Slashdot Mirror
Stanford Breakthrough Could Make Better Chips Cheaper
angry tapir writes: Researchers at Stanford University have come up with a new way to make chips and solar panels using gallium arsenide, a semiconductor that beats silicon in several important areas but is typically too expensive for widespread use. "[I]t can cost about $5,000 to make a wafer of gallium arsenide 8 inches in diameter, versus $5 for a silicon wafer, according to Aneesh Nainani, who teaches semiconductor manufacturing at Stanford. The new Stanford process (abstract) seeks to lessen this thousand-to-one cost differential by reusing that $5,000 wafer. Today the working electronic circuits in a gallium arsenide device are grown on top of this wafer. Manufacturers make this circuitry layer by flowing gaseous gallium arsenide and other materials across the wafer surface. This material condenses into thin layer of circuitry atop the wafer. In this scenario, the wafer is only a backing. The thin layer of circuitry on top of this costly platter contains all of the electronics."
One of the very first papers I read for a VLSI design course had one of the weirdest final sentences I have ever heard, from a geeky see-my-smarts cross between physics and car geeks. As I recall, it was something like this:
"And then, of course, there is the problem of gallium arsenide, which is the Wankel Engine of the semiconductor industry."
To which the class (a bunch of undergrads wading into the delightful bliss and head-scratching geekery of academic journals for the first time) collectively and perplexedly said "WTF?"
I apologize if this was explained in TFA and I missed it; but I was left wondering why gallium arsenide would be so dramatically expensive. A quick look shows that even the scammers selling 'gallium bullion' in small quantities are charging under a dollar a gram for the stuff(at allegedly very high purity); and arsenic certainly isn't terribly pricey. Silicon, of course, is really abundant, and still fairly cheap once you've coaxed the oxygen out of the quartz-form you typically find it in; but not lower cost enough to explain a wafer-level difference as large as the one that exists.
Are gallium, arsenic, or both markedly more difficult to purify enough to serve as reliable semiconductors? Is growing sufficiently flawless crystals large enough to be cut into wafers too error prone to get good yields? Some other unpleasant aspect of processing or handling the material?
Wasn't Intel announcing ga.as. as their new technology some weeks ago, for their sub-10 nano chips? I guess they must have solved the cost problem, too.
No, they are making a wafer, building chips on top, remove thin top layer to sell the chips, and reuse the bottom part.
You need a wafer with the same crystal structure and lattice constant. If there is a mismatch
between the inter-atomic distance (aka lattice parameter or lattice constant), the atoms
deposited on this wafer will try to adjust to this lattice.
If the layer is thin, the deposited crystal will in effect be compressed or expanded. While this is OK from a mechanical and
crystal point of view, the electronic properties of the grown semiconductor will change. E.g. the bandgap (energy distance
between filled and empty energy levels) will shift, which will change the electronic properties of the material.
If/when the layer becomes thick (~5 atomic layers), the grown crystal will (try to) revert to its native lattice constant. However, there's
no ordered way to do this, so the grown crystal will contain lots of defects, or on the worst case, become amorphous
(loose its crystal structure alltogether). Defects destroy the material from an electronic point; it provides ample opportunities
for electrones and holes to recombine. It increases the leakage current and power dissipation and alse change the electronic band structure.
I haven't read the article, but what I don't understand how they intend to separate the circuitry from the wafer... it's not exactly
a tape that you can peel off, or a thick slab which you can hammer away.
The article follows the youtube presentation and the summary is, for once, accurate (i.e. does not introduce new errors).
The trouble is that the presentation is utter BS. The GaAs devices are NEVER made out of a solid GaAs wafer; the process starts with a plain silicon wafer, on which GaAs is grown epitaxially. The secret sauce is, and always has been, how to minimize the defect density at the Si/GaAs interface.
Such a wafer is more expensive than the plain Si one, but not 1000x more! Oh, and every purchaser would kill to get $5 8" wafers...
Since the Stanford guys are no dummies, I guess that the announcement was deliberately made to sound ridiculous. For what purpose? Time will tell.
To get the single-crystal purity of the surface layers, they need to perfectly match the crystal dimensions of the substrate. Making it out of the same thing is aa easy way to achieve that.
That makes sense if you're building devices directly on the wafer, but wouldn't the three sacrificial layers interrupt that?
=Smidge=
I remember a joke from about 20 years ago.
Gallium arsenide, the semiconductor of the future, and always will be.
Turns out to be true afterall.
I recall an AMD engineer presenting at MICRO in 2012 telling us that one of the problems with making wafers too thin is that they tend to curl up. I'm not sure whether the warping is inherent in the silicon or doesn't occur until after all the circuit layers are put on top. Regarding the article, the wafer doesn't start out thin. The circuits are formed, and then chips are (in a manner of speaking) shaved off the surface, exposing fresh GaAs to make another set of chips.
Not to mention pretty much everyone has a GaAs amplifier chip in their cell phone. Also CD and DVD drives use GaAs-based lasers.
Silicon has many distinct advantages over GaAs for logic. To many to go in to here.