Take this $15m and make us some ultra-energy-efficient superconductor chips, scientists told

Researchers in the US have received a $15 million National Science Foundation (NSF) award to develop superconductor chips that ought to be much faster and use significantly less energy than the hardware the world today relies on for computing. A team at the University of Southern California's Viterbi School of Engineering is …

[Adam Azarchs]

"a fair price"

Obviously tung in check, but that tends to be how early stage basic research goes. You keep the spending low until you know enough to know more specifically what you should be prioritizing. A lot of early stage ideas are never going to go anywhere. And once the researchers have a proof of concept, there's still a much, much larger amount that will need to be spent to get it to commercial scale production.

[Persona]

Re: "a fair price"

If it looks like its going to work IBM will spend serious money researching it. If it's a commercial possibility Intel will spend vast amounts of money getting it there.

[David M]

Déjà vu

I'm sure I remember Josephson junctions being 'the next big thing' back in the ... 90s, or 80s, was it? It's nice to see them back again. But I'm now wondering how big your supercomputer has to be for the energy saving of this new technology to outweigh the energy cost of the substantial cryogenic cooling plant that would be needed to get everything down to 4.2K. It will be interesting to see how this develops.

[Pascal Monett]

Re: Déjà vu

Indeed.

I once had the opportunity to go do helpdesk stuff in a warehouse that stored milk products (ice cream, etc). It was cooled to -30°C.

It was in winter, outside temperature was around 0°C, but sunny. I spent all of twenty minutes there.

When I left the place, I had to take my coat off. For a few minutes, I felt like I was on the beach, what with the sun and all.

Of course, 0°C caught up with me pretty quick and I put my coat back on after a few minutes.

Still, I don't think there'll be a lot of helpdesk people poking around racks that are cooled at 4.7°K.

I can't imagine the amount of energy that is going to be needed to cool an entire server farm down to that point.

[ThatOne]

Re: Déjà vu

> I can't imagine the amount of energy that is going to be needed to cool an entire server farm down to that point.

Came here to say the same thing. I have a feeling their emphasis on "energy efficiency" is just hollow marketing speech: They conveniently ignore the extraordinary amount of energy you'll have to spend to keep those superconducting processors at the required 4.7 K.

4.7 K is really, really cold, keep in mind liquid nitrogen is a balmy 77 K. IIRC only liquid helium would be able to do here, and only inside a multi-stage and multi-shell insulation, which won't be cheap to build and operate.

[Ian Johnston]

Re: Déjà vu

Keeping stuff cold isn't a huge problem. Sure, the cryostat will be a bit complicated, but MRI scanners only lose about 4% per month. Since they typically contain 1700 litres, which is about 200 kg, and the enthalpy of vaporisation is 23.3 kJ/kg, that's 0.04 * 200 * 23.3 = 186.4 kJ/month = 70 mW.

On the other hand, superconductors are only lossless for persistent DC currents. AC use is lossy, because the flux in them changes, which produces a voltage, which combines with the current to give a power loss, albeit a non-resistive one.

Spent several years of my life a developing a standard test for AC losses in superconductors ...

[Cuddles]

Re: Déjà vu

"Keeping stuff cold isn't a huge problem. Sure, the cryostat will be a bit complicated, but MRI scanners only lose about 4% per month. Since they typically contain 1700 litres, which is about 200 kg, and the enthalpy of vaporisation is 23.3 kJ/kg, that's 0.04 * 200 * 23.3 = 186.4 kJ/month = 70 mW."

Not really sure what you're trying to calculate here. Heat isn't lost from coolers due to loss of coolant (unless something goes really horribly wrong). Your fridge loses close to 0% of its coolant per month, but you'll notice the back still gets quite hot. That's because cooling one place can only be done by using energy to transfer heat from one place to another, and that energy is itself converted to more heat in the process. Nearly 50% of the power use of an MRI machine is the cooling system. You're around 5-6 orders of magnitude out on how much power is required for cooling.

To take an example, the LHC uses around 40MW for the cooling system that uses 120 tonnes of helium. A similar system scaled down to 200kg would then take 66 kW. That's likely a bit of an overestimate since a widely distributed system will have more losses. But even assuming an order of magnitude gain in efficiency for a small system like an MRI, you're still looking at at least kW of cooling power. If you want to build an exa-scale supercomputer, you'll be back up into the MW region again.

[RobThBay]

Re: Déjà vu

Why not put it in orbit? The James Web telescope is operating at these supercold temps.

The next problems would be power and data transfer rates.

[Eclectic Man]

Re: Déjà vu

Sadly even the JWST needs a cooling device to reach below 7K for its experiments:

https://webb.nasa.gov/content/about/innovations/cryocooler.html

"Webb's MIRI instrument carries detectors that need to be at a temperature of less than 7 kelvin to operate properly. This temperature is not possible on Webb by passive means alone, so Webb carries an innovative "cryocooler" that is dedicated to cooling MIRI's detectors."

However, I wonder whether these Josephson junctions could be made using ceramic 'high temperature' superconductors.

https://en.wikipedia.org/wiki/High-temperature_superconductivity

[ThatOne]

Re: Déjà vu

> Sadly even the JWST needs a cooling device to reach below 7K for its experiments

Indeed. First of all, space isn't really "cold", it's rather characterized by "absence of heat", big nuance. Second, vacuum is a very bad heat conductor, which means that whatever heat you create will be a pain to get rid of. Unlike on Earth, where air convection eventually carries off any excess temperature, in space heat can only radiate off, very slowly...

You'd have a hard time keeping temperatures as low as 4-5 K. Don't forget that's only a couple degrees above the absolute temperature minimum, which means about everything around you tends to be hotter.

[Tom 7]

Re: Déjà vu

The JWST just needs to be colder than the electromagnetic radiation it is trying to receive. Josephson Junctions rely on a superconduction. We've seen that at 203k!!! Quantum Computers might need 4.2k but a Josephson Junction computer can run a lot warmer than that. If its power loss is low enough (and if the whole thing is superconducting then we're talking EM radiation alone) then its quite feasible to run a whole data centre in a very small volume with very little cooling required compared to a 'full size data centre'.

[Eclectic Man]

Re: Déjà vu - ASIDE JWST mirrors in alignment

See the image here from ESA:

https://www.flickr.com/photos/europeanspaceagency/52036285993/in/feed-37440125-1651169895-1-72157721616820865

"Alignment of the James Webb Space Telescope is now complete. After full review, the observatory has been confirmed to be capable of capturing crisp, well-focused images with each of its four powerful onboard science instruments.

Upon completing the seventh and final stage of telescope alignment, the team held a set of key decision meetings and unanimously agreed that Webb is ready to move forward into its next and final series of preparations, known as science instrument commissioning. This process of setting up and testing the instruments will take about two months before scientific operations begin in the summer."

[Tom 7]

Re: Déjà vu

IIRC Josephson's Junctions need superconduction not 4.2k so we could be up to 203k or so. The thing to remember is the bigger you go the less it costs per unit volume to keep it cool. They're up to a million or so junctions on a chip at the moment so around 68020 level of complexity so I'd bet we're close to a point where we could see some useful stuff in the pure digital realm - I think most of the cryogenic goes to quantum computing stuff but there does seem to be a digital option opening up as a result, I cant find any figures on gate switching power but I can imagine it could be very low and it might be possible to replicate a whole datacenter computing power in a small cool room.

[Neil Barnes]

Re: Déjà vu

But I don't know enough to know whether I need to worry about the energy required to make a superconducting region change state. After all, bog standard CMOS uses very little power when it's not being clocked; it's the state changes that suddenly eat power.

[vtcodger]

Ones? Zeros?

and they store logic values of zeroes and ones

Real distinguishable ones and zeros? Not quantum? How 2020.

[DS999]

If this works out

Look for the major commercial application to be wasting electricity mining bitcoins.

[AnoniMouse]

The speed of light remains unchanged

>> ... the team thinks the SuperSoCC could provide at least 100 times faster performance at the same energy level as CMOS-based chips.

As Conway and Mead articulated clearly decades ago, a critical factor for performance is "distance": the time it takes for a signal to traverse a given distance imposes limits on performance for any given feature size. Even with X-ray lithography we are approaching physical limits of practical feature sizes.

[Richard 12]

Re: The speed of light remains unchanged

It doesn't have to all clock together.

At the dawn of computing there was "delay line" memory, where you clocked data in at a far higher frequency than the propagation time along the delay line, thus effectively storing bits along its length. Briefly.

Perfectly feasible to have a pipeline of multiple clock cycles across the chip, spaced purely by the propagation time.

Pretty sure some chip designs already do that.

[Big_Boomer]

Moore's "law"

I'm sorry, but this was NEVER a law or even a guideline. At best it was one mans prediction/guess as to how CPU processing capabilities would evolve over time. For a few years it was a self-fulfilling prophesy that certain companies felt that they had to follow. They might as well have been basing their progress on horoscopes or cafeteria tea leaf readings. I'm not taking anything away from Gordon Moore's achievements but one prediction does not define the man. As for superconducting CPUs, first we need high-temperature Super Conductors. Cooling stuff to -umpty degrees is impractical and VERY expensive.

[Torben Mogensen]

The theoretical limit

The article says that they might be "approaching the theoretical limit of energy efficiency". I assume they mean the Landauer limit, which puts a lower bound on dissipation when performing operations that lose information. For example, an AND gate has two inputs and one output, so it loses a bit over one bit of information on average (9/16 bits to be precise). But it is possible to go under this limit if you don't lose information, i.e., if all operations are reversible. In this case, there is no lower limit. And you can do surprisingly much with only reversible operations -- though you sometimes need to output extra "garbage" bits in addition to the result bits to preserve reversibility.

Still, the Landauer limit is several orders of magnitude under the current energy use of semiconductor gates, which also dissipate more in wires than in gates, so superconducting wires and gates will reduce the power dissipation quite radically. As for cooling, this is proportional to power dissipation, so while superconducting chips need to be cold, they don't tend to heat their environment very much, so in a well-insulated container, the cost of keeping them cold may not be too bad.

[Stuart Halliday]

$15 million won't pay for much

What are they hoping to fund with just $15 million. That'll hardly pay for the Coffee machines...