Slashdot Mirror
NTT Verifies Diamond Semiconductor Operation At 81 GHz
Anonymous Coward writes "This story over at eetimes.com reports of a semiconductor made of diamond that is able to run at 81 GHz." Mmmm, foreshadowing.
← Back to Stories (view on slashdot.org)
Edit: I should have read the 'article' on slashdot instead of just clicking the link. Sigh, I'm an idiot :(
Vacuum tubes are still used as the final amplification stage for TV and radio broadcast transmitters. They're the best thing able to handle the power efficiently, even today. Try building a semiconductor transistor with a gate width measured in centimeters (compared with microns); it's tough.
81GHz is the switching speed of the transistor, not the processing speed of a resulting PC. Some of the reasons are:
* CPU's perform a large number of transistor switches in a single clock cycle.
* The rise/fall response time must be much smaller than the switching time.
Don't be uninformed...oh wait this is slashdot. Vacuum tubes are still used in RF broadcasting, especially digital TV because the are able to reach the power levels necassary to broadcast a 50kW radio signal at low enough distortion to cleanly transmit the digital signals.
This lengthy article gives a fascinating history into how the DeBeers cartel has created artificial scarcity in the diamond market and convinced the western world that a "Diamond is Forever". Before the 19th century, no one ever had to spend 6 weeks salary on an engagement ring!
Even today, high power RF amplifiers (as used for terestrial radio and television broadcasting) are usually built around vacuum tubes. We are talking about >> 10,000 watt RF amplifiers - something like 30,000 to 50,000 watts for a typical commercial FM radio station.
These very high power RF amplifiers are difficult to build with (current) solid state electronics due to the limitations of amplifying bandwidth and power handling. The article is just suggesting that diamond semiconductors could tackle such high frequency, high power applications.
There are some really great uses for vacuum tubes. Here's a couple:
1) High quality audio reproduction. Any home audio freak will tell you nothing sounds like a sweet tube amp. There is both anecdotal and scientific evidence for the superiority of tubes versus semiconductors. Why then do we use semconductors as audio amps? Price and size. For a home theater amp, semi's cost anywhere from $100 to $900+, and tubes cost anywhere from $500 to $20,000.
2) High frequency amplification. Good for rf transmitters. They have many other high frequency uses as well.
Don't discount the tube!
You can't legislate goodness. Let each to his own destiny, by will of his freely made choices.
DeBeers is shitting a brick over it too, because that means its nearly impossible to tell a diamond from the ground from a lab one, except the lab one is even purer. The good part of this is the tech industry has far more muscle and clout than DeBeers does. DeBeers is truly an evil company sown on the blood of africa and putting them out of business would do the world a favor.
In fact, the only way for this technology to become realistic is for large scale lab diamond growing like I mentioned above. Its still many years off.
-
One word, Cost
...but give it a few years...
...but give it a few years...
I can only assume that the writer was thinking in the short term (i.e. immediate application, not whats coming several years down the road)
QDR (Quad Data Rate) memory is already a reality. It's already being used in enterprise class routers (and probably a few other places I am unaware of). So why do we still use DDR on the desktop? Because QDR is not yet cheap enough to be practical in the mass market.
I imagine that the devices that will initially make use of this Diamond technology will probably have a hefty price tag. Much more than your average consumer is willing to spend (possibly more than many of them make in a year)
I think most of us have seen this enough times to recognize that It's just the same old familiar technology curve at work. Technology which is new, exotic and expensive today, will be plentiful, cheap, and obsolete tomorrow.
well, one of diamond's characteristics is high thermal conductivity, so presumably generated heat could easily be carried away with heat-sinking technologies.
make world, not war
Vacuum tubes are used for high power broadcast transmitters.
If you're going to make a big UHF/VHF/FM/SW/AM transmitter, you are going to use power tetrodes.
For instance, a pair of Eimac 4CW1400KG/X-2242 are rated at 4,600KW, continuous. The tube uses water vapor cooling and recovers some energy by using the superheated steam from the tube to drive a steam turbine generator set.
The diamond devices are intended for power output stages of broadcast transmitters. I somewhat doubt that they will replace ultra high output beam tetrodes for 50KW and larger transmitters.
From the fine article:
It is targeting devices with an operating frequency of 200 GHz and an output power of 30 W/mm.
That's the output (RF, I assume) power of the chip. Not the total power consumption/dissipation of the device, which I would guess would be more like 200-100W. Serious cooling is necessary, of course, but I hear the diamond doesn't vary nearly as much with temperature as Si does, so heat is less of a problem.
everything in moderation
Because you have to run those signals over wires, which do a really crappy job of conducting a high speed signal. On chip cache is certainly fast - just expensive (real estate and fabrication errors)
At the sort of frequencies we're currently using, circuit tracks look more like inductors and capacitors than bits of wires. They essentially act as antennas, and there is a massive amount of effort spent in trying to avoid those effects.
The basic idea is that (unfortunatly) there are just a few players out there, but (fortunately) they're big players. They intend to revolutionize computing the same way the mosfet did, etc. I don't know all the science and stuff, but basically they're getting able to make high quality, super good, diamonds synthetically, fairly reliably, and fairly cheap.
Most importantly, they're able to make the diamonds without DeBeers. I don't remember which companies are focusing on which side of things (jewelry vs. semiconductors), but I think the moral of the story is that progress is being made, and the diamond semiconductor revolution does NOT have to include DeBeers, which is a good thing for all involved.
I was a little surprised nobody mentioned this story that was posted recently here.
If this man and his product really pan out, we could see some eally exciting advances in the semiconductor industry. But there could be a billion dollar enterprise that might think otherewise.
A quote from said artice:
But De Beers wasn't backing down. Throughout 2000, the cartel accelerated its Gem Defensive Programme, sending out its testing machines - dubbed DiamondSure and DiamondView - to the largest international gem labs. Traditionally, these labs analyzed and certified color, clarity, and size. Now they were being asked to distinguish between man-made and mined. The DiamondSure shines light through a stone and analyzes its refractory characteristics. If the gem comes up suspicious, it must be tested with the DiamondView, which uses ultraviolet light to reveal the crystal's internal structure. "Ideally the trade would like to have a simple instrument that could positively identify a diamond as natural or synthetic," De Beers scientists wrote in 1996, when the company unveiled plans to develop authentication devices. "Unfortunately, our research has led us to conclude that it is not feasible at this time to produce such an ideal instrument, inasmuch as synthetic diamonds are still diamonds physically and chemically."
Faster switching speed does have benefits in power reduction.
One on the main causes of heating in semiconductors is the switching performance. Whilst a transistor is "on", voltage accross it is zero (or near to), current high, power dissipation (equals voltage * current) is low. Whilst a transistor is "off", voltage accross it is high, current is zero, power dissipation low. However, during the transition from on/off, voltage and current levels are both intermediate, hence power dissipation occurs. Faster switching response times means less dissipation during switching.
One reason which another poster mentioned is the data transfer over the bus between the CPU and Main Memory, this is usually a few inches which means the signal can take more than 10ns to travel along the bus (which is a significant amount of time in chip design).
Another reason is that SRAM is used in a CPU for cache - its VERY fast but takes up more silicon per bit and is very expensive per bit.
Main memory is generally made of DRAM which is slower but also much smaller so you can get a much larger amount of memory onto a chip and much cheaper.
It's not that the latest technology isn't used in memory, it's just that its very expensive so it's used within the CPU as a cache while main memory will be slower in order to balance space vs cost for the machine to still be both affordable and usable.
Once the price drops, the cache technology gets put into main memory and a newer faster one replaces it in the cache.
The other big thing is that most of the advances in CPU speed are not due to the chip tecnology but due to design, especially pipelining.
CPUs go through a series of stages (eg fetch-read-execute) and the CPU can take advantage of this by running each stage while the next stage is still running.
This trick can't be taken advantage of in memory as memory does not contain several stages - hence pipelining increased cpu speed by something in the region of 5-10x while not increasing memory speed at all.
It's mainly new design tricks like this that have made most of the speed advances, which is why processor speed increases at such a larger rate than memory speed.
...that after we reach that 300 GHz barrier, and it is not possible to have CPUs faster than that, we will already be headed down a path where parallel computation is the focus.
:) at the same time!
:p Maybe I have a shot at the next-next generation CPU. ;)
I am thinking along the lines of DNA computing or quantum computing, where the CPU(s) cunch numbers in efficiently massive parallelization. Instead of solving all the factors of 15 in a linear, one by one approach, we will have computers that go about all the possible factors simultaniously, and in one "operation" report back all the true and false factors. Either with many quantum atoms, numerous DNA strands, or even an large number of traditional pipelines in a CPU. It is this direction of massive parallelization that I think we are headed.
Heck, it is already like that with fiber optics. Copper could send one signal (2.7x10^8 m/s), then they went to fiber which was faster (2.99 x 10^8 m/s). Once we reached that theoretical, and practical, barrier, we got "faster" through parallelization. No, we don't just send 1 or 3 or 10 light signals down a fiber, we can send up wards of 380 different signals (well, in my lab
I just wish that I had thought of sending more wavelengths down a fiber long ago... if so, I wouldn't need to be reading slashdot for fun.
Outcome #1 has already happened.
This article talks about two companies (one based in Florida, one in Boston) that have both developed separate methods for manufacturing diamonds. Both are gem quality and one may be completely indistinguishable from natural diamonds. (The magazine cover isn't bad, either.)
Very few people are understanding what the article is saying
The research teams have been able to fabricate semiconductor gates. In other words, they have probably been able to make a couple lone transistors (on/off electrical amplification switches) on a substrate lying in a lab with very controlled conditions -- long way off from computer processing.
You can run Doom on this about as easily as you can run Quake with your bedroom lightswitch...
I explained it exactly above. If you underclock, rate of switching goes down, but the response time is the same. This means that in every unit of time, there will be less of the tranistions that cause heat dissipation.
Response time provides a maximum limit to switching speed. A device that is capable of switching fast, but is operated at a lower speed (underclocking), will produce less heat.
There are some very undesireable things about semiconductors. They are low power devices. They don't work well at high frequencies. Couple these faults together and you let out the magic smoke on higher frequency applications (mostly Sat-Comms).
There are work arounds for the low power problem. In my job, (US Navy Electronics Technician) I've worked on an LF transmitter that could crank out over 150KW. It was all solid-state. The workaround to not cook silicon? It used about a freaking million amplifier circuit cards. I think it might have been more efficetive to just use 4 PA tubes but whatever.
Now the problem is high frequency and high power together. Consider the semiconductor. Two (slightly) different materials with a depletion region in the center. Well that's basically like a capacitor. Capacitors tend to pass higher frequency signals. If the signal is getting passed, it is not getting amplified. This problem is called inter-electrode capacitance. Tubes suffer from the same downfall. They dont just resemble capacitors, they are capacitors to a degree.
The tube world has to use some pretty crazy devices to amplify signals at high frequencies. These methods cannot transfer to the solid state world. For more information google for "klystron", and "travelling wave tube".
But because the issue of inter-electrode capacitance cannot be easily solved with workarounds. The only way to have a high frequency, high power amp, is with a tube. With higher quality semiconductors, this will no longer be true.
I wish there was some there was some way that I could be outside playing basketball, in the rain, and not get wet.
Apollo Diamond is now making near perfect crystal diamonds by vapor deposition. Their product has fewer flaws than natural diamonds. Since the diamond jewelry industry has been making a big deal out of "flawless" diamonds for a century, they're stuck - the industrial process is better than the natural one. Semiconductor process technology has been making near perfect crystals of silicon, quartz, sapphire, ruby, etc. for years, after all. This is just the next step.
Sapphires used to be rare gems. Not anymore. Linde Chemical started making synthetic star sapphires in the 1970s. Then sapphires went into volume production. Then the patents ran out. This is where the sapphire industry is now:
A few years, and bulk diamonds will be on the Home Shopping Channel.
Enhancement mode MOS transistors have characteristics very close to those of ideal pentodes, and should therefore give even better results (no transformers.) But that doesn't suit the guys (always guys) with the "golden ears" and the bullshit filter bypass.
Panurge has posted for the last time. Thanks for the positive moderations.
Negative; the GM operation was shut down because all they could produce cheaply with their hydraulic presses was diamond powder. They actually were to the point where they could make contiguous crystalline structures bigger than dust; however, the cost far exceeded that of the DeBeers extortion and international crime fee diamonds. Though GM abandoned the project for purely financial reasons, I'm sure that DeBeers was happy about it nonetheless.
As for tube amps in high-power situations, that's still the norm. The reason tubes fell to discrete transistors was mainly due to the fact that tubes have to be heated to work right. While several tube heaters in a small radio mean serious inefficiency, a 200W tube heater coil in a 200 KW radio transmitter means that all of 0.1% of your broadcast power is used for the tube heater - no big deal. Add to that the fact that large transistors are very expensive and the difficulty of moving heat away from the junctions in something that large, and tubes are still the natural choice for really high-power applications.
That's it. I'm no longer part of Team Sanity.
Mobile processors can usually alter their clock on the fly, but this requires tight intergration with the motherboard circuitry which is traditionally responsible for generating the CPU clock (which could probably also be programmed to overclock).
Religion is a gateway psychosis. -- Dave Foley
Tube sound is different than solid state sound. I have a old Dynaco ST 70. Blow the shit out of anything solid state. If you do not believe me. Check out the newsgroup rec.audio.tubes. Also, tube amps just report the distortion figures correctly. Not like the Future Shop garbage of today.
I hope you're being sarcastic;
Actually, I'm being entirely serious.
the only area where there's even a difference between the output of tubes and transistors is when they're overdriven.
That's a damn lie. All the recording artists I know use a tube preamp on their vocals -- and how do you overdrive a vocal track other than screaming into the mic?
Matthew G P Coe
http://mgpcoe.blogspot.com/
perhaps you should go read the article titled :(
it will explain why your thinking is typical, and also incorrect
i used to think the same thing as you cuz thats the popular view, but that doesnt make it correct.
I KUT J00 M4NG!!!
I would like to point out 2 things:
1) SRAM is actually Static RAM. It's very vast but it also requires a LOT more transistors per bit than DRAM - Dynamic RAM. I do believe that SRAM also consumes more energy than DRAM (i'm not absolutly sure). Don't expect SRAM to be use in Main Memory anytime soon (unless people are willing to pay the same for 100M as they pay today for 1G - and i'm being optimistic here)
1b) Note that EDO memory, DRAM, SDRAM, DDRAM are all different technologies based around Dynamic RAM. The biguest difference between them is not the way the bits are stored but the way they are accessed - both the "comunications protocol" with the memory chips and the speed with which they respond to requests (there is more to memory than just MHz)
2) Actually, improvements like pipelining don't affect the maximum clock frequency of a microprocessor (the GHz thing) very much. What they do improve is the average ammount of processing work that can be done per-clock-cycle.
To put things in another way, if somebody made a 3GHz 386 processor it would have less than 1/10 of the processing power of a 3GHz Pentium 4 even though the clock speed would be the same.
There was an article in the most recent Pop Sci or Discover (I can't remember which) abotu two companies that have successfully made large-karat diamonds synthetically. One company in Florida, Gemology I think, hastered the hydraulic press and can produce a 3-karat diamond, with few flaws, for $100. Another company out of Boston, I believe, uses a plasma deposition method that produces better-than-nature flawless diamonds... 3k for $15. And this latter process promises to be able to deposit not just chunks (i.e. jewelry), but wafers (i.e. semiconductors!)
Of course, the preseident of the latter of the two companies was at a diamond conference and was told by a DeBeers fellow that what he was doign was a good way to get a bullet in the head!
Wired 11.09: The New Diamond Age discussed on Slashdot earlier. Actually the link to the eariler /. story was posted above under "foreshadowing".
Future Wiki -- If you don't think about the future, you cannot have one.
Mobile processors can usually alter their clock on the fly, but this requires tight intergration with the motherboard circuitry which is traditionally responsible for generating the CPU clock (which could probably also be programmed to overclock).
That's not correct, actually. The CPU reference clock is generated outside of the CPU, but it gets multiplied inside the CPU. So, when a mobile CPU wants to slow itself down, it just reduces the multiplier, while the reference clock remains the same
Never underestimate the bandwidth of a 747 filled with CD-ROMs.
The article writes about problems with the diamonds purity. However, September's Wired has an article about manufactured diamonds for this purpose. One of the diamonds created by a plasma carbon process can be used to grow diamonds in a wafer shape for processing. They have succeeded in creating a positive charge with Boron and also a negetive charge also using Boron in a process. This allows npn or pnp transisters. Because they are grown instead of mined and DeBeers does not control them, they are also cheap enough to be a mainstream computing resource. These diamonds are flawless and perfect size and shape. Unlike trying to use mined diamonds, you dont have to find diamonds that match because they are all grown the same.
there is a corporation in Boston which is developing ultra-pure diamonds using a vapor disposition techinque
You're thinking of Apollo Diamond, which plans to use revenues from selling vapor deposition gemstones to fund research into diamond semiconductors. There's a nice writeup about synthetic diamonds at E2.
However, in many markets, synthetic diamonds sold as gemstones have to be labeled as synthetic, giving De Beers an out: "A diamond isn't forever if it was grown in a lab five days ago."
Will I retire or break 10K?
Is the diamond transistor really even all that special? IBM announced a 210 GHz transistor a long time ago. Any wonder why the PPC 970s are kicking the crap out of anything Intel has to offer? [Sorry, I couldn't resist ;)]
1) SRAM is actually Static RAM. It's very vast but it also requires a LOT more transistors per bit than DRAM - Dynamic RAM. I do believe that SRAM also consumes more energy than DRAM (i'm not absolutly sure). Don't expect SRAM to be use in Main Memory anytime soon (unless people are willing to pay the same for 100M as they pay today for 1G - and i'm being optimistic here)
The typical SRAM structure is a 6T circuit (That is 6 transistors), while DRAM is 1T. DRAMS does however need to be refreshed with regular intervals as the capacitor that stores the bit is prone to leakage. This means the DRAM can never idle at virtually 0 power consumption.
SRAMs therefore consume a lot less power than DRAMs when there are significant idle cycles.
Chips in these frequency ranges are analog - low noise amplifiers, mixers, and power amplifiers. Commercially available chips are available up to 100 GHz or more. These chips typically have no more than 20 or 30 transistors, if not much less. The chips are ussually based around GaAs or InP processes.
The current limitation of these chips is power. The leader is TriQuint, which produces chips that produce 1 to 4 watts around 40 GHz. Thermal limitations are important - GaAs is a terrible thermal conductor. And these analog amplifiers are biased with transistors in conduction, so the efficiencies are on the order of 15% - they generate a lot of heat. (There are other limitations as well, of course, having to do with breakdown voltages,gate width, and switching speed.)
Up until now, the option for high power is a good old fashioned vacuum tube - the traveling wave tube. They have several problems - poor linearity, high noise, the need for kilovolt power supplies, and reliability. Also, they're not cheap to make.
All this to say, diamond is an exciting prospect for analog power amplifiers, and it wouldn't take very many transistors to really make something valuable.
I'm away from my reference books at the moment - does anyone have a comparison of the electron mobility in diamond versus GaAs?
(My associates would consider me remiss in my duties if I didn't mention their high power solid state amplifiers, at Sophia Wireless
It's not wasting time, I'm educating myself.