Forums

Why Hasn't This Been Done with Silicon Carbide

Started by John Savard December 12, 2014
On Friday, December 12, 2014 1:27:22 PM UTC-7, John Larkin wrote:

> GaN is big in RF, especially military stuff. Lots of people are > investing big time, and shipping parts. Macom, Cree, Triquint, > Nitronics, Amcom, IR, probably more. > > I haven't seen a SiC or GaN fet with avalanche ratings.
This inspired me to do a search on gallium nitride. One thing I saw was that, at MIT, someone had made a hybrid chip which was a silicon chip with some gallium nitride transistors for the parts that needed to be extra fact. This is nice, but BiCMOS is much cheaper - a CMOS chip with a few ECL circuits for the parts that need to be really fast. And while SiC goes up to 500 degrees C, silicon on insulator can go up to 300 degrees C - and that's what AMD has been using for its microprocessors for some time. Another thing I found in that search was a mention of a processor TRW is working on for the Advanced EHF satellite program - so at least the military apparently has processors with higher than normal clock speeds. This wasn't GaN, it just happened to be mentioned as a future possibility in a speculative comment in the article. John Savard
On Friday, December 12, 2014 1:58:32 PM UTC-7, Jon Elson wrote:
> Tim Williams wrote:
> > So for example, the Cray-1 or whatever it was that was constructed > > entirely from GaAs and ran at a blazing 60MHz -- was made in old fashioned > > NMOS. No CMOS, no low voltage, no ECL. Almost as hot as ECL though.
> The Cray-1 and Cray-1S were made with traditional ECL, although a custom > fast version somewhere between ECL100K and EclinPs types, with about 750 ps > propagation delay. That's why the static DC power dissipation was about > 100KW. The clock was 80 MHz.
Yes, but the Cray I wasn't made from Gallium Arsenide. Seymour Cray did make a failed attempt to make a successor to the Cray I that was GaAs, and maybe that's the "whatever it was" he was referring to. John Savard
On Saturday, December 13, 2014 12:10:21 AM UTC-7, John Savard wrote:
> On Friday, December 12, 2014 1:58:32 PM UTC-7, Jon Elson wrote: > > Tim Williams wrote: > > > > So for example, the Cray-1 or whatever it was that was constructed > > > entirely from GaAs and ran at a blazing 60MHz -- was made in old fashioned > > > NMOS. No CMOS, no low voltage, no ECL. Almost as hot as ECL though. > > > The Cray-1 and Cray-1S were made with traditional ECL, although a custom > > fast version somewhere between ECL100K and EclinPs types, with about 750 ps > > propagation delay. That's why the static DC power dissipation was about > > 100KW. The clock was 80 MHz. > > Yes, but the Cray I wasn't made from Gallium Arsenide. Seymour Cray did make a > failed attempt to make a successor to the Cray I that was GaAs, and maybe > that's the "whatever it was" he was referring to.
Apparently he was talking about the Cray 3. One was delivered, it was the first Cray to use gallium arsenide chips - but I haven't seen a reference to the logic family it used, however. John Savard
On Saturday, 13 December 2014 16:21:47 UTC+11, meow...@care2.com  wrote:
> On Saturday, December 13, 2014 2:32:59 AM UTC, Bill Sloman wrote: > > On Saturday, 13 December 2014 12:01:46 UTC+11, Gerhard Hoffmann wrote: > > > Am 13.12.2014 um 01:36 schrieb Bill Sloman: > > >=20 > > > >> Gigabit Logic once made GaAs logic chips, long gone now. > > > > > > > > I used them - quite successfully. ECLinPS is now almost as fast > > > > > > > > http://www.onsemi.com/pub/Collateral/MC100EP016A-D.PDF > > > > > > > > looks as if it can be cascaded to count at 600MHz, rather than the =
800MHz we got out of the GaAs equivalent.
> > > > > > >=20 > > > Try these: > > > < http://www.onsemi.com/pub_link/Collateral/MC100EP016A-D.PDF > > >=20 > > Isn't that exactly the URL that I posted? > >=20 > > My problem - back in 1988 - was that I needed a 32-bit wide counter clo=
cked at 800MHz. The plan was to do most of the counting in 100k ECL which w= as going to be tricky, and - in the event - never worked, because the guy w= ho did the detailed design of the relevant board didn't bother reading the = bit of my specification that discussed how to get the 100k ECL to work as f= ast as was needed, and an unrealistic schedule meant that the design review= s that would have caught this had had to be by-passed to meet the schedule.
> >=20 > > Comically, a completely unrelated screw up meant that that board never =
worked - some idiot in the drafting office okayed a redistribution of the i= nner layers of the board which meant that none of the critical printed circ= uit tracers had the characteristic impedance (or the immunity from cross-ta= lk) that they needed.
> >=20 > > The board was persuaded to sort of work by rerouting all the critical p=
aths into subminature coax (1.1mm OD Filotex - it's still available from Fa= rnell, but the name has changed). By the third go-around, the board worked = adequately, but half as fast as had been originally intended - nobody who m= attered was prepared to risk changing the logic to make it go that fast, re= flecting the fact that the original demand for 800MHz had been management-i= mposed to allow the boss to claim a 10psec granularity in time resolution w= hich we'd never actually needed, nor would have been able to prove that we = had.
>=20 > I so don't miss all that
I would have been glad to miss it. I wrote the GaAs proposal entirely tongu= e-in-cheek, as a way of sending up the 10psec granularity requirement, as g= etting us into over-priced single-sourced components. Sadly, it could be ma= de to work, though it took us so long to get it working that our market had= gone away before we had gear that we could sell into it ... --=20 Bill Sloman, Sydney
On Saturday, December 13, 2014 12:13:29 AM UTC-7, John Savard wrote:

> Apparently he was talking about the Cray 3. One was delivered, it was the first > Cray to use gallium arsenide chips - but I haven't seen a reference to the > logic family it used, however.
It used diode logic, and the diode logic drove differential amplifiers. The transistors were D-MESFET - depletion-mode metal semiconductor field-effect transistors. John Savard
On Fri, 12 Dec 2014 23:07:59 -0800 (PST), John Savard
<jsavard@ecn.ab.ca> wrote:

[snip]
> >This is nice, but BiCMOS is much cheaper - a CMOS chip with a few ECL circuits >for the parts that need to be really fast. >
[snip] Yep, I was a designer on an Atmel custom RF chip... incoming at 5.5GHz... taken down to ~500MHz with SiGe PECL, then translated to CMOS. ...Jim Thompson -- | James E.Thompson | mens | | Analog Innovations | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | San Tan Valley, AZ 85142 Skype: skypeanalog | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at http://www.analog-innovations.com | 1962 | I love to cook with wine. Sometimes I even put it in the food.
On Fri, 12 Dec 2014 23:07:59 -0800 (PST), John Savard
<jsavard@ecn.ab.ca> wrote:

>On Friday, December 12, 2014 1:27:22 PM UTC-7, John Larkin wrote: > >> GaN is big in RF, especially military stuff. Lots of people are >> investing big time, and shipping parts. Macom, Cree, Triquint, >> Nitronics, Amcom, IR, probably more. >> >> I haven't seen a SiC or GaN fet with avalanche ratings. > >This inspired me to do a search on gallium nitride. > >One thing I saw was that, at MIT, someone had made a hybrid chip which was a >silicon chip with some gallium nitride transistors for the parts that needed to >be extra fact. > >This is nice, but BiCMOS is much cheaper - a CMOS chip with a few ECL circuits >for the parts that need to be really fast. > >And while SiC goes up to 500 degrees C, silicon on insulator can go up to 300 >degrees C - and that's what AMD has been using for its microprocessors for some >time. > >Another thing I found in that search was a mention of a processor TRW is >working on for the Advanced EHF satellite program - so at least the military >apparently has processors with higher than normal clock speeds. This wasn't >GaN, it just happened to be mentioned as a future possibility in a speculative >comment in the article. > >John Savard
GaN fets are in production and widely used in RF apps. There are some great super-wideband "pallet" RF power amps available now, and the military boys love GaN. A GaN power fet blows away any bipolar or mosfet, for sheer speed and low gate drive requirements. The Gm/Cin ratios are awesome. GaN is usually grown on some substrate material, silicon or SiC or, lately, diamond. -- John Larkin Highland Technology, Inc picosecond timing laser drivers and controllers jlarkin att highlandtechnology dott com http://www.highlandtechnology.com
On Fri, 12 Dec 2014 22:25:34 -0800, Robert Baer
<robertbaer@localnet.com> wrote:

>John Larkin wrote: >> On Fri, 12 Dec 2014 02:35:43 -0800 (PST), John Savard >> <jsavard@ecn.ab.ca> wrote: >> >>> Today's microprocessors are all made using the CMOS logic family. >>> >>> It has the advantage of using a minimum amount of electricity, since - except for leakage currents, which are becoming more important as transistors and wires shrink - electrical power is only used during changes of state. >>> >>> However, the performance of CMOS circuits is limited by the slower P-type FET branch of them - this can be helped by going to Germanium, which has high hole mobility, or by using stretched silicon, or by using domino logic... the IBM CELL processor used an alternative approach instead of domino logic which also worked. >>> >>> And the fastest logic family used to be ECL, because the transistors didn't saturate. But it was an energy hog. >>> >>> Silicon carbide can tolerate high temperatures, and it's a semiconductor. The trouble is that silicon carbide crystals are riddled with defects, so it's a challenge to even make decent single transistors out of it, let alone microprocessors with millions of transistors! >>> >>> However, the thought occurred to me that surely there must be some material that forms nice single crystals, free of defects, with the same interatomic lattice spacing as silicon carbide. One could use wafers of _that_ - presuming it's also an insulator - and using chemical vapor deposition, produce good silicon carbide transistors in large numbers per die... >>> >>> No doubt there are good reasons why that is harder than it seems. >>> >>> John Savard >> >> Compound semiconductors, in addition to defects, tend to make only >> N-type devices. So, no equivalent of CMOS. >> >> Silicon is unique in that SiO2 makes a great implantation mask, so the >> photolithography process works really well. >> >> I don't think anyone has sold even an opamp from SiC. There are >> microwave ICs made from InP and GaAs and such, but tend to be >> relatively simple, with distributed amps being the high end of >> commercial products. >> >> http://www.semiconductor-today.com/news_items/2014/AUG/KTH_010814.shtml >> >> http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=6019027&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D6019027 >> >> That is NOT very impressive so far. >> >> >> Gigabit Logic once made GaAs logic chips, long gone now. >> >> SiC fets on the market now are slow, due to high gate contact >> resistances. They need a lot of gate voltage swing, too. GaN seems to >> be better. >> >> >...idiot mumblings here... > Sapphire substrate for isolation, silicon Nfets and Germanium Pfets >sputtered / CVD on for CMOS-like circuitry?
Problem is, CMOS is cheap and keeps getting faster. Things like cell phones keep trending to custom cmos, even in the RF sections, which, tragically, results in end-of-life on all sorts of cool discrete GaAs parts. We are approaching the limits of CMOS scaling, the end of Moore's Law, but it's possible that things will level off around 7 nm features and nothing will replace silicon. I was looking at this famous painting from 1514. https://en.wikipedia.org/wiki/File:D%C3%BCrer_Melancholia_I.jpg See the hammer? After 500 years, it looks just like the hammers sold today at Home Depot. It's good enough. -- John Larkin Highland Technology, Inc picosecond timing laser drivers and controllers jlarkin att highlandtechnology dott com http://www.highlandtechnology.com
On 2014-12-13, John Larkin <jlarkin@highlandtechnology.com> wrote:
> On Fri, 12 Dec 2014 22:25:34 -0800, Robert Baer ><robertbaer@localnet.com> wrote: > > > https://en.wikipedia.org/wiki/File:D%C3%BCrer_Melancholia_I.jpg > > > See the hammer? After 500 years, it looks just like the hammers sold > today at Home Depot. It's good enough.
http://www.homedepot.com/b/Tools-Hardware-Hand-Tools-Hammers-Claw-Hammers/N-5yc1vZc98p I don't see any there with the handle riveted to the head, and the material is different too, -- umop apisdn
On Saturday, December 13, 2014 12:44:17 PM UTC-7, John Larkin wrote:

> See the hammer? After 500 years, it looks just like the hammers sold > today at Home Depot. It's good enough.
I don't think it will take 500 years for it to become possible to make a reasonable microprocessor from, say, Indium Phosphide if silicon levels off. There will be an interregnum during which the period of explosive growth we've enjoyed will be but a memory, but it won't be replaced by _absolute_ stasis. One of the things that is likely to happen is that people will have time to look seriously into new architectures that exploit parallelism somewhat more efficiently. John Savard