jaw@ames.UUCP (James A. Woods) (03/10/85)
# Snuff epigrams.
The first major application of ballistic transistor technology
is the Belle II project at Murray Hill. Since K. Thompson discovered
that each speed doubling of the standard alpha-beta chess machine adds
70-100 rating points, Bobby Fisher, last seen at 2800+, has been hired on
retainer to check out the playing strength of the new VLSI machine.
The claim for the $100,000 Friedkin prize is not far off, making the
recent Karpov-Kasparov battle pale in comparison. Word has it that the
high-level evaluation function has been converted in the last few months
to use the new Karmarkar linear programming algorithm.
Of course, the ballistic gate speeds (100 femtosecond range) combine with
a sub-picosecond cycle time and the speed of light to require a single-chip
implementation. Where this design dedicates much RAM to transposition table
lookup, another research group is implementing a general purpose CPU,
the speed of which (no messy parallelism for the first go) has obvious
consequences for the future national GNP. (The Japanese have their own HEMT
[High Electron Mobility Transistor] program.)
In fact, Seymour Cray, long disassociated from Cray Research
(viz. recent Fortune magazine piece) flew in several weeks ago to aid in
this effort. Heretofore, he was slaving away at a "slow" gallium arsenide
version of a breadbox-sized Cray 3.
The importance of these events cannot be overstressed in the wake
of a failed attempt by the DOD to impose national security restrictions.
Comments about "the book being thrown out" which were originally
stifled by Nobel winner and lab chief Arno Penzias have leaked out
to not only the technical community but to the general press. However,
what is too late for executive level intervention is not too late to
affect the price of AT&T's stock. Curiously, it has remained stable
because Wall Street analysts see product development as too "blue sky".
-- James A. Woods {research,ihnp4,hplabs}!ames!jaw (or, jaw@riacs)
P.S.
Little known anecdotes about Seymour Cray Dept.: not only does he
build a new type of sailboat every year before setting it afire (!!) on Lake
Michigan, but he is the constructor of the world's largest boomerang
(300 lbs. plus), the problem here being that there is no one strong enough
to test out its aerodynamics!abeles@mhuxm.UUCP (J. Abeles (Bellcore, Murray Hill, NJ)) (03/12/85)
> The first major application of ballistic transistor technology > is the Belle II project at Murray Hill. ... > Of course, the ballistic gate speeds (100 femtosecond range) combine with > a sub-picosecond cycle time and the speed of light to require a single-chip > implementation. ... (The Japanese have their own HEMT > [High Electron Mobility Transistor] program.) > ... Heretofore, he was slaving away at a "slow" gallium arsenide > version of a breadbox-sized Cray 3. ... However, > what is too late for executive level intervention is not too late to > affect the price of AT&T's stock. Curiously, it has remained stable > because Wall Street analysts see product development as too "blue sky". > > -- James A. Woods {research,ihnp4,hplabs}!ames!jaw (or, jaw@riacs) A few comments: As a researcher in the field of high-speed III-V optoelectronic devices, I am a little curious as to what exactly J. A. Woods is talking about. I understand from my colleagues that AT&T Bell Labs was embarrassed several weeks ago by a mistaken article (possibly in Business Week) suggesting that AT&T BL had transistors switching in the femtosecond delay per gate regime. This is not true as far as I know. For years, people have been trying to speed up the switching speed of logic circuits, and the very fastest to date is about 10 picoseconds delay per gate, achieved using the technology known variously as HEMT, SDHT, MODFET, and TEGFET, all of which are the exact same thing and which have been around for about five years or more. Note that there is no sub-picosecond delay per gate circuit reported in the literature. The 10 psec speed corresponds to a cooled device operating at liquid nitrogen temperature, or 77 degrees Kelvin (above absolute zero). As to Cray slaving away at "`slow' GaAs," the HEMT/SDHT/MODFET/TEGFET are all based on GaAs technology. As to the Japanese having their own project, this is of course true. The name HEMT was coined by them. The concept originated at pre-divestiture Bell Labs and was demonstrated here before being made into an actual transistor structure in Japan. As to the stock price not going up because it's too "blue sky," I think a good justification for the stock price remaining stable is that there wasn't any advance and furthermore the practical value of these devices is not proven. In this business, everyone knows the old saying, "GaAs is the material of the future, and it always will be." For cognescenti, I will mention something about ballistic transport, also known as hot electron transport. By the way there was an article on this by Prof. Holonyak of the Univ. of Illinois appearing in Physics Today years ago (like, 1978 or earlier). "Ordinary" GaAs transistors having one micron long gates are thought to depend upon ballistic phenomena for their operation, according to device modellers such as Dr. W. R. Curtice of RCA. Probably the most well-known research group seriously tackling the ballistic electron transistor concept (which I would define as a transistor achieving greater speed or higher cut-off frequency than "conventional" FET structures for the present purposes) is that of Professor Lester Eastman of Cornell University. They hope to achieve 94 GHz (a popular frequency for certain military applications) amplification using their vertical FET. That would be roughly equivalent to a 10 picosecond operation, with the achievement of being able to extract power at that speed, something the present demonstrations don't really do (fan-in = fan-out = 1 for those demonstration ring oscillators). As to the chess people making a dedicated chess processor, or using a new superfast processor for playing chess, I have no information, particularly because that work is being done at AT&T Bell Labs, not Bell Communications Research. However, anyone interested in progress made in integrating GaAs devices to make useful processors, etc., should attend the International Solid State Circuits Conference, held in January every year alternating between San Francisco and New York. This year it was in NY. You will quickly find out about the difference between rumors and facts! --J. Abeles High Speed Device Research Group Bell Communications Research, Inc. Murray Hill, NJ mhuxm!abeles The above comments are informal and do not necessarily represent any official position of Bell Communications Research, Inc. Affiliation given for identification only.
ron@brl-tgr.ARPA (Ron Natalie <ron>) (03/12/85)
> The first major application of ballistic transistor technology > is the Belle II project at Murray Hill. Since K. Thompson discovered > that each speed doubling of the standard alpha-beta chess machine adds > 70-100 rating points, Gee, does he still say that? He used to back 6 or 7 years ago, expecting great results as technology advanced. We asked him about it at the Second Toronto UNIX conference and things weren't progressing as he had predicted.
riddle@ut-sally.UUCP (Prentiss Riddle) (03/13/85)
Umm... before any more of you post flames or serious responses to this
article, you might consider re-reading it. I know it didn't come out on
April first, wasn't signed by Bruce Springsteen and doesn't have kremvax in
its path, but I think it nevertheless constitutes a rather vigorous pulling
of the leg.
After all, isn't that what net.rumor is for?
--- Prentiss Riddle ("Aprendiz de todo, maestro de nada.")
--- {ihnp4,harvard,seismo,gatech,ctvax}!ut-sally!riddle
--- riddle@ut-sally.UUCP, riddle@ut-sally.ARPA, riddle%zotz@ut-sallyjewett@hplabs.UUCP (Bob Jewett) (03/14/85)
> > Of course, the ballistic gate speeds (100 femtosecond range) combine with > > a sub-picosecond cycle time and the speed of light to require a single-chip > > implementation. > > -- James A. Woods {research,ihnp4,hplabs}!ames!jaw (or, jaw@riacs) > For years, people have been trying to speed up the > switching speed of logic circuits, and the very fastest to date is > about 10 picoseconds delay per gate, achieved using the technology > known variously as HEMT, SDHT, MODFET, and TEGFET, all of which are > the exact same thing and which have been around for about five years > or more. > --J. Abeles mhuxm!abeles Nope. Try Josephson junctions. Gate delays less than 10ps have been reported by the Japanese. IBM reported 13ps delay in a string of OR gates, 6ps of which was propagation down terminated superconducting transmission lines. The 13ps number translates into ~40ps gate delays with reasonable margins and fan-in/fan-out. Bob Jewett hplabs!jewett
gwyn@brl-tgr.ARPA (Doug Gwyn <gwyn>) (03/15/85)
> > > Of course, the ballistic gate speeds (100 femtosecond range) ... > > > > about 10 picoseconds delay per gate ... > > Nope. Try Josephson junctions. Gate delays less than 10ps ... "About 10ps" is compatible with that and both are far from 100fs. IBM was working on Josephson junctions when I was a youngster. What is the holdup?
john@x.UUCP (John Woods) (03/17/85)
> > > > Of course, the ballistic gate speeds (100 femtosecond range) ... > > > about 10 picoseconds delay per gate ... > > Nope. Try Josephson junctions. Gate delays less than 10ps ... > "About 10ps" is compatible with that and both are far from 100fs. > IBM was working on Josephson junctions when I was a youngster. > What is the holdup? Didn't you hear? They filled their NSA contract shipments, and tore down the assembly line... -- John Woods, Charles River Data Systems, Framingham MA, (617) 626-1101 ...!decvax!frog!john, ...!mit-eddie!jfw, jfw%mit-ccc@MIT-XX.ARPA Campus Crusade for Cthulu -- It Found Me!
chongo@nsc.UUCP (Landon C. Noll) (03/20/85)
In article <434@x.UUCP> john@x.UUCP (John Woods) writes: > > > > > Of course, the ballistic gate speeds (100 femtosecond range) ... > > > > about 10 picoseconds delay per gate ... > > > Nope. Try Josephson junctions. Gate delays less than 10ps ... > > "About 10ps" is compatible with that and both are far from 100fs. > > IBM was working on Josephson junctions when I was a youngster. > > What is the holdup? > Didn't you hear? They filled their NSA contract shipments, and tore down > the assembly line... The NSA must grab most IBM projects. I once thought it was only marketing that sat on many nifty IBM internal projects. :-) And dont forget the NSA snarf on the notes/methods behind the 56 bit and 128 bit "DES"! chongo <The Data Encryption Standard; from now on consider it like Sys V> /\42/\ -- no comment is a comment.
steve@hcradm.UUCP (Steve Pozgaj) (03/28/85)
I would suggest to Mr. Abeles that, prior to writing his opus response, he would have been well served to yank out a dictionary and look up the meaning of the word "humor". That could then be cross-referenced in any thesaurus to yield words such as "joke", "anecdote", "wit", etc. That, in turn, may have provided a clue as to the intent of the original posting (at least, that was how I read it). Steve "I still know how to laugh" Pozgaj, Human Computing Resources Corp.