Why IBM's Superconductor Computer Failed
Вставка
- Опубліковано 26 вер 2024
- Notes:
8:20 - Note that with this graph, presume that the first part of the red line is vertical, rather than horizontal. My fault for being terrible at drawing graphs. My thanks to viewer Craig for clarifying.
Links:
- The Asianometry Newsletter: www.asianometr...
- Patreon: / asianometry
- Threads: www.threads.ne...
- Twitter: / asianometry
Back in the 1970s, I remember a corporate meeting where the IBM salesman told us that if we didn't buy IBM computers, we would be left out of the coming revolution in "cryogenic computers". I also remember conversations with colleagues who sincerely believed that IBM was lightyears ahead of its competitors. So even though the project failed from an engineering perspective, it was immensely successful - for over 10 years - as a marketing tool for IBM.
kinda like ai today or "fuzzy logic" 20 yrs ago, or "nano technology".
the next big thing that won't happen.
bur they'll surely be the first to have it...😀
@@ivok9846I remember “fuzzy logic”being a thing from 40 years ago in Japanese consumer goods such as washing machines, toasters, etc. in the 80’s. Hadn’t thought about that in a while.
@@mattbland2380 i was thinking about late 90s in stuff available in central europe, but rounded it to 2000s, because it wasn't really available in beginning of 90s.
search for "fuzzy logic washing machine yumpu", they have image from 2003..."news " from some magazine
it seems that's still big in india....
@@mattbland2380 I bought a Japanese rice cooker last year with Fuzzy Logic! It's still around...
Jim
My family was Philips ASML , Cryogenic Engineered Computing to get the superconducting state, IBM only failed to produce the super computer on Josephson laws.
These Cryogenic Engineering we still use, needed for quantum computing now.
Back in 1978, I was studying physics in High School. Our teacher brought a group of us to a lecture by a team from IBM. They talked about Josephson Junction and how that was the future of computers. Throughout high school, college, and into my engineering career, I waited for superconducting supercomputers. I followed the superconducting engineering news and the tiny breakthroughs. The idea of high-temperature superconductors was brought up as another source for the magical machines.
It's been 46 years since my introduction to "the future," and I am still waiting.
Thank you for the information and the memories from the past.
Good share. I guess sometimes the future is.... The future. Lol
I remember that at about 1979-1980 where they said it could operate at 500 MHz when at the time we had around 2-6 MHz on our microprocessors and the Cray-1 was 80 MHz. Then it took like 14 years until the DEC-Alpha reached 200 MHz.
The Josephson junction is the main problem with the superconducting computer, but it is by no means the only one. It killed that old computer; it killed the IBM quantum computer.
IBM used to seem to have unlimited funding and to be at the center of so many technologies back then. What happened?
Great video again Jon.
Mismanagement and saturation in the computer market in the early 90's? I bet on the computer market, at that time a lot of tech gigants were born.
Also, IBM is not gone, they just went quiet to the main public as they no longer sell home computers, but the datacenter and specialize hardware world is dominated by them (up to a certain extent)
Strategic shift to software and services, while dropping low margin hardware business units.
They still are a pretty heavy hitter as R&D goes. It’s just most of what they work on is licensed to other businesses. Now, they aren’t nearly as big as they used to be. That’s true, but a lot of their tech is used as parts of most electronics you use today. It’s just their contributions make up many small parts used along many other small contributions from other companies. Whereas IBM used to make most components on their own, they are now just licensing new technologies aimed way up the supply chain from a typical consumer.
Accountants and MBAs took over from the guys who made stuff. Same story everywhere.
Same thing that happened to Boeing, the suits and bean counters took over and pushed the engineers and scientists aside from any decision making capacity.
One reason IBM's computer failed is when you cool the repairman down to cryogenic temperatures, they are unable to make repairs. The thermal stresses of warming and then cooling the circuits usually destroy them.
It's called "the path of least resistance" (pun intended) which governs pretty much everything, from nature to science. Water, and electricity will always take the path of least resistance, whether we like it or not.
@@BillAnt ...wut? Is this a joke I'm too dumb to understand?
@@m2heavyindustries378 - Move it along buddy.. Nobody asked your opinion. smh
Very interesting. In 1976 I finished my PhD in Superconductivity and interviewed at IBM T J Watson Research where I was offered a Post Doc (which I declined in favor of a permanent professional position at Shell Research), and yet had no idea of the effort IBM was putting into their superconducting computer in this timeframe. Only with your video did I learn the back story of what they were attempting, and how it all ended up (making me glad that I moved on). I still remember though, the intellectual caliber of the IBM scientists that I met there. Thank you for all the work and then sharing. I have learned so much, on a variety of subjects, from your channel.
Here in Scotland the word dram is usually used to describe a small glass of whiskey, and DRAM (pronounced Dee RAM) is a form of memory. One of those things will increase your memory, while the other will likely decrease it.
yup. worked in memory development for a time, we always called it "dee ram" and "s ram" "dram" sounds strange.
Yea, in the US, I've always heard it pronounced Dee-RAM. He does pronounce SRAM the way I've always heard it. When I was a kid I used to call HTML Hotmail 😅
No, a dram in Scotland is a small glass of whisky, not whiskey.
@@chestertonic HotMetal
If a narration pronounces it "dram" as the measure, it's highly likely the video is using TTS; there is nobody home.
No computer-savvy youTuber would pronounce the item other than "dee-ram"
Fascinating stuff. When looking at the history of technological progress, it's easy to forget the huge efforts that sometimes went into dead-end ideas.
And that is dangerous as some politicians want *_all_* research to be profitable and cut public funding following news of the inevitable and expectable failures.
This shifts research to the private sector and what is good for a specific company isn't necessarily good for society.
A very interesting video. My master's thesis was on the "punch through' problem which made resetting from the 1 state a matter of probability and not a certainty. I later was an intern in the IBM Yorktown Heights project. It was nominally intended to develop a very fast signal processor for the Navy. Josephson junctions in the voltage state have current oscillations. The frequencies are exactly proportional to the applied voltages, so they define the volt. The problem is that there is no direct control over the phase of the junction as it resets to zero and there is a metastable state at 180 degrees when you want 0. That state has a decay time, and you simply have to wait for it to "certainly" go away or you risk spontaneous 1s from your gates.
I thought you said that Dudley Buck died of an enema until I listened to that part again!
I'm really glad you have good captions. I've been seeing more and more larger UA-camrs stop using them for some reason
Electrons are half spin particles when superconducting happens a pair of them act together to form an integer spin so they go from Fermi-Dirac to Bose-Einsteinian behavior.
I'm sure that's what you tell all the girls
@@Milsparro He's fucking sure it will work next time. 😂
Yes. And that's why you have cooper pairs. It's simpler to say that it goes from Fermion to Boson behavior. No need to make it too obscure.
@@jannegrey593gotta get that wordcount
@@Mordecrox true
A room temperature superconductor (RTSC) walks into a bar.
The bartender says "we don't serve RTSCs in this bar."
The RTSC turns around and walks away, without showing any resistance.
I would love to see a video on saphire on silicon that HP designed awhile back! I find it super interesting and there is like no videos on it! Love the channel and keep up the amazing work!!!
here is an (i believe) easy analogy for broken symmetry in superconductors: when i am at my normal range of temperatures, i behave in expected ways. i put on a sweater if i am too cold, i turn on AC if i am too hot. when i am supercooled to a temperature below -40 using liquid helium, i no longer react those ways because i am dead. frozen to death is the odd, supercooled state of Me which does not conform to the laws one can derive from my normal range of temperatures. Broken Symmetry.
There were working superconducting chips in the USSR towards the end of it, but they also did not work as reliably or as fast as people had hoped. Eventually many of the researchers from Moscow went to Stony Brook University, establishing a significant superconducting electronics project there.
Some of the same know-how goes into designing chips for the quantum computers, like those made by the D-Wave in Canada.
So far superconducting circuits really shine only in very special applications, like national-level resistance and voltage standards, super-sensitive magnetic field sensors (with which one can even measure the minuscule magnetic fields produced by the working of the brain!). But for ordinary computation, silicon is hard to beat.
SFQ is great, but has significant memory issues. Especially random access is "challenging". EDA tools are also absent, some researchers here have created their lithography masks for their SFQ designs using inkscape. Though I have heard there are users in the defense sector, but have no clue what they are doing with it.
It's probably useful for extremely low latency computing
SFQ integrated circuits can be easily designed using Cadence or even open source EDA tools such as KLayout
Look at the Isake diode. Discovered in 1957. It uses tunneling at room temperature. Its speed power product is 100,000 times better than CMOS. It is trivial to make Threshold Logic Gates using tunnel diodes, reducing the gate count by about 30%. Resonant Tunnel Diode circuits today run at 33 THz with 5 W. One problem is the hyper abrupt junctions requiring tolerances of 5 atoms. I used them in 1968 but none of our test equipment was fast enough to measure the circuit's behaviour.
A friend who designs quantum computers tried to explain Josephson Junctions to me and failed. Now I finally get it. Thank you!
its interesting that current flowing forever in a superconductor ring feels 'wrong' somehow, but the idea of a planet spinning forever seems intuitive
I think in both cases, the true phrase would be " for very,very,very..................very long time"
Gallium Arsenide was another nail-in-the-coffin for JJ's. GaAsFETs were a hot research topic in the early 1980's and at the time they were often thought to outpace CMOS, and for awhile Vitesse Semi had a niche market for very high speed GaAs logic. There were so many innovations with CMOS that it's held it's ground for decades.
GaAsFET technology definitely has panned out in other ways though. These transistors are used for a lot of sensitive and low-noise analog RF circuits, for example. They didn’t work out for computers though.
used in high-end uniq military computers , at least in memory bus switches.
All hail the JJ! Electro-metrology nuts represent!
Love this channel, so many interesting and random subjects covered. But I had to lol at around 15m when the picture of dennard is on the screen, with the message in the lower left corner saying 'references and sources go here'. I guess that's a template you use a lot ;)
I remember my father trying to explain this to me when I was about ten years old.
We need a video on RSFQ pls.
You really are the best. Amazing work.
I believe you have an error regarding Dennard's law. You need about the same electric field for the switch to work, but that field should be proportional to the voltage divided by the separation between electrical contacts on the transistor. A smaller transistor requires a smaller voltage to keep the same ratio, and therefore less power to operate if the amount of current flowing is the same.
I recall that the Josephson Junction was used as a super-stable and accurate voltage reference using a frequency to voltage conversion configuration.
Being based on fundamental principles and very controllable inputs, it is still the standard.
en.wikipedia.org/wiki/Josephson_voltage_standard .
Another outcome of the Josephson effect is that it is the basis for the Josephson voltage standard - the SI reference standard for the volt.
@17:30 That’s wild, I hadn’t thought about that song in forever, but I just heard it in a bar a few days ago, and then it shows up here. What a classic.
On The name of the nobel laureate called "Giaever " as last name:
it is a Norwegian name pronounced more like "Yeah-where" with the sound of the A in "sad" instead of the E sound 'Ye' :D
so: Ja-ver , or "Ya-wher" , Yæ-wher, - ( æ or ae is the nordic char for the wovel sound in "sad" the english word for sadness :)
Glad to see D.B. Cooper makes an indirect cameo in this video...
I am not sure that the NSA was financing specifically for cryptographic purposes.
The NSA is financing mathematical stuff that have to be:
* Not cryptography. They are the expert in crypto, not the people they finance.
* Have potential application, however distant, to their work. So fine for combinatorics, but not for differential geometry.
I believe the problem in understanding symmetry is because people learn things in the wrong order.
Before understanding symmetry, you must be familiar with the Noether Theorem.
Also, it must be clear that the law of conservation of energy is NOT a First Principle; the law of symmetry IS the First Principle.
Impressed IBM stuck with it so long. Such is the nature of cutting edge research.
At 7:58 the axes are wrongly labeled. The vertical axis should be voltage and the horizontal be current. Or both are voltage, then the horizontal is the switching voltage, and the vertical is the voltage on the superconducting wire.
THANK YOU! I was starting to doubt myself
I interviewed at TJ Watson in the early 1980's for a job in the IC/transistor fabrication clean room. I spoke to one of the scientists working on the Josephson junction devices by the name of Dr. Ting. I also spoke the one of the project managers named Blum.
I didn't get the job but I guess that's okay since the project was shut down. 🥶
There was an article "A Data Center in a Shoebox" published recently by the authors from imec. It seems that there is some work continuing in this direction.
One of the difficulties to any superconductor technology is the extremely low temperatures needed to achieve the superconducting phenomenon. The technology requires the use of a liquid helium refrigerator running all the time.
I'm confused about one thing. That first superconducting switch was composed of a superconducting niobium wire wrapped in a coil of copper wire. It was supposed to be super fast switching. If it's relying on the copper wire, isn't it limited by the speed of the switching used to turn the current in the copper wire on and off, so it's still exactly the same speed as a non-superconducting switch.
What have I missed?
IBM had guts to be a pioneer. Now there is only fiscal economic policies and the company is in ruins.
Hewlett Packard Labs was developing the “memristor” but managed to keep it only in the lab and not able to make any practical, manufactured devices.
The failure of the Josephson junction computer was a terrible blow to the scientists and engineers at IBM. It was a awful feeling. IBM was so invested in it.
This is really interesting and I am literally eating popcorn while watching this 🍿
2024, they have improved them a lot. They can now make these superconductor insulator superconductor devices operate up to the THz range. While not practical as a computer element, they are used for extremely high gain low noise mircowave amplifiers mixers and otherther devices for radio astronomy and other extreme mission critical uses. The speed change is mainly due to improved fabrication technique and metal oxide superconductors, thout they still use helium cooling as a means to keep the noise floor low. ❤
this was the 1960s so they thought they were just going to be able to put in on Pluto in the year 1980 so cooling isn't an issue anymore but when that didn't happen the project fell apart
I wouldn't rule out someone figuring out how to make these practical, but I also won't be holding my breath waiting.
I think this was an informative video, but at 7:57 , I think you switched Voltage and Current in the shown Graph. I think the Current should spike at around 0V , taper of to the current density and then should slowly rise again with higher Voltage.
Exactly. The graph is not showing what the narration is saying.
awww man that beeping is killing me
Awesome video, thank you
The only place I can see these things used is in quantum computer controllers. The main event needs milikelvins, so putting the controller chip right outside of it at 4K would not be too much of a hassle.
Edeeema. At first I thought you said enema.🙂
By far the most common avenue towards quantum computers is by using quantum bits based on Josephson junctions.
Wow. I just happened to have this pop up as a suggested video less than 15 minutes after it came out!
wow man that's so amazing
@dziban303 Hey, in the world of Asianometry, when you're not a subscriber, it actually is pretty amazing! 🤣
I think you should make a video on SMIC's quadruple patterning. This is just mind boggling.
Open ended pipe, with water only half filling the full diameter of the pipe…. Is gravity driving the flow, not so much pressure driving the flow…. 😃
But, point well made…
Happy Sunday!
Go Asionometery!
Brian Josephson still alive!!!
And only 84: en.wikipedia.org/wiki/Brian_Josephson
RSFQ has not become the dominant supercinductor digital logic because of a high static power dissipation, parallel dc biasing, and insufficient energy consumption advantage over CMOS if cryocooling penalty is taken into account. It is enormously fast, though, with complex circuits clocking at 50 -70 GHz and above.
Saying photons gain mass is slightly misleading. Instead I would say gains momentum or energy rather than mass. Quantum tunneling allows electrons to exist in same space simultaneously… so fast it makes mass nearly meaningless.
1:15 electrons moving in a circle should lose energy due to synchrotron radiation.
Interesting to learn about the cryotron and its role in early computing. With advancements in both superconductors and silicon, how do you see the future of these technologies evolving in parallel?
You got me with the Brian McKnight reference.😂
Heyo, a small mistake at the 8:16: The characteristic curve shows the current on the X-axis and the voltage on the Y-axis, otherwise the zero-voltage condition would not apply. Maybe try to correct that:)
Completely wrong on Josephson junction switching. The S to resistive (N) state switching is very fast, now subpicosecond. The R to S switchinf is slower due to a slower discharge of the junction capacitance. Easily can be made fast, 1-3 ps, by resistive shunting of the junctions
I always laugh when you call d-ram dram, like its a Whiskey or something lol :)
Almost all the modern quantum computers use josephson junctions
Fascinating as always 👍
Do you think we'll ever have solid state cooling capable to making a mobile phone's internals cold enough to exploit the resistance drop?
Close to my PhD work on cryogenic STT+SOT MRAM 😀
Another great video, thank you 🙂
Thomas the tank got a Super Conductor..... Choo Choo
Yay !
Ringo Starr !
Kind of unrelated but what keeps something floating and spinning on a super conductive magnet in a vacuum from violating the second law of thermodynamics? I know magnetic fields do no work thus the floating does not use energy, but assuming the item is spinning, it has kinetic energy. That movement creates flux in the magnetic field and thus some work is done. Does that flux create a tiny amount of heat in the magnet itself and therefore the magnet will get hotter and eventually become non superconductive (without active cooling and thus energy being removed from the system [which in turn uses energy])? I am not an expert at this at all, so I may be wrong but for some reason I thought of it while watching the beginning here. Anyone know for sure?
God I love these videos.
What about the use of the Josephson Junction as a very accurate and fixed voltage source for calibration purposes? I think NIST uses them for this purpose.
I wonder.... will the cryogenic computing (and all the "hardware stuff" overhead, say pumps etc) fail again and in a similar way in the era of quantum and pqcrypto?
Did they figure out the reason they randomly flip? Anything to do with what's happening nearby?
Josephson junctions seem so promising; it's too bad manufacturing physics made it untenable.
True. Like photonic switches and fiber optic busses, also ahead of their time though gradually making inroads.
I believe the chart at 08m23s has wrong orientation. It shows zero current at different voltages at superconducting. It should be zero voltage…
Excuse me sir, i was wondering if superconducting materials have a linear shift in resistance as temperature decreases past the transition temperature
Dudley Allen Buck The Cryotron Files by Iain Dey and his son Douglas Buck, He died just weeks after his 32nd birthday, Louis Ridenour died the same day.
Oooowee, love your videos, ty so much ❤️
Was the memory limited to 1000 k bits as shown on screen, or k bytes as in the voice version?
Wait.. Josephson Junctions and the physics for it is 1930s/60s old? I had no idea. I known they have been used in some various research applications
And superconductivity was discovered in 1911.
Could you do a video on racetrack memory? Maybe it's too early.
You should review IBM's quantum computer ideas. After 40+ years of improvements, there is little to show. This will never change. If IBM people tell you otherwise, they are just full of shit.
The superconducting computer and almost anything with Josephson junctions works, nearly.
I wonder if this could be revived to work with Quantum computers...
Wait, turned Sultan? What
Great vid. Interestingly enough it was the Josephson supercomputer computer idea that led to Heinrich Rohrer, Gerd Binnig, Edmund Weibel, and Christoph Gerber into inventing the nobel prize winning scanning tunneling microscope as described in Cyrus Mody's book Instrumental Community.
Or maybe it did work but the only working one went to the NSA?
bandgapped computing and graphene semiconductor advancements have made all of this pointless and paved the way to create true qbits :TLDR
They should have triet put it down in the ocean bottom to reduce noise from cosmic rays.
Microsoft tried submerging computers in the sea (in a huge metal tank filled with oil inside), but it was for improved cooling, not for reducing cosmic ray-induced errors
Bryan McKnight wow haha good one
Thanks!
Baffling, how did anyone not easily see from the beginning that a computer that needed to be submerged in a liquid helium would never achieve mass adoption?
It was the 1960s. Computers were the size of refrigerators; the PC revolution and smartphones were not on anyone's radar.
@@fwiffo
Those computers never needed anything like liquid helium, and they knew what liquid helium is, it never is a substance that the average Joe has to deal with in every day.
"Bell Labs" again!!!!!
"Asian parent's dream" lol
By the way, did you ever make a video about Silicon Carbide? It's an interesting material with a lot of history.
Shame. Picoseconds switching is ~1000× faster than current transistors, that would be revolutionary!
Speaking of supertech that didn't go anywhere, what happened to memristors?
They are doable - I read that someone made them at home from sulfur compounds - but it appears they can't be easily and reliably mass-produced like other types of memory
Thanks 👍
I believe at 8:20 the Current and Voltage axis labels are swapped from what they should be?
Yes.
Спасибо за обзор, очень интересно получилось)
How much did SEEQC sponsor?