Slashdot Mirror
Super-fast Transistors On the Way
nbannerman writes "The BBC is reporting about a new kind of transistor, that recently set a world record of 110Ghz. From the article: 'To achieve the speed gain, researchers at the University of Southampton added fluorine to the silicon devices. The technique uses existing silicon manufacturing technology meaning it should be quick and easy to deploy.' The apparent applications for this process include mobile phones and digital cameras."
They're talking about transistors, not entire processors. High speed transistors are needed for the RF front-end, where analog signals up to 1 GHz or so are encountered. These signals require devices that can switch at speeds significantly faster than the signal frequency. Formica
The faster the chip cycles, the higher the communication frequency can be. It is difficult to do noise-reduction calculations on ultra-high frequency communications without chips that cycle at the rate of data transmission.
11 GHz chip != 11 GHz processor. They're mainly talking about analog chips - i.e. op-amps, oscillators, high speed muxes, etc. Chips like these: http://www.maxim-ic.com/solutions/cellular_handset s/index.mvp?pl_pk=14
http://www.analog.com/en/subCat/0,2879,770%255F851 %255F0%255F%255F0%255F,00.html
No, because whenever Slashdot covers these ultra-high-frequency transistors, they never bother mentioning that there's a huge difference between transistors optimized for logic (always on/off, usually very high drive levels and low gain, fast switching of square waves) and transistors designed for RF signal amplification (Usually designed for linear amplification of sinusoidal or modulated sinusoidal signals, lower drive levels with higher gain, and no one cares about the switching time, just the highest frequency sinusoid at which the device exhibits gain.) In essentially every case, the article is covering amplification of a signal at the record-setting frequency, not operation of a logic gate at that frequency.
There is also a very good chance that while the manufacturing process may be suitable for single (relatively) large tranistors (perfectly suitable, and often desireable for RF), it is not suitable for integrated circuits with multiple tranistors and other components on a die. Gallium Arsenide is a perfect example of this - The IC industry gave up on it pretty quickly because it was simply too difficult to make integrated circuits with it and the performance benefits for logic circuits weren't worth the costs, but manufacturers of RF transistors are still putting large amounts of effort into GaAs and plenty of commercial products exist. (Yes, there are still issues with GaAs technology and a lot of companies still don't trust GaAs in their products except in low-volume high-performance applications, but it's not like logic circuits where nothing exists on the market.)
Same thing with IBM's big SiGe push - great for RF but doesn't seem to have made any inroads to logic, probably due to cost issues and technical problems that make SiGe potentially unsuitable for logic but don't really affect their RF performance.
retrorocket.o not found, launch anyway?
And before anyone brings up that TFA does mention "clocking", the impression I get is that the writer of the article isn't very technically literate and doesn't really understand the difference between RF circuitry and clocked logic circuitry. See the comment about mobile phones operating in the 1 GHz range - even the fastest smartphones have a CPU clock speed of only 400-500 MHz at most, but mobile phones have been operating with RF carriers close to 1 GHz (specifically 800 and 900 MHz) for 15-20 years, and the 1.8 and 1.9 GHz bands have been in use for close to a decade too. Satellite communications systems frequently operate in the 10-20 GHz region. I don't see any case where the researchers are directly quoted talking about using their new developments for logic circuitry, but a few where they are implying using the new stuff for RF.
retrorocket.o not found, launch anyway?
The gate will only switch faster for the same, or less, current if the gate capacitance is decreased - i.e. you make the device physically smaller. If the capacitance stays the same then you must bung in current faster to achieve faster switching. Faster switching may reduce the power cycle (time), but on its own it doesn't reduce the power requirements.
I believe this technique would speed up MOSFETs as well because they are saying that the added fluorine doesn't allow the boron to diffuse into the silcon as much. This means you'll have a cleaner line between the p-type and n-type dopped regions. In terms of MOSFETs you could inject the flourine under the gate so when you dope the silicon to create the source and drain you won't have overlap you normaly get under the gate. This means you could reduce the gate to drain and gate to source capacitances which kills the high frequencies.
Sure, it might be the fastest silicon BJT, but as TFA alludes to, there are InGaAs HBTs that are functionally equivalent to BJTs and have cutoff frequencies of 710 GHz. Specifically, I'm talking about the one discussed in this paper by Milton Feng's group at the University of Illinois.
Um, sure, but you've got the wrong substance in mind.
Fluorine: http://en.wikipedia.org/wiki/Fluorine
Fluoride: http://en.wikipedia.org/wiki/Fluoride
There would be some pretty serious differences betweent the two. Neither is good for you to ingest, but one is just REALLY BAD to get anywhere near you at all!
No Comment.
You're both right.
You are talking about basic c*v**2 current, and he's talking about shoot-through current during the transition. Though one normally doesn't fuss too hard about shoot-through unless slew rates are really slow. But then again, it wasn't that many years ago that device standby leakage was nearly negligible, instead of being a substantial fraction of the active current, like it is today. For that matter, the scope traces I've seen of high-speed clocks look a heck of a lot more like a sine wave than a logic pulse, but at this point we're stressing capabilities of the measurment electonics, too.
The living have better things to do than to continue hating the dead.
Jesus fucking Christ, it's GHz not GhZ, you babboon.
RTFA
This isn't about CMOS, for a change. This is about analog power amplification and the 100GHz figure quoted is either the maximum frequency of current or power amplification. Too bad the BBC doesn't say.
Most cell phones contain one Gallium Arsenide bipolar transistor to amplify the signal going to the antenna. This faster Silicon transistor would open up other transmission frequencies but it wouldn't make that game of Alchemy play any faster.
A sig! What kind of person do you think I am?
Actually, the article says that they created a BJT transistor
"The research was carried out using a simple type of transistor known as a silicon bipolar transistor."
Processors use FET transistors because BJT transistors need current to bias them all the time. These transistors would consume way too much power to make any sort of processor (especially for mobile devices). As others have commented, this would only be useful for the analog processing of the output transmitter.
Like pi? Try 10,000 digits.
Yeah, except TFA says the gains were achieved with modified BJT technology, which is not CMOS. In addition, the faster that you switch COMPLIMENTARY (that's the C) MOS structures, the larger the shoot-through current (this is the current that flows between the power supply rails as each transistor in the complimentary structure is temporarily partially conducting). In microprocessors and memory cells, these are responsible for huge transient current requirements, and get worse as the clock frequency is increased.
The reason that the development is significant is not from a microprocessor standpoint - it means that the front end amplifiers and mixers that have to run at the highest frequencies can be fabricated using more cost-effective manufacturing techniques. This is assuming that the article is correct in stating the development concerns BJT's. Hell knows why they showed a photo of a non-populated circuit board, but hey, it's the media. Guess you have dial your expectations lower.
This here's my BIPOLAR TRANSISTOR. I design analog circuits with it and it's got an Ft of 110 GHz.
I don't design digital circuits with bipolar devices. I design digital circuits with CMOS devices. Bipolar sucks power but it runs fast. CMOS sips power but it run's slower.
And if I'm going to design anything usefull with it, that thing is going to operate at about 1/10th of the cut-off frequency (Ft).
This ain't about 110 GHz CPUs.
This is about Op Amps and Phase Lock Loops.
I know that Circuits 101 was a long time ago for some of you folks, but really.
If you are looking at a little more clarity on why mobile phones, you need to think about radios and signal processing. To digitally sample a signal, you need to sample it at least twice the highest frequency (this is the Nyquist frequency). If you want to create or receive a cell phone signal (around 2 GHz) you need to sample it around 5 GHz, and to digitally process what you receive, you need to be processing at these higher speeds. Without such speeds, receivers and transmitters need to use analog electronics to modulate slower digital signals up to 2GHz, and analog electronics are not flexible. I said 2 GHZ, but some cells are 1.8GHZ, some 1.9GHz, and so on. If you do it all digitally, then changing cellular systems and frequencies becomes a software issue rather than a hardware issue. Now you can use a single phone for CDMA and GSM networks. Carriers can upgrade or switchover their networks without having to get people to trade in their phones. The people really salivating over this are probably the DOD's JTRS software radio people who are trying to make a single radio to handle all military waveforms under 2 GHz (assuming the war didn't swallow their funding).
"I love his boyish charm, but I hate his childishness" - Leela
Just a few weeks ago there was an article about IBM in conjunction with Georgia Tech, supercooled reaching 500ghz, room temp was at about 300ghz.
Hos is this new one a world record at 110ghz?
Check this out. TFA's stuff is slow, and pisses away power like it was water. This stuff... they can make a functional transistor by bouncing a single electron off force-field walls. One electron. To test it at full speed, they first need to figure out what to use as a THz scope.
As always, all IMO. Insert "I think" everywhere grammatically possible.
Who reads BBC news for scientific discovery?
_ Record_fT.pdf
Summary:
http://eprints.ecs.soton.ac.uk/12112/
pdf:
http://eprints.ecs.soton.ac.uk/12112/01/2006_Kham
"Fix it"