistors on the layers below through a vertical hole called a "via". It isn't manufacturable without some filler material to make a planar surface for the next layer of wiring. Imagine that done 8 times. You have 8 planes of this maze of metal wires a few tenths of microns apart, with a bunch of vertical pieces which span one or more of the planes. Even if it could be manufactured, the slightest jolt would annihilate it.

Mechanical issues also rule out many other materials with lower dielectric constants. Anything with a high "coefficient of thermal expansion"(quantifies expansion of a material with temp change) is out because expansion and contraction would bend and break the wires. So, the trick is to find something with a low dielectric constant, which has suitable mechanical qualities as well.

Nature's Orgy

[Khyber]

"Hydrogen merrily grabs other hydrogens, carbon loves a pair of oxygen, so and and so forth."

It's a regular atomic orgy!

fragility

[jstoner]

So these chips have holes in them... are they structurally weaker? Do we need to worry more about dropping devices made of them?

I suppose it's possible they might be stronger, but that would be a coincidence... they didn't do this for strength.

So, 30% faster and 15% more efficient ...

[constantnormal]

... but how much CHEAPER is it?

As semiconductor devices scale smaller and smaller, making features on the chips becomes increasingly expensive as we transcend the range across which past lithography-based fabrication techniques operate effectively. This seems like it might have the potential to produce some features at a much lower cost than via masks and multiple layers of etched handiwork. And it might have room to scale smaller while maintaining the same cost profile. So it's possible that the biggest percentage improvements might be related to cost.

Of course, all that will have to wait for actual engineering experience to know for sure, but it certainly seems like it might have potential in that area.