Maybe the principles of electron microscopy could be applied to CPU manufacture.

For decades electron-beam lithography was the wave of the future. When I was an intern (we called me a co-op then) at Bell Labs Murray Hill in 1971, my office mate at one point was working on design rules for a proposed electron beam lithography project. We oohed and ahed at the prospect of a one micron square contact, and the implied tens of ohms of resistance (for reference, three years later as a first-year design guy at Intel my contact design rules called for them to be drawn at 6x6 microns--I think they printed/etched larger, but don't remember for sure).

IBM even ran appreciable production with electron-beam litho mostly using it as a way to make one-off wafers with the desired mix of seldom-used dice (not a bad use by the way).

The killer problem is/was throughput. When the beam current gets high enough to get a useful writing rate, the charge of the beam itself is enough to spread the beam ("space-charge effect"). Getting the writing rate where you need it takes stunningly high modulation frequencies, and higher with each passing year as inherently parallel "optical" approaches relentlessly press onward. Electron-beam exposure is not even close to competitive now. And the optical has long since left the visible, inchig into UV, and threatening to get to extreme UV (a cheesy renaming of soft X-rays, rather the way the medical folks took the N out of NMR on the way to making us all comfortable with MRI machines).

Actually carving stuff takes even higher beam currents, so would have this problem in spades if it got past the other problems. It is done, for point repairs of prototypes (in a gadget called an ion mill), but I've heard no one before you suggest it as a serious manufacturing technique.

---

Yep, I remember not too long ago an issue of a popular computing magazine had a picture of a 486 chip on the cover with the big headlines "fastest computer ever - 50 MHZ!" and in the article it was saying that it would be hard to beat this speed!

I remember buying a Computer Shopper 15 years ago that profiled the 486 25mhz & 33mhz flavors and discussed some oem OC set-ups that were getting an unheard 66mhz! The article stated "more computing power than anyone could possibly need" LOL! I wanted one REAL bad, but was just a pennyless new college grad w/ a 286 that was little more than a typewriter. Spent more time playing old Wolfenstein and Wizzardry floppy games than writing papers ; )

---

Maybe the principles of electron microscopy could be applied to CPU manufacture. I think they coat the specimen then somehow get the electrons to discriminate features. Years ago the resolution of electron microscopes was about 10 to 20 angstroms (one to two nanometers). If it gets much better it'll be possible to see individual atoms. Maybe the electrons could be made to carve features and make the CPU.

I used to work for Rockwell International, next to their Newport Beach Fab -- a couple of decades ago. They were doing Electron Beam Lithography then.

---

I remember buying a Computer Shopper 15 years ago that profiled the 486 25mhz & 33mhz flavors and discussed some oem OC set-ups that were getting an unheard 66mhz! The article stated "more computing power than anyone could possibly need" LOL! I wanted one REAL bad, but was just a pennyless new college grad w/ a 286 that was little more than a typewriter.

Still got a couple of 66mhz chips in a cuboard somewhere. Doubt they work anymore-what would be the point i ask myself after wondering if they still work. :)
Recently dusted of an old amstrad cpc and that does work. Games seemed so good back then

Was it IBM or someone? used concentrated/focused magnetic field (i think) to move individual atoms into a word. Not very practical though i guess.

I remember buying a Computer Shopper 15 years ago that profiled the 486 25mhz & 33mhz flavors and discussed some oem OC set-ups that were getting an unheard 66mhz! The article stated "more computing power than anyone could possibly need" LOL! I wanted one REAL bad, but was just a pennyless new college grad w/ a 286 that was little more than a typewriter.

Still got a couple of 66mhz chips in a cuboard somewhere. Doubt they work anymore-what would be the point i ask myself after wondering if they still work. :)
Recently dusted of an old amstrad cpc and that does work. Games seemed so good back then

Actually, I have a couple of 66MHz machines that are still in service.

---

Was it IBM or someone? used concentrated/focused magnetic field (i think) to move individual atoms into a word. Not very practical though i guess.

That was using Scanning Tunnelling (electron) Microscopy. Individual xenon atoms were arranged by the STM on a graphite surface to spell out IBM. The STM was then used to scan that area to show the results.

Somewhat different to SEM (Scanning Electron Microscopy).

STM and the related AFM in the family of SPM techniques may yet prove useful for data storage and computation.

Regards,
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

That was using Scanning Tunnelling (electron) Microscopy. Individual xenon atoms were arranged by the STM on a graphite surface to spell out IBM. The STM was then used to scan that area to show the results.

Somewhat different to SEM (Scanning Electron Microscopy).

STM and the related AFM in the family of SPM techniques may yet prove useful for data storage and computation.

Regards,
Martin

Thanks

Was it IBM or someone? used concentrated/focused magnetic field (i think) to move individual atoms into a word. Not very practical though i guess.

That was using Scanning Tunnelling (electron) Microscopy. Individual xenon atoms were arranged by the STM on a graphite surface to spell out IBM. The STM was then used to scan that area to show the results.

Somewhat different to SEM (Scanning Electron Microscopy).

STM and the related AFM in the family of SPM techniques may yet prove useful for data storage and computation.

Regards,
Martin

This post reminded me of this interesting article.........

Theres also the one where they created microscopic working cogs (like those in a clock). I'll try and find it... Verry intersting thread this :)

[EDIT] not sure this is it but i'll keep looking
http://www.newscientist.com/article/mg17924074.900.html

[EDIT]http://www.memsinfo.jp/whitepaper/WP59twoPP.pdf

And about that liquid helium coolong- I don't think its to ridiculus to think OCers would love to try a basic refridgerated setup. Doubt its anymore risky than water cooling. All works on the same basic principle.

intel also says they are working on a 10 cores cpu and a few dozen cores cpu perhaps for racing with ibm nano chip tech.

i think 1thz is approachable in a decade. 10 years ago 66mhz-100mhz was cool speed.

---

Mandtugai!

i think 1thz is approachable in a decade. 10 years ago 66mhz-100mhz was cool speed.

1THz will more likely take ~20years, not ~10.

Look at your own numbers. 100MHz 10years ago, 2-3GHz now. => ~20-30x.
2-3GHz now, ~60-90GHz in ~10years. Not even close to 1THz.
Even this is way too high a prediction.

A more accurate way to predict is to extrapolate "Moore's Law".
CPU performance 2x every 18-24months.

10years= 120months =>
120/24= 5x, 120/18= 6.66x =>
2-3GHz -> 10-15GHz on the low side, 13.33-19GHz on the high side.
=> ~12.5-16.16GHz

...and remember, this is based on =all= of the performance increase coming from clock rate increases and not from any other technique.
Given that the entire semi-conductor industry is presently doing everything they can to get performance increases =without= having to increase clock rate, the low 10year projection, ~12.5GHz, is probably still too high.

~6.25GHz is probably closer what will be the "standard" clock rate in ~10years time {This ignores the overclocking fanatics...}.

---

I have to agree with Eric here.

In fact, CPU manufacturers have been hard pressed to ramp clock speeds as much as used to be the case.
This is due to us getting close to the limits of what's physically doable. Much smaller and we'll be down to a few atoms per transistor.

So rather than pump in ever more juice to be able to get that little bit more in raw GHz, they're going a different way. Not higher clockspeed, but higher levels of parallelism - hence dual-core and quad-core and xxx-core CPU designs as well as going back to having a specialized processor for every subsystem vs. doing it all on the CPU.

Look at the "Cell" CPU that the PS3 uses - it's a technologically comparatively simple piece of chippery compared to, say, any recent P4 or A64. That isn't to say it's worse in any way, it's just different. It can do 7 things in parallel - not very complex things, but that's where your 8th core comes in that takes care of ordering things and keeping the other chips fed well.

This is a radically different concept. At a frequency of 3.2 GHz, such a machine has a theoretical max compute power of (AFAIK) 200 GFlops (Billions of floating point operations a second). A typical 3.x GHz P4 or quick A64 has less than 10 (less than 5 even I believe).

So GHz ain't everything, it's what you do with them.

Sun also just premiered a new platform design that uses a massively parallel chip that's not clocked very high but can achieve amazing performance given the right sort of work.

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

In fact, CPU manufacturers have been hard pressed to ramp clock speeds as much as used to be the case.
This is due to us getting close to the limits of what's physically doable. Much smaller and we'll be down to a few atoms per transistor.

To be more precise, the limits we are reaching are those that is doable
a= w/ photo-lithography, and
b= w/ the materials we currently use

We've been using 193nm based photo-lithography for far longer than the industry thought we'd be able to, and we still have a few tricks like EUV and immersion left, but for the most part PL is definitely in the region where incremental improvements are going to cost exponential amounts.

Same for traditional doped Si based ICs.

We are =no where= near the limits of physics, or even the limits of Material Science. We simply need to use new materials and need an economically sound manufacturing method for them.

Once we transition from PL to nano-assembly, all of the present "limits" will look quaint in retrospect.

---

when pll 200mhz was speed of the time intel sayd they were reaching 455mhz in the lab in late 90's. but too soon it reached 2-3ghz.

but now chip industry is just coming on the edge of nano tech not even sufficient nano tech mechanisms and complexities put into production and development practice and they are acheiving already .5thz. so i would say 1thz is very probable. i am not sure whether Moore's law concerned about nano tech development factor.

there are more nano level construction methodologies expected to be discovered and launched into chip production practice and many unknown possibilities in it.

---

Mandtugai!

Sorry, Eric,

I should have stated "close to what's physically possible using Silicon".

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Sorry, Eric,

I should have stated "close to what's physically possible using Silicon".

Wasn't intending to pick on you or sound harsh, just increase the precision and clarity of the discussion. :-)

Some leap to extreme conclusions given the least amount of imprecision or provocation. This has been particularly true in the IT industry and press lately about "the death of 'Moore's Law' ".

...so I like to do what I can to keep things accurate.

---

A year or so ago I saw one transistor per square micron on CPUs. I took the area of the chip and divided that by the number of transistors to get that value. Now I can conclude two or three because of progress. That looks to be the size of a wavelength of visible light. Still it is called 65-nanometer technology - that might be the closest two wires can be without causing a short, I don't know. I don't know what Intel and AMD are using to make those patterns - are they using light? If so maybe they could use UV or soft X-rays, if special lenses could bend them like light. The Chandra X-ray satellite focuses X-rays by means of grazing reflections - don't know if something like that could be practical with CPUs or not.

---

...it is called 65-nanometer technology... I don't know what Intel and AMD are using to make those patterns - are they using light? If so maybe they could use UV or soft X-rays, if special lenses could bend them like light. The Chandra X-ray satellite focuses X-rays by means of grazing reflections - don't know if something like that could be practical with CPUs or not.

IC's are presently made using 193nm light and a process very much like the Chandra grazing reflection idea.

The amazing thing is that we are etching 65nm features using 193nm light. Evidently the industry thinks we are going to be able to do this for 45nm and possibly even 32nm as well.

At some point we will have to start using shorter wavelengths of light, possibly even EUV (Extreme Ultra Violet). We also have a technique called immersion waiting in the wings.
These tricks will probably allow PL to get down to ~11nm or so.

However, we are getting to the point where PL is getting more and more expensive (or just plain hard to do) for each improvement in scale. That means the successor to PL has to be found and made economically viable.

The indications are that the best candidate presently for that successor is nano-assembly: instead of etching IC patterns onto wafer substrates, we literally build or grow IC's one atom or malecule at a time.
Nano-assembly will be as big a revolution in semi-conductor manufacturing as PL was before it.

---

...it is called 65-nanometer technology... I don't know what Intel and AMD are using to make those patterns - are they using light? If so maybe they could use UV or soft X-rays, if special lenses could bend them like light. The Chandra X-ray satellite focuses X-rays by means of grazing reflections - don't know if something like that could be practical with CPUs or not.

IC's are presently made using 193nm light and a process very much like the Chandra grazing reflection idea.

The amazing thing is that we are etching 65nm features using 193nm light. Evidently the industry thinks we are going to be able to do this for 45nm and possibly even 32nm as well.

At some point we will have to start using shorter wavelengths of light, possibly even EUV (Extreme Ultra Violet). We also have a technique called immersion waiting in the wings.
These tricks will probably allow PL to get down to ~11nm or so.

However, we are getting to the point where PL is getting more and more expensive (or just plain hard to do) for each improvement in scale. That means the successor to PL has to be found and made economically viable.

The indications are that the best candidate presently for that successor is nano-assembly: instead of etching IC patterns onto wafer substrates, we literally build or grow IC's one atom or malecule at a time.
Nano-assembly will be as big a revolution in semi-conductor manufacturing as PL was before it.

In only 5 short years, we'll be 2 and 3 generations ahead of todays 65nm process w/ 32nm being common place and 22nm coming online. To put this into perspective, .13 & .18 micron processes were only 2 and 3 generations ago. These generations of chips will be almost unblieveably powerful compared to what we know today as a 22nm-based die will be able to fit perhaps a couple dozen cores in an essentially, massively-parallel CPU. If heat can be controlled as effectively, imagine a 24 core cpu running at 6-8ghz, not to mention the likely scores of improvements in cache architecture, bandwidth, pipeline, execution, threading, latency etc. Certainly at least an order of magnitude different from comparing a 65mn, 2 core Conroe Extreme vs. the .18 micron 1.33 G4 writing this comment. 5 years is just around the corner.....
For those interested in more detailed reading, here's a few PDFs I found readily available online.

http://homepage.mac.com/gecko_r7/.cv/gecko_r7/Sites/.Public/Intel%20Tech%20Presentation%205.06.PDF-zip.zip
http://homepage.mac.com/gecko_r7/.cv/gecko_r7/Sites/.Public/Lithography%20Data%2004.PDF-zip.zip
http://homepage.mac.com/gecko_r7/.cv/gecko_r7/Sites/.Public/Lithograpy%20Roadmap.04.pdf-zip.zip

---