This Quantum Computer is Better Than Your PC

so one of the biggest questions I get is always along the lines of Ian when will we ever see quantum computers as useful if you've watched my previous content you'll know that quantum computers have applicability in physics chemistry biology combinatorics and hopefully artificial intelligence as well but we've yet to see a useful calculation done by a quantum computer that's impossible for traditional digital computers to do in this video we're going to blow the lid that quantum computers are now outperforming other computers with useful problems not simply in just utility but wall time as well the leading paper in nature this month this paper says exactly that and I'm here to tell you all about our new Computing future [Music] foreign so let's address the elephant in the room right away we've already seen Google announce a Quantum Advantage back in 2019 which was an extremely limited demonstration of the power of quantum computing in that demo Google was able to run a Quantum build algorithm to simulate random circuits so it's not entirely a useful workload and simulating a Quantum circuit and Quantum a computer seems obvious but they showcased it would take classical Computing 10 000 years to match 200 seconds of quantum computer for that problem now while Google's announcement was groundbreaking it failed to address the utility of quantum computers which aim to solve real world problems the test Google did was synthetic at best or too abstract it from reality at its worst instead today's paper in nature addresses the boundary of quantum utility can a quantum computer be useful on a real world workload that a traditional computer cannot the paper is a collaboration between IBM using its 127 Cubit superconducting dilution refrigerator quantum computers that we've already seen in one of our recent lab tours and academics in California and Japan using traditional supercomputers they were showcasing a real world algorithm in fact the story is actually a bit more adversarial than the paper makes out to be more on that later but let me talk about the problem they were solving first so in any given material particularly Metals there can be magnetism and each atom has a magnetic alignment sometimes the magnetism can be permanent such as in a permanent magnet or it can be induced by an electric current or an electric field when dealing in Material Science one angle of calculation along with stress drains thermals and everything else is how well these materials interact with each other magnetically the same way a big aircraft manufacturer will simulate how other forces interact with each other on its finished product and magnetism needs to be taken into account as well what this paper tries to solve is something called the transverse 2D icing model in English this means a calculation over time on the effects of a magnetic field applied at right angles to a magnetic structure a one-dimensional calculation for this is relatively trivial and is done on classical supercomputers today but in two Dimensions it becomes very complex to model the behavior at the subatomic level this means traditional computers need to apply lots of tricks and techniques and assumptions that have to be made in order to get a result out thankfully a way to solve this in a Quantum domain has been researched and effectively if there are enough qubits a const computer doesn't need to make all those mathematical shortcuts or assumptions so this is where the competition between Quantum and classical comes into play this wasn't a case of finding an existing problem and trying to best it with a classical computer IBM actually reached out to the academics actively looking at this problem and others like it giving them the parameters and asking them to best it the researchers were excited for the challenge to try and push their systems to the limits using optimizations and other classical Computing techniques ultimately in this case the quantum computer had won despite months and months of work the classical computers could not match the results from the quantum computer both in terms of utility and speed to that end what took the quantum computer five and a half minutes the classical computer achieves the worst result in 30 hours so what sort of Hardware did IBM actually use I mentioned earlier this 127 Cubit processor which they call the eagle all 127 qubits are laid out in an architecture called a heavy hex which looks like a brick wall as shown here each connection point is a qubit and each one can come to entangle with its nearest neighbors with the right microwave pulse this generates what is called a two qubit gate and it is these Gates they're Quantum circuits for the calculation are built on typically when talking about a Quantum calculation we refer to the number of qubits or number of gates and also the gate depth the gate depth is the measure of how many times a two qubit gate can be used and this is limited by noise and noise is a perennial problem with all quantum supercomputers due to manufacturing differences no two qubits have the same coherence time the amount of time it can stay up and this is similar to how quickly dram cells might lose charge and it means each qubit has to be profiled at manufacturing for how long it can stay in this coherent state in this case the eagle revision 3 chip used had coherence times between 127 microseconds and 288 microseconds it doesn't actually sound like much but it is really good for this type of system and to put into context it only takes half a microsecond to set up and run a 2 qubit gate so even with those numbers IBM kept it conservative around the qubit for 60 circuits so 127 qubits ran for 60 circuits which we call the gate length and we call this a 1 to 127 by 60 circuit IBM has announced a goal of providing 100 by 100 circuits by the end of next year and it sounds like they're getting close to that already so I could go into the fact that these qubits and gate lengths enable 2880 c not Gates and the quantum calculation comes through stitching these Gates together and so on but I want to touch on in each trick IBM is doing here to get a result I said quantum computers are hideously affected by thermal noise it disturbs the result and as such that you this noise can Cascade through calculations and the errors compound until the result is meaningless there's a lot of work going on for error correction in quantum computers but it still stands that you need more physical qubits to implement error correction overall resulting in fewer logical qubits for calculations in this instance error correction like that wasn't possible due to that qubit overhead so IBM turned to error mitigation specifically zero noise extrapolation the noise in these systems is actually quantifiable and IBM can inject extra noise and run the problem with different amounts of Noise with enough results the end result can be extrapolated to the equivalent zero noise result this method has lots of applicability and although it multiplies the time needed for calculation we're still orders of magnitude better than conventional computing interesting to see how this technique works as we expand into bigger chips with more complex problems as part of today's announcement and the front page in nature I got a chance to sit down with Coley daughter abinav kandala and Katie pizzellato IBM's director of quantum theory about the paper and the utility of this announcement beyond the models in the paper I'm the manager of the quantum capabilities and demonstrations team and I'm Katie pizzelletto and I'm the director of theory in Quantum computational science right right that's a great question and and really the the first part of trying to approach anything that's beyond classical is to really firmly establish confidence in your results in a verifiable regime so with the kind of circuits that we can run we have a very nice knob where we can change a parameter and we're immediate so even though the circuits are very large in in volume those specific choice of parameters are very interesting verifiable fights so you can compare against those points and and and then build confidence in your device and the other mitigation methods and then you move on to re you know to the parameter spaces that are that are Beyond exact verification right so so a large part of the work is really in in verifying and building confidence that that these machines can do something accurate yeah I mean I think that's the interesting part about this paper and something that we talked about last time we talked about a lot internally is that getting these accurate getting these accurate values out of quantum States is what we're doing in most of our application work algorithm application work you're trying to get these accurate values and these load up these short depth circuits so while this is an example using the icing model and you know that was picked because of you know it being widely studied it's a it's a it's a nice way to map to the system there's a lot of other applicable things that we believe are out there that you can do by measuring accurate values out of these short depth circuits and that's what we're doing in almost all cases in IBM and outside of IBM in in where we're looking at applications today we're experts in in building quantum computers and and and trying to figure out methods and tools to to extract the most out of these systems uh but we're certainly not experts are are on the classical simulation side of things so we really wanted to find you know a world-leading group in in classical tensor Network algorithms uh for the simulation of quantum circuits and and we actually had an intern from that group uh with us over a summer who helped us connect and and they were very interested uh I think initially the idea was more to to kind of Benchmark and test our results but um it got into more interesting territory there was a lot of you know very interesting back and forth we had the opportunity to work with a very talented graduate student Sir John's in the group um and yeah I think over a period of time their their methods were able to inform what experiments we had to run next our experiments were helping them nudge and tweak their their methods so it was a very fun I think that we're going to continue to see this Quantum classical right we do not believe that there's a moment in time that says okay we're here Quantum one everybody pack up right Quantum and classical are going to continue to be partners and I think that a part of what's exciting is what is going to happen next to to come back and look at this problem as well so I think there's a lot of exciting Milestones ahead and I don't think it's a it's a one marker Moment In Time yeah I mean if you actually speak to the graduate student um who did a lot of the work on the classical simulation his reaction is well this experiment you know for the first time it it made me think a lot more about what changes I need to make to my classical simulation algorithms right so so I think it's it's it's quite uh a symbiotic relationship where uh where what they're beginning to speak on their end will motivate us to to change things on the experiment uh you know we're very excited that our our next Generations of Hardware will have even further improvements um that that should enable even larger circuit volumes make make us you know enable us to run even harder circuits we can be more careful about about the circuits we choose you know now begin to tailor them to to actually be hard uh from a complexity Theory perspective um and and so this back and forth will keep going I'm also you know optimistic that once the paper is published till the other groups uh in in the classical simulation side of things that'll that'll really try and you know uh see if they can verify our results and we really hope to actually see if that can happen we want to know if if parts of our unverifiable results were indeed correct or not uh and and I think people will hopefully latch onto that I don't think there's go I do not well I don't think there's gonna be a time Quantum is not going to replace classical I think we're all on board with that right um from a full-in perspective will there be parts that Quantum that you just go to Quantum for and not classical potentially um I think that you even in this even in this experiment the quantum peace still had a lot of classical pieces you know that you know even not just comparing to classical like purely classical but Quantum still does a lot with classical to get the quantum result there's a there's a lot of other you know kind of co-processing that goes along there um will we get to a point where that's gone and Quantum only does certain things maybe I mean I can see it I can see that you only rely on Quantum for certain parts of the problem you you know continue this kind of workflow where Quantum is just integrated um I think IBM you know at IBM we believe if we do it right there will be parts of that that Quantum should be another accelerator another co-processor that you use in your workflow for what you need it for um when that day comes I I don't think we can put a number on that but I I do think this is the beginning I mean I I think if I was gonna say what I what I really hope this is a big part of the beginning of is for people to start looking at Quantum as a tool uh you know the the history of of classical you know Evolution was we had gains in the devices and when we had gains in those devices we found you know a computational you know we found something to to drive the computational side of that and that's what I think um this is really the start of I hope is that this is a tool and we need to think about how to use it and we need to think about where to integrate it and where it is going to be valuable so many thanks to Katie and to abhinav for coming on the channel to explaining to me what this paper really means this paper is really information dense uh it took me at least to the second line before I had to break out and try and understand what some of these terms meant but it's an interesting read and you can check the link out in the video description if you have access to Nature now I'm it's hard to explain just how important having a high profile paper in nature is to uh to somebody who hasn't published a research paper for it is the cream of the crop it's these are the high profile journals that effectively can change where research is headed this will get a lot of attention not only for just what it's doing inside the paper but for the noise it's going to produce so congratulations to the team and again Link in the video description now overall on this the world of quantum Computing ever since that original Google announcement has had a tricky path in the media we are by far not even close to ubiquitous Quantum Computing but it's clear that we have a path to bring these qpus into the data center as traditional accelerators and what's going to take time is building the software backends to use what these computers can provide what's also going to take time is building more qubits of course as I've said the paper is in nature and do check out that link in the description but I should point out here that IBM has also announced a 100 million dollar partnership to build a hundred thousand Cubit computer in the next 10 years effectively a Quantum supercomputer now super users do need names I have a few ideas but do let me know yours in the comments down below [Music] [Laughter] [Music] thank you

Channel: TechTechPotato

Views: 13,676

Keywords: Quantum computers, Quantum utility, Real-world problems, Quantum advantage, Quantum computing, Traditional digital, Nature paper, Superconducting dilution fridge, Material science, Magnetic interactions, Transverse 2D Ising model, Quantum vs classical, Quantum entanglement, Qubit coherence time, Forward error correction, Qubit chips, Computing applications, Biochemical research, Software development, Qubit scalability, Data center accelerators, Supercomputer, IBM

Id: aqfVmF-M2vc

Length: 15min 58sec (958 seconds)

Published: Wed Jun 14 2023