Tuesday, August 23, 2016

Techy Tuesday - How Do You Build Those Things, Anyway?

Semiconductors, that is.  Everyone knows the world is awash in electronics, from things like iPods to Fitbits, and really far more than one could ever list.  I'd hazard a guess that few people except those in the business have an idea how the integrated circuits in these things are made.  This is a topic that could fill a good-sized series of these posts, but tonight I want to focus on an improvement to the processing of raw semiconductors into integrated circuits (ICs) just announced.

To begin with, most of what you're familiar with is made from elemental Silicon.  It starts out as the most pure substance known to man, but pure (intrinsic) silicon isn't very useful by itself.  It's rather resistive, which is why Silicon is called a semiconductor - it's not a very good wire.  What makes Silicon useful is that tiny amounts of alloying elements are added to it ("doping"), producing either too many electrons in a given volume (N-type material) or too few electrons in a volume (P-type - also said to have "holes", places in the electron cloud that should have an electron but don't).  This ability to customize the conductivity of Silicon is why it's so useful and why this has become the silicon age.   

Over the last 30 years or so, new materials have been developed that offer improvements in performance over silicon transistors.  The III-V (three five) materials were among the first to market.  III-V materials typically give higher electron mobility, a characteristic of their atomic structure, and are often used in High Frequency and microwave applications.  Gallium Arsenide (GaAs) was the first commercial success, and the current industry darling Gallium Nitride (GaN - pronounced with a short "a", like in can).  Another material that has a lot of interest is Silicon Carbide, SiC.  SiC doesn't achieve the high frequency performance of GaAs or GaN, but it has excellent high temperature performance because the compound has very high thermal conductivity.  This makes it a natural for parts intended for car engines, at the end of well-drilling rigs, and other places where the temperatures exceed the 150C or so that Silicon will tolerate.

There's just one problem with SiC: it's very hard.  In mineral terms, it's between sapphire's Mohs hardness of 9 and diamond at 10.  So if one has a crystal of SiC, how does one cut that thing into wafers for further processing?  Any rockhound, tombstone, or granite countertop maker would recognize the current method: a saw made of thin wires carrying a slurry of diamond grit, kind of a hi-tech bread slicer, cuts the wafers free.   
As you can see by the picture, this is a time consuming process.  I find it remarkable that over the diameter of the crystal being cut (4 to 6" is common for SiC), the wire only wanders enough to cause a surface roughness of 50 millionths of a meter.  Which has to be ground away in that final step. 

The big innovation is that Japanese ingot processing equipment manufacturer DISCO Corporation has come up with a laser-based technique to slice wafers out of the SiC ingot, producing 50% more wafers through reduced material losses while slashing production times by a factor of six.
Dubbed KABRA (for Key Amorphous-Black Repetitive Absorption), the patent-pending process uses a focused laser to form an amorphous layer of SiC decomposed into its constituents silicon (Si) and carbon (C), which becomes the base point for separating the wafer through cleavage.
They save material because the diameter of the wire and the diamond grit combine to lose about 200µm per wafer of SiC.  The diamond is focused below the surface of the crystal, turning the point it scans into separate silicon and carbon atoms; the wafer is cleaved off resulting in half the loss of the SiC.
With half the loss of the diamond wire saw method, this has to drop the costs of SiC components.  Add to that the lowered cost of processing by taking 1/6 the time of the previous method, and SiC parts are about to get quite a bit cheaper. 

I debated doing this story because it's rather deep into the weeds for those of you who aren't even tangentially associated with the business, but I love this sort of story.  We have a problem - if we can get enough production on SiC transistors, we can make a butt load of money - so we get a bunch of clever folks together and say, "make it better".   It's what engineers do, and it's why this is the most dynamic industry in the world.

Oh.. By the way.  The people among the farthest from the semiconductor business know silicon carbide now, but don't know it.  It's sold as the diamond simulant Moissanite.  Exactly how they produce those clear, white crystals to cut into imitation diamonds when the chemically pure semiconductor is that off-yellow color is their trade secret. 


  1. The Hughes Aircraft facility I worked at had a semiconductor fab on the second floor. They made lots of IMPATT, Gunn, and Tunnel Diodes, and were experimenting with GaAs production. There was a "special" alarm in the building that was sounded if there was a Silane or Arsine leak.

    If you heard that alarm go off, you immediately dropped (literally) whatever you were doing and got OUT of the building as fast as you could.

    I think they finally started getting some functional GaAs products right about the time I was laid off (late 1989).

    They made all the IMPATT, Gunn, and Tunnel Diodes for all of Hughes Aircraft, but were plagued with repeatability problems. They'd make a whole "boat" of wafers and maybe a third of them would make really good diodes. The rest would be "OK" diodes or complete duds.

    AFAIK, they never got their issues sorted out.

    1. That's around when I was first working with GaAsFETS in the satellite communications world. We'd sort noise figure by measuring room temp noise and then when exposed to liquid nitrogen (which isn't zero, but we used it as our reference). Process variations were there, but a large portion of the parts were usable. The FETs were from Mitsubishi.

      I put a relatively high NF FET into a 2 meter transverter. It was about .5 dB NF at 144.

      Pretty primitive times, comparatively speaking.

      I figure if there's a couple of requests, I can go farther into the weeds on how ICs are made.

  2. I probably should have been a bit more specific on how I graded the diodes.

    The "Really Good" ones were world-class, better by far than anything else available, while the "OK" ones were comparable to what other companies were doing.

    The "Duds", of course were duds...didn't work at all, but the wafers could be made into pretty jewelery.

    Then they set up a complete new semiconductor manufacturing on speculation that Hughes (Radar Systems Group, or "RSG" in El Segubdo) would win the contract for a phased array radar upgrade for the F-16.

    Much to their astonishment, RSG declined to bid the contract, and Hughes Torrance was on the hook for about 10 million of 1989 dollars.

    Last time I was at the building was around 2002 or so. It had been sold, stripped, and turned into the Pacific Sales outlet where I bought my Sony HiDef TV set.

    Really funny to walk through the same door, and instead of being greeted by a security guard and being made o sign-in, I was greeted by a display of the newest gas ranges with built-in grills.....

  3. GrayBeard, I am a hamradio operator since 72. If I am correct, an IC is made using some kind of photographic process. Please elaborate some for a simple manual laborer.
    Also, I live near Shelby,MI which has a place called Shelby Gemstone, which makes artificial diamonds. I think they use some type of process similar to what you describe here.
    One more thing, I started out in ham radio using transistors, building my own transmitters. NPN and PNP types were quite familiar to me back then. I am amazed with what they are able to do with IC circuits now days. Brilliant engineers, and amazing technicians.

    1. I got my novice ticket in 1976, so not far behind you, but started playing with electronics about 8 years before that. I worked with tubes and transistors. As a kid, I had to break them open to see what was inside. The vacuum tubes were easier to see!

      Let me work on trying to do a piece on how the IC processing is done.

    2. I should have said that I also used Vacuum tubes as well. Just as output tubes. The old 6l6 was mostly used by me, since that was the one I had on hand and the transmitter that I built. But we learned about them all back then, of course.

  4. It's called "Photolithography".

    Once the logic of the part is designed electrically, they convert it into a layout pattern of all the individual transistors, resistors, diodes, and capacitors.

    This layout used to be done by hand, on a clear base sheet, using different colored tape, similar to how printed circuit boards were made. The completed artwork was called a "Mask", and was then optically reduced, the exact opposite of "blowing up" a film negative.

    When it got down to a certain size, they'd then use it like a slide, and shine a light through it and project the pattern on a silicon wafer that had been treated with chemicals. The wafer would be "exposed" for a length of time, and then "developed", and then etched.

    The whole process back in those early, dusty days was very similar to making a printed circuit board.

    Since that time, the features have gotten smaller and smaller, to the point where they had to stop using white light, and go to monochromatic light of shorter and shorter wavelengths so that the light wouldn't diffuse around the edges of the mask, and "blur" the "print" they were making. I'm sure the process has changed some over the decades since I saw it being done, but that's how it was in Ye Dayse of Olde".

    1. thanks. I thought it was kind of like that, but was fuzzy on any details. Brilliant people continue to innovate, and I am always amazed.

  5. Last I worked with SiC (for spacecraft power converters a few years ago), it could support reasonable NMOS but had poor or non-existent PMOS. High temp is the redeeming feature of SiC but if P devices aren't feasible, it's back to single family logic as in early 70s.

    SiC seems to make more sense for specific components ... diodes or discrete transistors for example ... except in specific cases where cost and effort are justified by the required application.

    Been away from the more exotic processes since about 2011/12, but as you suggest, GaN is said to hold promise. But then I'm more on the analog side of the business and haven't kept up with the latest digital manufacturing techniques (1000 transistors is a pretty large circuit in my world).

    All these years ... from 12AX7s onward ... and moving electrons around is still as fascinating as when I stuck a fork in an outlet to see what happens.


    1. I think, like you say, SiC is best for discrete devices. If all they can make is N-channel FETs, that's not a really bad handicap for a power supply design.

      I've had hands-on some 10GHz GaN power transistors and they're pretty impressive. Gain at 10 GHz isn't the easiest thing to come by, and these parts had both impressive gains and power output. A single device for 125W output (pulsed) with internal matching to 50 ohms. As well as anything at 10 GHz can be made to be "no tune".

  6. You guys passed me by a few posts ago, I think. But anonymous, I hear you about the fork in the outlet. My first endeavor as a kid was using an old motorized can opener to build a mini arc welder. And SiGraybeard, thanks again for this topic. Fascinating.

  7. ... using an old motorized can opener to build a mini arc welder

    Was that deliberate? I know a lot of guys who have used the 110 outlet as a welder, but few who did it deliberately. Not sure I know anyone who did it deliberately. ;-)