Intel does not use SOI, but IBM does

Massive computational power on the cheap! What's becoming the limiting factor is our ability to use it. I'm not just talking software, I'm talking "wetware." I see conventional cryptography becoming obsolete thus obviating a practical implementation of quantum physics cryptology.

Moore's law ends when feature size equals atom size. I don't think we'll shrink things much below that.

So...how do we invest in this company?

I think that was by 2030 we would augment our intelligences by 10,000 and in 15 years (~2020 ) computers will have the power of 1 human brain. It will be like the greasy taxi driver on MTV said sometime back..."They do that thankin' for ya"

Way to go - the Brits save the world - again ! ;-]
(only joking..............)

Way to go - the Brits save the world - again ! ;-]
(only joking..............)

Don't put all of your money in Mears.

If true, the capability to be used sub-10nm is impressive...but there are likely other R&D efforts that will do just as well.

60% nmos/80% pmos is not viable competition for high-k in the near term...at IEDM 2007 intel demonstrated a 96% reduction of CMOS and a 99.9% of nmos gate leakage. IN all likelihood, they will have a competitive technology for mears when we get to sub-10nm.

see, for instance:

http://maltiel-consulting.com/IEDM_2007_Highlights_maltiel_semiconductor.htm

These results are for penryn's HKMG process, which started shipping in november. I meant pmos and nmos transistors, not cmos and nmos.

see, for instance:

http://maltiel-consulting.com/IEDM_2007_Highlights_maltiel_semiconductor.htm

These results are for penryn's HKMG process, which started shipping in november. I meant pmos and nmos transistors, not cmos and nmos.

What about the concept of using (man-made) diamond, instead of silicon, to make CPUs?

I recall an article from 2003 which discussed new ways of making synthetic, yet perfect, diamonds ... now popularly called "cultured" diamonds. Two outfits doing this are Apollo Diamond, and Gemesis.

Anyway, the article stated that such cultured diamonds, if cut into tiny wafers, could replace silicon for CPUs, and withstand heat that would usually cause silicon to melt.

Has this gone anywhere? Or is it secretly going somewhere?

(article link)
http://www.wired.com/wired/archive/11.09/diamond.html

Always thought it was interesting that Moore's observation of only a couple of data points describing the past performance of the chip industry has become a whip to beat it towards its future performance. You know when I look back I realize that the days are shorter -- no, wait -- I didn't say that!!

Good article today. I was waiting to learn when Google was going to use MST-layered technology to finally make possible cool-able microsatellites that Team Cringely could launch into low-earth orbit to provide their worldwide free 700MHz data service (for iPhones), but I came to the end unfulfilled.

I'm just being snarky. BTW, the double-posts come from PBS reporting a posting error, but the messages really do get through. Check before posting again, folks.

There's actually someone (besides me) who remembers the PDP-8?

Great article, Bob. Highly entertaining, very informative, and it is fascinating to learn something about an area of technology I have not been paying attention to for the last three years.

Great article Bob! nice to read something actually about technology, and not Apple/Google/Microsoft conspiracy theories ;-)

Erm, if it means I can download and watch porno movies in HD then I'm all for it! :-)

Doubt we'll have robot bosses, as although the hardware will be capable, the software will not. A.I. has underdelivered so far.

Speaking of A.I., did you hear about Chris McKinstry's suicide a while back? He did the Mindpixel project. If you ask me, I think Mindpixel was just a scam to get people to work for free and then fake his own death so that the fruits of the project could be passed along to his buddies -- data inputted by volunteers under the impression they were to be the recipents of Mindpixel Corp shares... I won't believe Chris is dead until I see the body brought back from Chile to Canada (where there will be several people waiting to piss on it.)

Thanks, Bob. Your article makes me want to move to Waltham, Mass. Mears probably has a couple dozen more nifty tricks up his sleeve -- a modern day Edison.

I'm looking forward to that Kurzweil world, believe it or not, when computers of today's desktop computing capability can be injected into my bloodstream to clean the plaque off my arteries in my old age. Boy oh boy won't that be fun! This was an informative article, now get back to the Apple/Google/Microsoft conspiracy theories!

- Pete O

I think Moore's Law will grind to a halt when the average desktop computer requires more power than what a dedicated household electrical circuit can deliver. Given that the top-end machines right now are sucking over 12 amps out of the wall we're closer than you think. If booting pops a breaker there isn't a lot of computing that gets done.

--chuck
http://chuck.goolsbee.org

Moore's law for only another 15 years? This is based on the assumption that chips have to remain flat (2-dimensional). However we are already seeing moves into the 3rd dimension, f.e. chip stacking.

New technologies, possibly including things like photonics and nanorods, will undoubtedly keep Moore's law alive beyond your projected 15 years.

-- 2.718

Funny, the work vs area = heat density issue also happens in scanning electron microscopy. You want to see a smaller area? Make the raster on the sample smaller - but beam current remains the same, so more heat is generated since the smaller area presents more resistance. Even with metal plating (to allow conductivity/backscatter) the heat is savage. Luckily the new low-current scopes have gotten around this. The IC industry has definitely driven development in the microscopy field.

Moore's Law is pretty irrelevant in real world applications with the exception of multimedia editing. While multicore CPUs may allow the Intel's of the world to stay on the curve, the ability of programmers to create S/W that can run truly parallel has not fundamentally changed in 20 years. [And I mean more than just forking a couple of threads here and there]



This is the true Achille's heel for at least the next decade. How do we utilize the 80/128/512 core processor? It's fundamentally a hard problem which has not become much easier with the current generation of programming paradigms. Frameworks like NVidia's CUDA or Cg (and others) helps a little but to properly exploit parallelism requires deep understanding of the underlying data dependencies and that is not getting any easier.

"How do we utilize the 80/128/512 core processor?"_______________________________________________________________________________

I think you are looking at the wrong tool to solve the right problem . I needs to be addressed at a different level .

There should be a corrollary to
Moore's Law: After age 50, human minds begin to deteriorate at a rate inversely proportional to the rate of Moore's Law, resulting in an ever-increasing gap in the ability of elderly humans to comprehend computing power.

Spencer wrote:How do we utilize the 80/128/512 core processor?
--

Spencer, you know that your niftly little laptop today is running 30-40 processes right now, with no real thread dependencies, right? I think I remember seeing a pretty standard windows workstation a few years ago that had 100+ threads running.

We already have simultaneous processing occurring in modern computers and operating systems.

A PDP-8 was the first computer I laid hands on at Berkeley in 1975. Booting up meant flipping a long row of switches on the front panel in a particular sequence to direct the computer to load more instructions from the rack mounted DEC tape drives. We got a multi-thousand dollar upgrade to boost the internal core? memory from 2K to 4K in order to reduce how often the PDP-8 had to swap data on and off the tape drives. Those high speed inch-wide memory tape drives really limited how much we could accomplish compared to the big iron priests with their spinning platters.

Another new technology should allow designing Peltier coolers into the chip itself: http://arstechnica.com/news.ars/post/20080110-startup-shrinks-the-peltier-cooler-and-puts-it-inside-the-chip-package.html



Probably not as ground breaking as a new fundamental way to bake the sand but surely a good stopgap until the next breakthrough.

The front bus of a computer has been 800Mhz for several years now (the bus is shared by all components of the mai board), therefore the Moore's law has ended a few years ago already.

This article is totally bogus.

Long time reader, first time poster-- This is way off topic, but:

Why are there so many (presumably accidental) double posts in the Comments? Bob, you're smart enough to write a bulletin board that would prevent this...

"I'm looking forward to that Kurzweil world, believe it or not, when computers of today's desktop computing capability can be injected into my bloodstream to clean the plaque off my arteries in my old age"

Just hope they are not running Windows!

________________________________

"Stef" is right; bus speed and other bottlenecks are a problem.

"The front bus of a computer has been 800Mhz for several years now"

Let's see, my computer purchased last year has a 1.3GHz FSB, and its recent replacement has a 1.67GHz GSB. In any event, Moore's "law" has not progressed evenly across all components. Memory speed lagged way behind for a time.

If I were going to make your argument, I'd make it based on processor. A bleeding edge computer five years ago was 3GHz, today it's 3.6GHz, and core-for-core probably only 50% faster. All the multicore shenanigans today could have been employed five years ago, but weren't.

Stef, Tom B, what part of "on-chip" CMOS didn't you understand?

The 800MHz front-side buss is on the board and it can (and will) be kicked up in speed as required.

Moore's law is not yet at end.

My left arm is longer than my right, therefore this article is *not* totally bogus. See what I did there? I made absolutely no sense - just like some other posts here.

Moore's Law has nothing to do with bus speed - it is a statement of the doubling of transistor counts on a CPU every 2 years (or so). Not speed, not bandwidth - just transistor counts.

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

In any case, my bus speed is 1600mhz; I have a 3.2ghz 8 core Xeon Mac Pro... if we *were* talking about bus and memory speed, or system speed in general, I will have to say that my computer definitely is at least twice as powerful as one of about the same price from January 2006 would have been.

@Stef: at the cost of sounding pedantic, may I suggest you read the following Wikipedia article carefully: >,
BEFORE deriding a whole article about ONE of Moore's Law various interpretations. The only thing that seems bogus here is you as a "intelligent reader" and a supposed "Expert".

@stef
sorry: http://en.wikipedia.org/wiki/Moore's_law

First of all, as a lithography engineer, I'd like to say that the lithography "problem" that you cite as "solved" is, in fact a major hurdle. There are significant restraints to moving into the 32nm node, and virtually no roadmap to go any smaller.

I haven't researched this MST technology you cite, but at first glance it appears to be way to selectively increase carrier mobility in the lateral direction in the channel of the MOSFET. While this is a novel idea, personally, I think that it's just another piece of duct tape that we in the semiconductor industry have applied to the MOSFET structure. We've been tweaking the processing so much in the past 20 years, and all it does is add process steps and make things more expensive.

I think that we are due for a fundamentally new device. There is some interesting work being done at Notre Dame and Rochester Institute of Technology on "tunnel transistors," which consume much less power than MOSFETs. It just might be that new device.

"How do we utilize the 80/128/512 core processor?"

Run a distributed computing project on it! That will make your purchase worthwhile!
(i.e. Folding@home, Seti@Home, climateprediction)

Not only is Dr Robert Mears very clever, he is also very attractive.

We get faster and faster CPU's, more storage etc, but we are all still doing the same things with our computers that we were doing 15 years ago. Sure games have come along in leaps and bounds, but why do we need 4,8,10 cores in our cpus when most of us are simply word processing or playing with spreadsheets?! It would be nice to see actual benefits of this increasing computing power.

What does this kind of processing power make available for us in the future?

What kind of applications will become possible with this kind or computing power?

What kind of innovations in operations have to occur to achieve these things?

Maybe the things we envisioned with neural networks and intelligent systems will suddenly be possible by different means; that is, allowed by pure cranking power.

Maybe we'll have house maids that can clean house, do laundry and cook food (but little else), per the Jeffersons - (as long as your clothes is put in a standardized hamper). That no longer seems that far out there. And that's change big enough, because it releases time to people to work and play more.

It seems like the action will be in military, energy and materials areas. Military, because that's the way it's always been; Energy because if we don't develop more sophistication here we go back to the dark ages; Materials, because like energy, it has a finite quality to it.

Thurstan Johnston : there are more CPUs than the one's under your desk.

You can think of it as more transistors in the same space using similar power to now or as the same amount of transistors in a smaller space using less power.

About a decade ago, some writers of an article in Scientific American made a shocking point:

What happens when fab labs are too expensive to build?

Wikipedia (ya, the one pundits loathe to quote) says it costs 1 to 4 BILLION dollars to build a new semiconductor facility.

http://en.wikipedia.org/wiki/Fab_(semiconductors)

Moore's law could come screeching to a halt not because the technology isn't there, but because nobody wants to come up with the money to make it happen.

BTW, the writers in the Scientific American article were somewhat prophetic. The claimed manufacturers would start coming out with different flavors of processors (e.g. 2-core, 4-core). This is much the same way automobile makers, (once everyone agreed there would be automobiles made), created different colors and styles.

Live long and prosper, Bob.

Your reader

Infogleaner

While the cutting edge of powerfulness is interesting there is another place where Moore's law will be closely followed.

How much computing can you get in your pocket for $200?

Some of the technologies mentioned above will be relevant here but the constraints on heat dissipated (Is it too hot to hold?) and power consumed (= battery weight/life) are much tighter.
Personally I expect PDA/Mobile phone/UMPC/music players to be the cutting edge as PC functions migrate down to be incorporated into these.

Soon these devices will have the power to support handwriting and voice recognition and natural language understanding and new ways of interacting with humans and intelligent objects.

If I was Google I would be investing in:
* Prosthetics (these will eventually lead to implantable computers.)
* smart cards (transit authorities are investing in smart cards for pay as you go fare payments. The EU is trying hard to impose one standard for these - this will be an ID card cum cashless payment card usable throughout Europe)
* Wimax in the third world, (with local partners who can sort the politics, the hardware, the billing but world wide partnering via big G).

Moore's law continues but at what expense? I see no discussion of the exponential increase in complexity of software algorithms that is required to take advantage of increased core complexity over time. Have you ever tried to optimize for a cache or instruction pipeline that is 11 stages deep? Every try to parallelize or software pipeline an algorithm that that does not naturally partition that way? Where are the tools that make it easier to program massively parallel machines? They are lagging by decades! The dirty little secret is that although Moore's law is moving along at an exponential rate, the means to leverage the performance is trickling along at a mere linear rate. I'll create a new law: "Bram's Law". It is the difference between Moore's law and the "true" performance level that can effectively and practically be achieved. This gap is increasing every year and frankly eroding the relevance of Moore's law.

Moore's law continues but at what expense? I see no discussion of the exponential increase in complexity of software algorithms that is required to take advantage of increased core complexity over time. Every try to parallelize an algorithm that that does not naturally partition that way? Where are the tools that make it easier to program massively parallel machines? They are lagging by decades! The dirty little secret is that although Moore's law is moving along at an exponential rate, the means to leverage the performance is trickling along at a mere linear rate. I'll create a new law: "Bram's Law". It is the difference between Moore's law and the "true" performance level that can effectively and practically be achieved. This gap is increasing every year and frankly eroding the relevance of Moore's law.

"So...how do we invest in this company?"

When they go IPO later this year, make sure you buy some shares.

I, for one, welcome our Skynet overlord.

I enjoyed the article immensely. You touched on something I was wondering about since I read about Intel's Hafnium high K developments going into the Penryn CPU. I am of the opinion that multi-core CPUs may be an evolutionary dead end.

Significant software investment is required to make these cores live up to their promise. I do not see any company making a credible attempt at defining a new architecture where many cores are useful. Instead I expect companies like Intel to press their technical advantage in process technology, meaning they may abandon the multi-core approach
and get back to raising clock rates again.

Could you put it to a poll of your readership?
The question would be something like:
Do you think multi-core processor trend will continue unabated or will design trends revert to 1 superfast core. Or perhaps there is an optimum number of cores?

Forwarded this article link to some of our Process Reliability Engineers (that study this sort of thing). One of them indicated that it sounded a little "too easy", but said that the technology sounded interested. He fed this link back to me that had more MST info. http://sst.pennnet.com/display_article/296979/5/ARTCL/none/none/1/Tackling-power/performance-trade-offs-with-silicon-channel-engineering/

"Significant software investment is required to make these cores live up to their promise. I do not see any company making a credible attempt at defining a new architecture where many cores are useful. Instead I expect companies like Intel to press their technical advantage in process technology, meaning they may abandon the multi-core approach
and get back to raising clock rates again."

I agree with this with caveats. The architecture and software is here, ... just not widely distributed yet. For example, the Haiku OS and BeOS are capable of scaling to at least take advantage of four-core systems. However, BeOS is in legal limbo now and Haiku is very obscure.

Similarly, to do good multi-threading, you really need to at least use some sort of dynamic message passing (Objective-C, which both BeOS and OSX/Cocoa uses) and a functional langage. The latter is no longer taught in depth in mainstream CS programs, and difficult to comprehend when coming from an imperative language background. Which most programmers use to make a living (including me). Erlang, for example, is suitable for this kind of app, but the syntax looks like obfuscated Prolog .

Having said that, I do enjoy having a dual core system. My Linux box might not be pushing both cores to its limit now, but it makes better use of it than say, my Windows box. If MST does make it easier to shrink cores, I can see Intel and AMD going back to pushing for higher core speeds ... but I hope I can still get n-core processors.

Wow. That. Was. Awesome.

What a great read. Thanks - pitched just at my level of understanding.

Too many cores will bottleneck at the memory access point, unless they have tons of cache - in which case they might do distinct tasks better but not be as much help to each other in pipelined tasks like video processing. Eventually they interfere at the disk or device level. And eventually the OS has to spend a lot of time organizing resources.



Some purposes will always need more horsepower; if each video filter and compression stage is another core, and data motion is somehow minimized (cache bank switching?), of course more is better. OTOH for normal people spending most of their time netsurfing and emailing, we already have massive overkill. For normal applications "more RAM and cache" gets better bang for buck than "more CPU cycles". (remember that the readership of this list is tilted toward power users, not normal people.)

Too many cores will bottleneck at the memory access point, unless they have tons of cache - in which case they might do distinct tasks better but not be as much help to each other in pipelined tasks like video processing. Eventually they interfere at the disk or device level. And eventually the OS has to spend a lot of time organizing resources.



Some purposes will always need more horsepower; if each video filter and compression stage is another core, and data motion is somehow minimized (cache bank switching?), of course more is better. OTOH for normal people spending most of their time netsurfing and emailing, we already have massive overkill. For normal applications "more RAM and cache" gets better bang for buck than "more CPU cycles". (remember that the readership of this list is tilted toward power users, not normal people.)

Moore's Law is all well and good, but let's face it: energy requirements and other issues will prevent the full extent of Moore's Law from being used.

Others already touched upon the energy costs, and with both energy requirements for computing as well as energy costs themselves both going upward, the usage of these chips in the future will be out of reach of most consumers. The ever increasing demand for energy will eventually sink all of these energy sucking devices, unless we make a massive switch to nuclear energy (unlikely) or something such as cold fusion ends up working (equally unlikely).

Assuming my math is correct...

15 years * 12 months / 18 month cycles = 10 Moore doubling cycles.

So taking a good desktop computer to be 4ghz, in 10 cycles, or 15 years, it would be 2048ghz.
That's an obscene quantity of processing power, even with inefficient software. I cant wait for the opportunities it unlocks!

Well, Theron, we could double ten times a different way and get 2000 processors with the product line differentiated by clock speed from 1.6 to 3.4 GHz. This would be on your typical desktop machine or laptop.

Or we could double the number of chips we get from a wafer(?) ten times and get 1000 dual core chips from the same amount of silicon we get 1 from today. It probably wouldn't cost 1/1000th as much (each one would still have the plastic box around it) but would be a lot cheaper.

What good is Moores Law if Windows keeps failing its Customers.

A major Drug company upgrades to Windows XP and a newer MS Office spends $90 Million dollars and now no one can put out a Word Document. If they can create the document they can not print it out. Basicly One week ago the companies computor system sort of worked. $90 Million dollars and one two days later they have an upgraded system that does not work at all. This is a SEAMLESS UPGRADE??

My sons new laptop which is not really hooked up to the Web now takes several minutes to come up after a couple months of use.

In two months my sons laptop turned into a useful tool to a tremendous time waster just to get it turned on.

I fight the same problems myself with my own laptop. I have no help desk to call either.

All this "progress" I believe is really slowing us down.

We upgrade to MS Office 2007 and could not seemlessly makes a new Document because all the icons changed. This is not progrees.

Can we apply Moores law to improve this?

What good is Moores Law if Windows keeps failing its Customers.

A major Drug company upgrades to Windows XP and a newer MS Office spends $90 Million dollars and now no one can put out a Word Document. If they can create the document they can not print it out. Basicly One week ago the companies computor system sort of worked. $90 Million dollars and one two days later they have an upgraded system that does not work at all. This is a SEAMLESS UPGRADE??

My sons new laptop which is not really hooked up to the Web now takes several minutes to come up after a couple months of use.

In two months my sons laptop turned into a useful tool to a tremendous time waster just to get it turned on.

I fight the same problems myself with my own laptop. I have no help desk to call either.

All this "progress" I believe is really slowing us down.

We upgrade to MS Office 2007 and could not seemlessly makes a new Document because all the icons changed. This is not progress.

Can we apply Moores law to improve this?

Could you please add to your ground rules, that persons should push the post button once and once only?

It is amazing that every time I look here, there are multiple posts everywhere. Very annoying. Maybe you should upgrade your software to remove them at source?

Thank you, and for some top presentation of ideas, always.

It sounds to me that MST may just have solved many problems. Adoption by the industry sounds iminent. Its great to see the big guys beaten to the post!

Well, I see why they do it. The multiple postings. I take back all my uncharitable thoughts.

Here is what happens.

- One posts, enjoying the ReCaptcha Captcha and its help in digitizing mediaeval books.

- interminable wait follows, from pbs.org.

- a minute or so later, you get a screen with just what is below. All fails, it says. Proxy, it says. Remote server problem, it says.

- My thought is that it tried to check back through my router, and got blocked. Or, it tried to check an aol.com email address I use to avoid junk bother, and was refused on that.

- only if you then hit the previous button on web browser do you see that your posting actually did occur.

Hope this helps to fix it. Message in all its beauty follows. I will now hit the submit button again.

N.

Proxy Error

The proxy server received an invalid response from an upstream server.
The proxy server could not handle the request POST /mt/cgi/comments.

Reason: Error reading from remote server

Do not under estimate 'Mears Technologies' and their work. This start up company is one to watch for the future. Their silicon work, discussed in the article, is just the tip of the iceberg. This is just the first of their commercial ventures, which will include solar and spintronics.

Dr Robert Mears is a recognised expert in his field and a brilliant scientist. Only in his mid 40's, I have no doubt that his best is yet to come.

Mears is presently a privately funded company which will be looking to go to an IPO later this year. When it does, make sure you grab a slice of this potentially huge cake.

Remember, you heard it here first.

If my desktop computer ever becomes that smart, it will surely want a more fulfilling existence than to sit on my desk waiting for me to do something with it.

Dr Robert Mears and his team appear to have sneaked in under the radar and surprised everyone.
The guy made the net possible.... is this even better !

Come on, Bob. Write a real technical version for those of us in (or formerly in) the industry. I'd like to have a good discussion on scaling, voltages and the physics really involved.



This week's pulpit has enough that isn't in it so that I can't decide whether the inconsistencies I see are your fault or that I've been out of the technology too long.

Straka - If you want some 'proper' technical info why not contact Dr Robert Mears direct. I've seen that his contact details are on the Mears website. He's obviously grinding the organ, the monkeys just dance.

Straka - If you want some 'proper' technical info why not contact Dr Robert Mears direct. I've seen that his contact details are on the Mears website. He's obviously grinding the organ, the monkeys just dance.

Wurzle:


Valid point, but each of us contacting Dr. Robert Mears doesn't scale well. I thought that Mr. Cringely's purpose in this blog was to provide such scaling opportunities.


I wasn't criticizing the Mears issues, just questioning some of the limits being proposed in the original article here. E.g., in CMOS, when you scale down, current is mostly proportional to the voltage swing times the capacitance of circuit nodes, and capacitance goes down by the square of the scale. I'll grant that local temperature is governed by the current divided by the area of interest, and may therefore be a wash. This is the kind of discussion that I think would be helpful.


And the concept of "gate leakage" is kind of odd. Any kind of "gate" leakage would seriously impact circuit operation. (Oh, I would also think that "gate leakage" is not a function of x-y scale, but of gate thickness and material - an entirely different consideration.) On the other hand, junction leakage could be significant and (if designed properly) would not seriously impact circuit functionality. That's where technologies such as SOS and SOI have been useful.


That's where my 25-year-out-of-date experience in fab runs out (I used to do EPROMS (2708s - VERY old stuff) and gate leakage and the Arrhenius equation was very important in data retention calculations), and I would like an illuminating discussion. We don't need to go to Mears for that.


VRY,...

No, Moore's Law ends when computational power for the dollar stops doubling every 18 months. This has nothing to do with the size of an atom.

Robert Mears for president!! Who needs a woman or an ethnic minority when you have this awesome Brit tempting us with his silicon titbits like the child catcher from 'Chitty Chitty Bang Bang'.

I'm a little tired of all the patter,
Can Bob Mears deliver the matter.
Is he gonna put Intel down
Or maybe he's just a struttin clown
Only one way to find out
Match the two in an even bout !

I'm a little tired of all the patter,
Can Bob Mears deliver the matter.
Is he gonna put Intel down
Or maybe he's just a struttin clown
Only one way to find out
Match the two in an even bout !

I'm a little tired of all the patter,
Can Bob Mears deliver the matter.
Is he gonna put Intel down
Or maybe he's just a struttin clown
Only one way to find out
Match the two in an even bout !

Bob said: "This extra chip heat comes generally from four sources. The first is simply reduced surface area; yes the voltage is lower, but if the ratio of old voltage to new voltage is less than the ratio of old surface area to new surface area from the previous product generation and manufacturing process, well then the chip simply has to get hotter, since it is dramatically smaller yet doing the same work. Voltages drop linearly while surface areas decrease as a far more rapid square function."

Not quite, Bob. Heat is a function of the POWER dissipated, which in turn is a function of the voltage powering the chip. Ohm's law tells us that power is voltage SQUARED, divided by resistance/reactance. So while voltages drop linearly, the power drops with square of the voltage drop FOR THE SAME LOAD, so lower voltage is a square function just like lower surface area.

What's happening is that the number of transistors is also increasing while the surface area and voltages are decreasing, so the load is not the same. This results in lower resistance/reactance, causing the power dissipated in the chip to rise, causing the heat to rise.

Right then chaps listen to me.
I think you'll find that Bob has sorted the heat issue as well.

Right then chaps listen to me.
I think you'll find that Bob has sorted the heat issue as well.

Williams man - your rap is crap
Goddam, I don't even think you're black.
This Mears dude, he aint no fool,
He's the daddy with the 12" tool.
Spread his silicon all over my chips,
He should be in LA, hangin' with the Crips
Instead of hidin' in Waltham MA,
With his white coated geeks in the lab all day.
Mears man, I can feel,
You're really close to that licensing deal
After that it's IPO, AIM, NASDEQ - way to go.
Later on, celebrating in my yacht, I'll be sipping tequilla whilst Intel rot.

Listen Dog you're way outta line
This Bobby Mears ain't got the time
To mix it with Intel and all the boys
They too pack impressive 12 inch toy
So bring it on blood man to man
And watch as Intel hide his can!

Who's watching the FCC auction?
Google and Verizon are playing chicken - seeing who can bid the most without for C block without going over the reserve price - what do they think this is, some punk ebay auction? What are the chances of Google sqeaking in with the reserve price at last second?

Hi - Nice article. It might be more helpful to think about the current rather than the voltage though. If you are using a bucket of electrons to represent a digit, then increasing the clock speed means that you need to empty and fill the bucket faster. This means more current, and you are then hit by the heat being proportional to I^2R, the current squared times the resistance. Note that it is the current SQUARED. Lowering the voltage means that you can use smaller buckets, and also means that there will be less leakage as you say.

Ciao

Steve

clock speed and current are all the same
low heat and dollars the name of the game
MST wafers are cool and juicy
Mears and his team have done it you'll see

If the Mears team can deliver their multi adoptional technology then why isn't the industry rushing to impliment it NOW