Slashdot Mirror
DARPA Creates 0.85 THz Solid State Receiver
hypnosec writes "DARPA, under its THz Electronics program, has designed a solid state receiver capable of THz (terahertz) frequencies thus inching towards the possibilities of transistor-based electronics that will operate at THz frequencies. The newly designed solid state receiver demonstrates a gain at 0.85 THz. This particular milestone is a stepping stone for the next target of 1.03 THz. Because of this achievement a host of DoD electronics capabilities can now be realized. One such application where this can be of use is for a sensor that will operate through clouds under a DARPA program dubbed VISAR."
Not just for airports any more. With technology like this, they can start minaturising the tech so every cop doing a stop-and-search can inspect the suspect. For, ah, weapons. Of course.
Article summary is incorrect.
Sorry, EE major and I get annoyed reading this kind of thing...
This is Slashdot, not Wikipedia.
now we need to go OSS in diesel cars
Another terrible article summary.
In 2010, a solid-state device at 0.67THz was achieved. In 2012, that effort is up to 0.85 THz. Progress is slow, but continuing.
Diode-type CMOS imagers for terahertz radiation have been built. Those convert terahertz radiation into DC, which can then be amplified by standard techniques. But diodes don't have gain. That's why the original article emphasizes that this new device has gain.
There are terahertz lasers, waveguides, antennas, and other components that work up there. The situation is much like radar during WWII; there were a few components that could do specific things at radar frequencies (then 60MHz to 1.2GHz), but general electronics wasn't there yet. Most of the electronics in radars of that period ran at far lower speeds. They still worked.
From the comments, I thought this was youtube...
The prefix "micro-" in "microwave" is not meant to suggest a wavelength in the micrometer range. It indicates that microwaves are "small" compared to waves used in typical radio broadcasting, in that they have shorter wavelengths. The boundaries between far infrared light, terahertz radiation, microwaves, and ultra-high-frequency radio waves are fairly arbitrary and are used variously between different fields of study.
--From Wikipedia
... or are they going to try to make a CPU/GPU core at this speed?
In the long run, maybe. In the short run you aren't going to like it. A very stereotypical microwave LNA MMIC operating around a factor of 100 lower then this device frequency (in other words, cheap and off the shelf) consisting of a couple transistors is biased much like a LED... couple volts, couple dozen mA. Lets call it 4 volts at .040 amps thats 160 milliwatts per device. For rounding purposes lets say a tenth of a watt per transistor. So if you have a roughly quarter million transistor original 386 a 10 GHz discrete 386 made out of microwave transistors would draw about 30 or so KW. Which is quite a lot of power. Of course you don't need low noise small signal performance or great fan in / fan out ratios... Regardless high speed individual devices certainly like their DC power.
The problem with making processors fast is keeping them fed with something to do. CPU tech always seems to lead memory/IO/algorithm design, I can't remember an era when the "memory guys" were waiting on the "processor guys" to catch up. With current tech a 1 THz CPU would merely spend 99.9% of its time in idle waiting for memory... But nothing in the world could run a NOP or an endless loop faster than that device.
"Science flies us to the moon. Religion flies us into buildings." - Victor Stenger
Incorrect. Gain is unitless (Vout/Vin), and decreases pretty much proportionally to the inverse of the frequency on amplifiers, so chip makes use something called the Gain Bandwidth Product (GBP) instead of the 'pure' gain, because it is a much more useful number for specifying actual transistor/amplifier performance in real live working conditions. And the unit of the GBP, is, you guessed it! Hz. Thus, a transistor with a GBP of 0.85THz will have a gain of 1 at 0.85THz, a gain of 2 at 0.425THz etc. When I see a gain with units in Hz I subconsciously think 'GBP' and don't even miss a beat...
See http://en.wikipedia.org/wiki/Gain%E2%80%93bandwidth_product if my explanation doesn't make sense.
Help I am stuck in a signature factory!
My understanding is that L1 cache is as fast at the CPU it is embedded in. Just make more of it and watch how we make real-time raytracing instead of relying on polygonal tricks for rendering 3D.
The Wise adapts himself to the world. The Fool adapts the world to himself. Therefore, all progress depends on the Fool.
The summary and linked press releases are light on details so here is what I gleaned from the photograph of the chip based on some experience in the area of microwave/mm-wave device and circuit work. There will probably be much more technical information in upcoming papers in the research literature.
Based on the photo of the chip on the linked DARPA page this is not a receiver, but a low-noise amplifier (LNA) which would be used as the front-end for an imaging sensor or communications/radar receiver. It would be straightforward to turn this into an imaging detector at this point by adding a detector after the LNA though I don't think this has one. For a synthetic aperture radar more circuits will be required, especially a mixer to downconvert the frequency.
The slashdot summary misquotes the article saying that the circuit has "gain of 0.85 THz" but should say "gain at 0.85 THz". The LNA appears to have 10 amplifications stages which is very large for a LNA, which suggests that the gain per stage is still quite low at 0.85THz. This is to be expected as the best per-transistor gain cutoff frequencies are not too far 1THz that I'm aware of. The circuit also appears to be built in coplanar waveguide (a metallized signal strip in the middle surrounded by two ground strips) which is easy to fabricate and good for a research environment but it has a higher loss than microstrip (a signal line above a ground plane).
Anyway that's my 2 cents.
With current tech a 1 THz CPU would merely spend 99.9% of its time in idle waiting for memory... But nothing in the world could run a NOP or an endless loop faster than that device.
This is incorrect, common DDR3 memory is already 2 orders of magnitude slower than a 3 Ghz CPU and they work just fine.
With a THz CPU you also have THz memory (registers, level 1 cache, etc.), if your algorithm fits in the cache you will have close to 100% of the performance.
All digital electronics is ultimately analogue. If you had transistors that could run at that speed, you could probably sample at (at least) a quarter of that speed. Make a very nice ADC/DAC for a software (de)modulator and fixed function DSPs.. Lots of digital applications on the digital side, if the power draw is reasonable.
I have determined that my sig is indeterminate.