It isn't clear to me from the press release whether these are similar to D-Wave's quantum annealing qubits, or if these are another type (e.g. more like a "true" qubit). Is there another source with more information?
Over at the Wikipedia it's reported that the latest D-Wave products have 2048 qubits. So, I would hope that announcing a processor with only 17 means a fundamental shift.
Of course, it could be marketing running away with it.
These are transmon qubits... they're designed for long coherence times, in other words, to store quantum information for long periods of time. D-Wave qubits do not have long coherence times-- by D-Wave's own admission. They argue that long coherence times are not necessary for quantum annealing. Many don't buy this argument.
After 20 years in VLSI, I have no clue about quantum computers, and haven't found a good source explaining them. Anyone have any idea how they work?
Like, how does one produce logic gates out of them? How do you create an IEEE 754 Floating point multiplier? I get that they're supposed to represent overlaid states on the same device due to Pauli exclusion, but how do you separate out states and use them in logic? Or save states? What happens if logic produces the same state for one device? Is standard metal used to send signals, or is it something fancy-dancy like neutrinos?
It's honestly the most exotic thing I know of in tech, just trying to even begin to get a handle on it, never-mind trying to perform useful work out of them.
You are not supposed to reproduce classical systems on a quantum computer just the same way you're not supposed to reproduce electric circuits on a classical computer, even though it's made out of them.
For me myself one of the easiest concepts to understand was quantum annealing (https://en.wikipedia.org/wiki/Quantum_annealing). It does not mean that's the strongest point of quantum computers, but it's a nice way to start.
I get where you're coming from with that analogy, but there is probably a better way to put it. Circuit simulation is a long-time application of computers (https://en.wikipedia.org/wiki/SPICE).
You're down the Turing machine rabbit hole - I implied you could do that. Just the same way you could simulate a processor using an another processor - it's just not as efficient.
By the way - you could as well simulate a quantum computer on a classical computer; it'd just be incredibly slow.
There are lots of different ways to build a quantum computer, so you might get conflicting answers about the materials and signaling mechanisms. But what they have in common is that they are probabilistic, and there are very few algorithms that take advantage of this to speed things up instead of just adding noise. Shor's algorithm for factoring numbers and Grover's search algorithm work by constructive and destructive interference of the wavefunctions of entangled particles, amplifying the probabilities of the right answers. http://twistedoakstudios.com/blog/Post2644_grovers-quantum-s...http://algassert.com/post/1718
Quantum computations are a case of reversible computations. If you're aiming to relate qantum computers to other kinds of computers, I suggest starting with the paper called "Synthesis and Optimization of Reversible Circuits - A Survey" [0]
It describes the kinds of computations that need reversible computations and kinds of devices that can perform them. That makes it easier to notice similarities, as quantum computers are not the only ones in this category. Notably, photonic and low power computations call for reversibility.
The first couple of pages of the paper also touch upon what can be done with them - while you don't have to throw away all you knew about logic gates, the reversible ones are their own thing.
If the water goes too deep, perhaps Mika Hirvensalo's "Quantum computing" helps. It approaches the topics of how to compute with qubits from mathematical side.
I still don't know the theory behind quantum annealing though - I don't think these two cover it.
Still, I learned that the main difference is that the state doesn't have to travel as with electronic circuits, but rather one would likely "apply" gates as "events" acting on the state. A qubit could then be any quantum object - a electron trapped in one place and being hit with photons for example. Place a couple together and hit them in a coordinated way and that event was your gate.
When it comes to saving states, it will be problematic. Of course, it's possible to "save" a qubit to another qubit - with some workarounds for the "no-cloning" principle.
But if you mean saving them to a classical system, it seems it's not possible at all. A qubit is a representation of a probability, and once read it's destroyed. You could try recalculating it several times and get an idea of the probability it represents. However, qubit readouts are not guaranteed to be independent from one another, giving another bit of headache here.
Disclaimer: I'm not a physicist - just a software engineer trying to understand the hype.
Ask you shall receive. Total these videos are about 30m but they're designed for someone that has taken Intro to Computer Engineering and Intro to Logic.
After reading the press release it seems to me its a Topological Quantum Computer[0].
They mention surface codes which are related to toric codes
and toric codes are a part of topological quantum computation [1]. A explanation of toric codes can be found at [3]
Their website says that they specialize in Topological Quantum Computers and Quantum Error correction (which are a feature of Topological Quantum Computers.
Also the link to the researchers website mentions similar projects . [2]
Slightly unrelated question but does anyone knows how quantum computing will change if Copenhagen interpretation is invalid and instead De Broglie–Bohm theory proves to be correct?
Yes, but if one interpretation is deterministic and one probabilistic doesn't it mean that the mechanics will change, and what we currently think is how quantum computers could work, will potentially make them faster/slower/impossible?
Pilot wave just moves the quantum magic/weirdness/conspiracy to a particle whose trajectory we cant see, that "rides" the pilot wave in a manner that grossly violates our notions of cause and effect. Under pilot wave theory, you can still build the same computer, but the explanation for the magic efficiency is explained in terms of the magical movements of the rider particle.
It's irrelevant. Interpretations try to create a model that explains why something acts the way it does. Under pilot wave theory, quantum particles act the same way they would under the Copenhagen interpretation.
I was under the impression that those theories agree about their predictions. Their relative merits come down to how well they explain things and how well those explanations jive with existing science.
So either:
- I'm wrong, and some phenomenon exists that can differentiate the theories.
- The only thing about quantum computing that will change if the Copenhagen interpretation falls out of favor is the contents of the textbooks
From what I've read (admittedly not enough) Pilot Wave theory is very deterministic and have some other weird quantities for example the observer does not really affect the result of double-slit experiment (if I understand correctly)
Thank you for asking this. I've been wondering about it for several years. My assumption has always been that if the latter is correct, then some kind of exponential increase in time to get a solution would be involved. I'd love someone who actually knows something about physics to give an answer that's understandable to people who haven't studied the subject since high school 30 years ago.
I have to disagree. EM drive is theoretically impossible with Copenhagen interpretation but with Pilot Wave not only it's possible but we can improve its efficiency. It would make sense that similar situation is possible with quantum computing.
We have no idea why the EM drive works. It doesn't seem to make any sense as it seems to violate conservation of momentum, but experiments have the ultimate say.
Pilot wave theory is actually a small change to the mathematical model in QM, but it is one which never affects anything that you can measure in the lab.
It is correct to say that all interpretations lead to the same predictions, but not actually correct to say they're the same mathematical model.
The Copenhagen interpretation is trivially invalid in the sense that it is incomplete: it doesn't provide a clear definition of what a "measurement" is.
The De Broglie-Bohm interpretation doesn't have this problem, because it does not distinguish between quantum interactions and "measurements"; they are one and the same.
Given the quantum computing chip space is getting increasing heated in competitions[1], can someone with industry experience and knowledge to compare Intel's chip with D-Wave's, Alphabet's and IBM's?
Hopefully someone can expand on these comments since I'm not an expert, but I know a little.
D-Wave [1] is shipping quantum computers with thousands of qubits. The difference seems to be that D-Wave qubits are not generalized qubits, but rather they are simply designed to solve problems using quantum annealing [1]. Annealing is just one algorithm which efficiently (but probabilistically) looks for the global minimum of a function, but it is prone to getting stuck in local minima. Quantum annealing exploits quantum phenomena to do this better and probably more efficiently [2].
On the other hand, we have "general" qubits: these can implement and carry out any arbitrary quantum computing algorithm (for example, Grover's search algorithm or Shor's factorization algorithm). It seems to me that researchers at Alphabet, Rigetti, Intel, IBM, etc. are trying to build these general qubits, since that's where we can finally unlock the full power of quantum computing. It's also much harder to build these general qubits, and that's why we only see 17 qubits announced here, or Google claiming to be working on 49 qubits [3].
The numbers come from quantum error correction and the goal to build a logical qubit (error protected qubit made of many actual qubits). In the surface code (planar version of a toric code), a single logical qubit protected with a code distance d = 3 corresponds to 17 physical qubits, and a distance d = 5 corresponds to 49 physical qubits.
I heard someone say that once we got to 50 qubits, we would be able to unlock the potential of quantum computers/actually solve problems that are nontrivial for traditional computers. If that's the case, why is google stopping just before that threshold?
What they're really proud of is essentially bringing Intel's manufacturing advantage to bear on an extremely new and different form of computing. Being able to iterate with new materials and architectures every few months is critical for testing, and it's there (as opposed to sheer numbers of qubits or whatever) that Intel hopes to outstrip the competition.
I was also surprised to learn that the cryo systems that get things down to a fraction of a kelvin are only the size of 50-gallon drums. I was picturing something much larger for anything that low. Now I know!
Would be fascinated to hear anything you (or anyone in a related field) has to say on the challenges for cooling to such low temperatures.
Is it essentially just heat-removal + lots of layers of isolated, effective insulation, or are there fundamental differences when you're going that low?
I'm not sure about what's used in all industry applications, but this video explains one implementation of a super-cooling setup: https://www.youtube.com/watch?v=7jT5rbE69ho.
I won't even try to summarise it - I'm sure I'd get it wrong! Fascinating stuff though.
I worked with NMR about 8 years ago in the uni. We used regular liquid helium to get to about 4.2 degrees K. The next step is He 3 isotope which is an order of magnitude more expensive but gets you an order of magnitude cooler (in degrees K). That usually runs in a closed system (heated He is not vented into the atmosphere) because of its cost.
The way cooking it was explained to me is that you try to slow down atoms with lasers, but I don’t know any of the details.
From memory: you fire a laser at a frequency that is only absorbed if relatvistically shifted the way an electron that goes towards the laser would observe. This, the laser delivers momentum only to atoms coming towards it, and is invisible to those at rest or going away. Do from multiple directions and you’ve reduced absolute momentum in every direction.
Claude Cohen danucci received a nobel and a wolf for this (and other things)
I work in this exact field. Yes. It's essentially a thermos. There's several cans nested within each other. Each can hangs from a different refrigeration stage. Typically you go from 300 K (the outer vacuum can), to 77 K (first pulse tube stage), to 4K (second pulse tube stage), then to 300 mK (still of dilution unit), then to base at 10 mK, though the 10 mK stage doesn't have it's own can. Everything is copper or gold plated copper where you want heat to flow, or G-10 (fiberglass) or stainless steel where you don't. There's mechanisms for thermally shorting stages together... sometimes it's a mechanical gripper that pinches an extension from a colder stage, other times it's stainless steel tubing that gets filled with gas (the heat conducting material) when you want to transfer heat, and then evacuated when you don't.
Wow. I could see something that size as per-workspace for specific use cases. Though what those use cases might be, I don't know. (Also, cost is probably the prohibitive factor right now...)
Still, I'd always imagined that quantum computers would start out as something that could only be done mainframe-like due to size. Perhaps not!
We owe it to Particle accelerators and MRI machines to let us cool to those temperatures. I also find it surprising that it is as small as a 55 gallon drum.
Amazing how this has progressed: in high school I was in a summer program with the guy who won the Science Talent Search that year with the research he was collaborating on in Nuclear Magnetic Resonance. (Note that the "Nuclear" has been removed from the tech for essentially PR reasons).
> I was also surprised to learn that the cryo systems that get things down to a fraction of a kelvin are only the size of 50-gallon drums. I was picturing something much larger for anything that low. Now I know!
Contrast that to the cooling D-Wave uses. It's not all cooling but it gives you an idea of the cooling apparatus: https://youtu.be/zDotDiK2UuY?t=50
In any practical chip design, isn't one required for the other? i.e. it is a superconducting chip, and that is a major engineering problem and a key difference between this and a semiconducting chip. Superposition just happens to also be how a qubit works.
> In any practical chip design, isn't one required for the other?
No. Superposition (in a qubit sense) is not necessary for superconductivity. Most quantum computer designs require superconductivity, but it is not a strict requirement. Photon based quantum computers don't need any superconductivity iirc.
> it is a superconducting chip, and that is a major engineering problem
Superconductivity is not a super high bar. The bigger issue from quantum computing is coherence time; and it just happens that low temperatures are critical for lengthening it.
Well, coherence and superconductivity are closely related. Superconductivity implies no interaction with the environment, which is necessary and sufficient to maintain superpositions.
> Superconductivity implies no interaction with the environment, which is necessary and sufficient to maintain superpositions.
I am not sure what you mean here. Superconducting magnets are often used to control plasma (part of the environment) so it doesn't imply isolation from the environment.
Superconducting qubits are only superconducting because josephson junctions happen to be good candidates to make qubits out of. Ion traps (which generally have much better coherence times than superconducting qubits) and photon based systems don't need superconductivity.
What I mean is that waste heat is equivalent to decoherence. If you use superconductors to do work, then they are no longer superconducting - that work introduces resistance. To do otherwise would violate conservation of energy.
I'm not sure how to make it more clear, but decoherence happens because some energy / information is transmitted into the environment. That loss of information is why superpositions collapse.
Maybe the confusion is what I mean by superconducting - I mean zero resistance circuits. It's not about what it's made of.
Well the point is whether or not this is a superconducting chip. It is. Since it is a superconducting chip, I'd like to know what OP meant when they said they mixed it up, because I don't see where they said superconducting where it isn't true and relatively relevant.
edit: also, I meant superconducting was (practically) required for superposition, but not vice-versa. I guess I misused the phrase one-required-for-the-other.
> Since it is a superconducting chip, I'd like to know what OP meant when they said they mixed it up, because I don't see where they said superconducting where it isn't true and relatively relevant.
I believe that everywhere superconducting was used it was correct and using superposition instead would be wrong. So I agree with you there.
Cooper pairs are merely the charge carriers in a superconductor. One stores quantum information in the oscillating supercurrent and voltage of a superconducting nonlinear resonator. The nonlinearity is contributed by a Josephson junction. There is a qubit called a cooper pair box but it’s not really used anymore.
I think there are analogies to the article the other day on fear-driven investments in automation industries. As well as post-WWI naval construction.
To wit, can they afford not to sink money into it, if their competitors are, even if success if less than certain?
The rewards that seem like they would be realized are generational enough so as to render much of the existing market obsolete. So to be caught without any expertise, on a risk-basis is more costly than seemingly dumping money down the drain.
If you measure the qubit prematurely the function collapses. In classical computing it might be analogous to breaking out of a for loop before it has run its intended course and then you're left with an incorrect (or incomplete) answer.
Or, more precisely, alters entanglement in an unkown way. Since the alteration is unkown, it interferes with execution and interpretation of the quantum algorithm.
Another interesting piece of news this week is that it seems that even 56-qubits can now be simulated, so the 49-quabit quantum computer Google planned to release this year likely won't achieve "quantum supremacy" anymore. To be really sure, we may have to wait for the 100-qubit one first, before declaring quantum supremacy.
80 comments
[ 5.4 ms ] story [ 146 ms ] threadBest contenders I see so far are supersingular isogeny ECC and NTRU or other lattice based algorithms.
Of course, it could be marketing running away with it.
Like, how does one produce logic gates out of them? How do you create an IEEE 754 Floating point multiplier? I get that they're supposed to represent overlaid states on the same device due to Pauli exclusion, but how do you separate out states and use them in logic? Or save states? What happens if logic produces the same state for one device? Is standard metal used to send signals, or is it something fancy-dancy like neutrinos?
It's honestly the most exotic thing I know of in tech, just trying to even begin to get a handle on it, never-mind trying to perform useful work out of them.
For me myself one of the easiest concepts to understand was quantum annealing (https://en.wikipedia.org/wiki/Quantum_annealing). It does not mean that's the strongest point of quantum computers, but it's a nice way to start.
By the way - you could as well simulate a quantum computer on a classical computer; it'd just be incredibly slow.
It describes the kinds of computations that need reversible computations and kinds of devices that can perform them. That makes it easier to notice similarities, as quantum computers are not the only ones in this category. Notably, photonic and low power computations call for reversibility.
The first couple of pages of the paper also touch upon what can be done with them - while you don't have to throw away all you knew about logic gates, the reversible ones are their own thing.
If the water goes too deep, perhaps Mika Hirvensalo's "Quantum computing" helps. It approaches the topics of how to compute with qubits from mathematical side.
I still don't know the theory behind quantum annealing though - I don't think these two cover it.
Still, I learned that the main difference is that the state doesn't have to travel as with electronic circuits, but rather one would likely "apply" gates as "events" acting on the state. A qubit could then be any quantum object - a electron trapped in one place and being hit with photons for example. Place a couple together and hit them in a coordinated way and that event was your gate.
When it comes to saving states, it will be problematic. Of course, it's possible to "save" a qubit to another qubit - with some workarounds for the "no-cloning" principle.
But if you mean saving them to a classical system, it seems it's not possible at all. A qubit is a representation of a probability, and once read it's destroyed. You could try recalculating it several times and get an idea of the probability it represents. However, qubit readouts are not guaranteed to be independent from one another, giving another bit of headache here.
Disclaimer: I'm not a physicist - just a software engineer trying to understand the hype.
[0] https://arxiv.org/abs/1110.2574
* https://www.youtube.com/watch?v=F8U1d2Hqark * https://www.youtube.com/watch?v=ZoT82NDpcvQ
They mention surface codes which are related to toric codes and toric codes are a part of topological quantum computation [1]. A explanation of toric codes can be found at [3]
Their website says that they specialize in Topological Quantum Computers and Quantum Error correction (which are a feature of Topological Quantum Computers. Also the link to the researchers website mentions similar projects . [2]
[0] https://en.wikipedia.org/wiki/Topological_quantum_computer
[1] https://arxiv.org/pdf/0904.4165v2.pdf
[2] https://qutech.nl/roadmaps/ http://dicarlolab.tudelft.nl
[3] https://physics.stackexchange.com/questions/29310/what-is-co...
So either:
- I'm wrong, and some phenomenon exists that can differentiate the theories.
- The only thing about quantum computing that will change if the Copenhagen interpretation falls out of favor is the contents of the textbooks
It is correct to say that all interpretations lead to the same predictions, but not actually correct to say they're the same mathematical model.
The De Broglie-Bohm interpretation doesn't have this problem, because it does not distinguish between quantum interactions and "measurements"; they are one and the same.
[1]: https://www.bloomberg.com/news/articles/2017-09-13/ibm-makes...
D-Wave [1] is shipping quantum computers with thousands of qubits. The difference seems to be that D-Wave qubits are not generalized qubits, but rather they are simply designed to solve problems using quantum annealing [1]. Annealing is just one algorithm which efficiently (but probabilistically) looks for the global minimum of a function, but it is prone to getting stuck in local minima. Quantum annealing exploits quantum phenomena to do this better and probably more efficiently [2].
On the other hand, we have "general" qubits: these can implement and carry out any arbitrary quantum computing algorithm (for example, Grover's search algorithm or Shor's factorization algorithm). It seems to me that researchers at Alphabet, Rigetti, Intel, IBM, etc. are trying to build these general qubits, since that's where we can finally unlock the full power of quantum computing. It's also much harder to build these general qubits, and that's why we only see 17 qubits announced here, or Google claiming to be working on 49 qubits [3].
[1] https://www.dwavesys.com/sites/default/files/D-Wave%202000Q%...
[2] https://en.wikipedia.org/wiki/Quantum_annealing
[3] https://www.bloomberg.com/news/articles/2017-07-17/google-s-...
[1] https://en.wikipedia.org/wiki/Toric_code
[2] https://www.nature.com/articles/s41534-016-0004-0
I heard someone say that once we got to 50 qubits, we would be able to unlock the potential of quantum computers/actually solve problems that are nontrivial for traditional computers. If that's the case, why is google stopping just before that threshold?
https://techcrunch.com/2017/10/10/intel-moves-towards-produc...
What they're really proud of is essentially bringing Intel's manufacturing advantage to bear on an extremely new and different form of computing. Being able to iterate with new materials and architectures every few months is critical for testing, and it's there (as opposed to sheer numbers of qubits or whatever) that Intel hopes to outstrip the competition.
I was also surprised to learn that the cryo systems that get things down to a fraction of a kelvin are only the size of 50-gallon drums. I was picturing something much larger for anything that low. Now I know!
Is it essentially just heat-removal + lots of layers of isolated, effective insulation, or are there fundamental differences when you're going that low?
I won't even try to summarise it - I'm sure I'd get it wrong! Fascinating stuff though.
The way cooking it was explained to me is that you try to slow down atoms with lasers, but I don’t know any of the details.
Claude Cohen danucci received a nobel and a wolf for this (and other things)
Wow. I could see something that size as per-workspace for specific use cases. Though what those use cases might be, I don't know. (Also, cost is probably the prohibitive factor right now...)
Still, I'd always imagined that quantum computers would start out as something that could only be done mainframe-like due to size. Perhaps not!
Contrast that to the cooling D-Wave uses. It's not all cooling but it gives you an idea of the cooling apparatus: https://youtu.be/zDotDiK2UuY?t=50
Edit: I was incorrect. Today I learned that Copper Pairs are being used as the qubits.
No. Superposition (in a qubit sense) is not necessary for superconductivity. Most quantum computer designs require superconductivity, but it is not a strict requirement. Photon based quantum computers don't need any superconductivity iirc.
> it is a superconducting chip, and that is a major engineering problem
Superconductivity is not a super high bar. The bigger issue from quantum computing is coherence time; and it just happens that low temperatures are critical for lengthening it.
I am not sure what you mean here. Superconducting magnets are often used to control plasma (part of the environment) so it doesn't imply isolation from the environment.
Superconducting qubits are only superconducting because josephson junctions happen to be good candidates to make qubits out of. Ion traps (which generally have much better coherence times than superconducting qubits) and photon based systems don't need superconductivity.
I'm not sure how to make it more clear, but decoherence happens because some energy / information is transmitted into the environment. That loss of information is why superpositions collapse.
Maybe the confusion is what I mean by superconducting - I mean zero resistance circuits. It's not about what it's made of.
edit: also, I meant superconducting was (practically) required for superposition, but not vice-versa. I guess I misused the phrase one-required-for-the-other.
I believe that everywhere superconducting was used it was correct and using superposition instead would be wrong. So I agree with you there.
Congrats Intel!
To wit, can they afford not to sink money into it, if their competitors are, even if success if less than certain?
The rewards that seem like they would be realized are generational enough so as to render much of the existing market obsolete. So to be caught without any expertise, on a risk-basis is more costly than seemingly dumping money down the drain.
If you measure the qubit prematurely the function collapses. In classical computing it might be analogous to breaking out of a for loop before it has run its intended course and then you're left with an incorrect (or incomplete) answer.
https://www.youtube.com/watch?v=p7bzE1E5PMY
If n is 20, you're not really beating a 6502.
Being quantum lets a chip scale its speed better. But that's about it. At this size, it's worse in every single way than a normal processor.
https://phys.org/news/2017-10-quantum-computingbreaking-qubi...