Full transcript
0:10Good morning. Can you hear me?
0:14Good morning.
0:16I know it's early.
0:19Once again, good morning.
0:22Thank you.
0:24Welcome to day five at IEEE Quantum
0:27Week. Uh it is my distinct pleasure to
0:29welcome you this morning.
0:31And what I would like to do is right now
0:34invite Dr. Greg Byrd who'd be talking
0:37about some of the papers that are going
0:39to be getting awards today. And we'll be
0:41following that up with Dr. Raj Hazra
0:43who'll be doing the keynote this
0:44morning. So, thank you. I'd like to
0:46welcome Greg.
0:51All right. Good morning. I am not a
0:54track chair for the uh quantum
0:56technologies and systems engineering
0:58track, but it is my pleasure to uh
1:01announce the best papers from that
1:02track.
1:04Uh beginning with third place,
1:07the paper is development of titanium
1:09nitride nitrite aluminum nitride base
1:12superconducting cubic components
1:14from uh Benedict Schuff, Moritz Singer,
1:17Harsh Gupta, Mark Tono from TU Munich,
1:21Si- Simon Lang, Daniel Lissane, and
1:24Johannes Weber from Fraunhofer
1:26Institute. Authors, please come up if
1:28you're here.
1:41Okay, they may not be here, but if you
1:43see them, congratulate them.
1:47The second place paper is precision
1:49frequency tuning of tunable transmon
1:51qubits using alternating bias assisted
1:54annealing. This is a group from Righetti
1:56Computing.
1:58Zichao Li,
1:59Joel Howard, Alán Aspuru-Guzik,
2:01Greg Steel, Cameron Kupus, Stefano
2:04Piletto, John Wu, Mark Field, Nicholas
2:08Shirac, Christopher Ekberg, Hilal
2:11Cankaya,
2:12Roger Katta, Josh Mutus, Andrew
2:15Bestwick, Kameshwar Yadavilli, and David
2:18Pappas from Righetti.
2:20If any of those authors are here, please
2:23join us.
2:54And the first place paper is engineering
2:56quantum states with neutral atoms.
2:58Jan Balewski from NERSC and and Lawrence
3:01Berkeley, Biljana Komijani,
3:04from QuEra, Katie Klimko, from Berkeley,
3:08Siva Darbha,
3:09Mark Hersam from Illinois, Pedro Lopes,
3:13Feng Li Liu from QuEra, and Dan Camps
3:15from Berkeley Labs.
3:54All right.
4:03Thank you very much.
4:18Thank you again.
4:20Congratulations to all the award winners
4:22and thank you Greg for the presentation.
4:25Appreciate it.
4:27For all of you, I think who don't know
4:29me, I am Anand Sankey. I am the vice
4:32president of marketing and
4:32communications from Quantinuum. And
4:35today I'd like to introduce you to Raj.
4:38Raj is the president and CEO of
4:40Quantinuum and has more than three
4:42decades of experience in supercomputing,
4:44quantum, and technical roles all across
4:46the globe. Prior to joining Quantinuum,
4:49he served as the general manager at
4:51Micron Technologies and spent 25 years
4:54at Intel Corporation leading enterprise
4:55and government group.
4:58Before joining Intel in 1995, Raj was
5:01with the Lockheed Corporation based in
5:03the NASA based in NASA's Langley
5:05Research Center and prides himself on
5:07building high-performing teams with
5:09growth mindset and a culture of truth
5:11and transparency.
5:13Raj has a PhD and a master's degree in
5:15computer science from the College of
5:16William and Mary in Virginia, United
5:18States. And a bachelor's degree in
5:20computer science from Jadavpur
5:22University in Calcutta, India. And he
5:24holds more than 16 patents. Please wel-
5:26welcome Raj on the stage presenting the
5:29keynote. Thank you.
5:38Someone has a clicker for me?
5:45That's it.
5:48Well, good morning. Now that we've
5:50sorted some
5:52infrastructure problems out.
5:54Uh good morning. Congratulations to the
5:58the winners.
5:59Uh it's a great I'm sure it's a great
6:02feeling, but more than that it's kind of
6:03great tailwinds to continue and
6:05accelerate research. Thanks, Anand, uh
6:07for that
6:09uh introduction. Um
6:11It's been fun for the last 30 years, but
6:14I'm yet to win an IEEE best paper, and
6:16the hopes for that are quickly fading
6:18given what I have to do uh for my job.
6:23So, good morning again.
6:24Uh it's my first time at an IEEE quantum
6:28conference, uh quantum week. Um and I'm
6:31excitedly here this morning. Um I heard
6:34this is about 1,600
6:37uh registrations of folks attending, and
6:40particularly heartening was to hear that
6:42it's a 60% 40% uh academic um or
6:47industry versus academic, and 250 plus
6:51uh students here, including some high
6:53school students. That's awesome. Um
6:55because this is the new frontier,
6:58uh and everybody who is here today
7:01working in this field uh are a little
7:03bit like Lewis and Clark from uh the
7:06state that I used to live in till about
7:07last year ago, which was Oregon.
7:10So, today, what do I want to do?
7:13I have uh 29 minutes and 4 seconds, the
7:16clock says, uh to talk to you about
7:20where we
7:22as an industry go.
7:24Uh we are obviously there's a lot of
7:26activity, there's a lot of
7:28experimentation, there's a lot of
7:29research backed up by decades of
7:32theoretical work uh in the community.
7:35Um but where do we go from here?
7:38Uh what paths do we see?
7:40Uh and what particularly we as
7:42Quantinuum
7:43are taking as our path forward. And so
7:46with that, I want to talk a little bit
7:47about
7:49this notion of getting to a state in the
7:52technology where we achieve universal
7:55fully fault tolerant uh, quantum
7:57computing.
7:59Why is it important?
8:01Why is it important we get there, right?
8:04Uh, we it's for multiple reasons, but
8:07the biggest one is
8:10we believe, and I think this belief is
8:13coalescing in the community, um, that
8:17the value from quantum computing, as you
8:20look at quantum being yet another and
8:22potentially a breakthrough way and a
8:25novel way to compute, uh, adding on to
8:28years of working classical compute, um,
8:31and the world of HPC and quantum coming
8:34together,
8:36um,
8:36that is only going to be enabled when we
8:39have quantum computers that are
8:42universal, that they they are not
8:43special-purpose devices or
8:45special-purpose for only one or two
8:47algorithms or use cases. And so that's
8:51the universal part, and then they're
8:52fully fault tolerant. That is more of
8:55the energy from the community go will go
8:59into the insights and the outcomes from
9:02it versus trying to see if you want to
9:05believe in those outcomes and insights.
9:09So that is why it has been for a while
9:11uh, a hope
9:13uh, and a prayer. It was a prayer 5
9:15years ago or maybe even 10 years ago,
9:17uh, when it really seemed like NISQ, the
9:21era of NISQ was here to stay forever,
9:24uh, and all of the approaches to
9:25achieving full fault tolerance were
9:28uh, more like holy grail or you know,
9:30Hail Mary's or whatever your favorite
9:32terminology is.
9:34Well, today's discussion on my part is
9:37to say it's no longer a hope and a
9:40prayer.
9:42It is now our responsibility and we see
9:46a path, a clear path to get there.
9:50The second important point is it's not
9:53just a clear path to get there
9:55in the next 50 years, 100 years,
10:00but in the next decade or less.
10:04And in particular, by the end of this
10:06decade, we believe we can get to a
10:09position where we can unleash the power
10:12of quantum computing on a large number
10:15of use cases
10:17that bring both
10:19quantum to the forefront of things you
10:22couldn't have done before
10:24or
10:25and or things you can do more
10:27efficiently or better now.
10:29And it's important that a new form of
10:32computing serve both purposes.
10:34So, without much further ado on the
10:37introduction, it's about the first time
10:38I've spent most of my time on the title
10:40slide, but I think that it's needed up
10:42some setting.
10:45Let's move on.
10:47So, we at Quantinuum, how many people
10:50here are
10:51familiar with Quantinuum?
10:55All right. For so those
10:57that aren't and and just like the
10:59airlines say, even if you're a frequent
11:01flyer, you should listen to the
11:03safety demonstration or whatever they
11:05call it,
11:06I'll do that. We've we are based out of
11:09Broomfield in the US and Cambridge and
11:12Oxford in in the UK.
11:15500 employees across eight offices. I
11:19already said US, UK, Germany,
11:23and Japan being being the major
11:26countries.
11:28We've got 375 plus of those 500-ish
11:32folks are PhDs and masters and we
11:35believe this is the largest
11:36concentration of quantum experts outside
11:38of academia under one roof.
11:42200 scientific papers. We are very, very
11:45committed to publishing what we do.
11:49This is part of our culture where we
11:51say, "If we are doing things
11:54uh
11:55that we want the peer community that to
11:58look at it, to review it,
12:00um and that builds our conviction
12:03then that we are on the right track.
12:06Uh it also makes us better when we get
12:08feedback.
12:10And we've got 100 plus patents as as a
12:12business that is uh obviously in the
12:14process of building value, creating
12:16value.
12:17Uh we take patenting as very seriously
12:20as well.
12:21Uh the cross-domain subject matter
12:24expertise that you see on the right is
12:26particularly important for us
12:28because this is the the lingua franca
12:31gap or the last 10,000-mile gap
12:34uh between computing and its use.
12:38So, we have
12:40a number of folks that are focused that
12:43are domain experts within the company on
12:46telling us how quantum computing is
12:48going to be used in quantum algorithms,
12:50for examples, applied to chemistry or
12:53natural language processing and
12:54artificial intelligence, condensed
12:56matter physics, obviously cybersecurity
12:58one of the prime early drivers for
13:01thinking about quantum computing uh and
13:04financial use cases. So, this helps us
13:08understand the context of the end user
13:12or the end customer a- and also help
13:14participate with them in accelerating
13:17their journey
13:18um on using quantum computing.
13:20Makes us a little bit unique in some
13:22ways where we aren't just building the
13:24infrastructure and saying, "I hope you
13:26can use it." Or even before that, I hope
13:28we got it right.
13:31So, with that,
13:33we deliver a full stack value.
13:36What we've taken on
13:38and it's kind of how we were formed by a
13:41combination of
13:43Honeywell Quantum Solutions, which
13:44focused on building state-of-the-art
13:46quantum computing hardware, and
13:48Cambridge Quantum Solutions, which was
13:50focused on writing the best, discovering
13:53the best quantum algorithms, creating
13:55the best software,
13:57bringing them together in a full stack.
14:00Um and that's how Quantinuum was formed
14:02in 2021.
14:04So, the result of that is we we bring
14:06the best of those two
14:08and the co-design of those two and the
14:11co-optimization of those two into one
14:13full stack.
14:15At the bottom, you will see Let me start
14:18from the top. Maybe it's easier to
14:19approach it from the customer
14:20standpoint.
14:22Uh from the top, you'll see essentially
14:26libraries. I call them libraries. Some
14:28people could call them applications. I
14:30tend not to call them applications
14:32uh for a particular reason, but these
14:34are domain-specific libraries. In
14:36Quantinuum, it's our molecular materials
14:38discovery,
14:39aka chemistry,
14:41uh libraries. We have other algorithm
14:43libraries in machine learning, Monte
14:45Carlo integration,
14:47uh quantum natural language processing,
14:50and we also host third-party software on
14:52the layers above, and I'll talk about
14:54that in a second.
14:56Uh the next layer then is the what we
14:58call the orchestration layer, right? So,
15:01if you're going to have hardware and not
15:02just one hardware, but multiple kinds of
15:05quantum hardware, what is your
15:07orchestration layer? This allows you to
15:09use these libraries, create workflows,
15:12manage workflows,
15:14uh collaborate across multiple folks and
15:18and sites,
15:19all based on the value of the of the
15:22powers of the cloud infrastructure and
15:24it's a workflow orchestration platform.
15:26A part of that is the ticket Quantum
15:30SDK, which is open source and has more
15:33than 1.6 million
15:35downloads.
15:36That gives you a total developer and
15:39deployment platform to take your Quantum
15:41algorithms of the workflow use cases
15:43you've delivered and now get ready for
15:47deploying them on on either simulators
15:50that are built in or real hardware as
15:53it's becoming more prevalent.
15:56The next layer is Quantum error
15:58correction.
15:59This is kind of that funky thing of
16:00whether you say that's the applications
16:02job or the software's job or the
16:04hardware's job and the answer in true
16:06Quantum spirit is yes and yes, right?
16:10Um
16:11and and what the reason I call it out
16:13separately is that obviously um
16:16we've built our hardware, which is
16:18Quantinuum systems, to not only be able
16:21to take our error correction methods
16:24that we develop and I'll talk about that
16:26in a second, but others as well and
16:28you'll see some examples of that
16:30particularly if you've been following
16:31the literature in the last
16:33couple of months.
16:35And then at the bottom is our hardware
16:38and other Quantum computers particularly
16:40gate-based uh Quantum devices.
16:43So when you look at the stack, you see
16:45two things.
16:47You see a connection right from I have
16:50this problem to solve to getting it
16:52solved on real hardware.
16:54And the second thing you see is choice.
16:58That is we aren't a vertical full stack
17:01walled garden.
17:03In the early days of Quantum and we
17:06believe we're still in the early days of
17:07Quantum, it is important to be able to
17:10experiment openly.
17:12And so we have folks that we work with
17:14who would want to take the algorithms
17:17that they're developing and run it on a
17:19trapped ion system or a superconducting
17:23uh system or neutral atom system and we
17:26allow that through the layers of
17:27software we build uh
17:29for them to do that. Of course,
17:33we believe
17:34that we do that in the hardware the
17:36best, but if that's not the case, we
17:38learn
17:39and we learn why.
17:41But we still allow that choice to go
17:44through. So, we are a full stack and
17:46from a solution {slash} software
17:48perspective, a platform-inclusive
17:51company. That is our fundamental
17:53approach
17:55to the world of quantum and the quantum
17:57ecosystem today.
18:00So, we've been
18:01following this philosophy
18:04uh over three generations of quantum
18:06computers, right? And the And the little
18:08drawing you see on the side is actually
18:10our latest uh generation system. It's
18:13called H2, system model H2. I'm holding
18:16it up. That is the actual trap.
18:18That's the front of it.
18:21That's
18:22the back of it.
18:24You can see the pin outs.
18:26Uh it's really that.
18:28Um and that hosts 56 qubits today.
18:31Uh very high fidelity qubits.
18:34Uh and over three generations now from
18:37H0,
18:39H1, and H2,
18:41we we've followed a philosophy of
18:43increasing scale
18:45while improving fidelity at the physical
18:49qubit level.
18:51Uh we were the first to demonstrate
18:53real-time error correction because we
18:55had uh
18:57um a capability in our architecture of
19:00something called mid-circuit
19:02measurement. So, instead of waiting till
19:04all the computation ended and then
19:06testing for errors, we could actually
19:09uh look at it in the or examine
19:11uh
19:12extract syndromes and and correct in the
19:15middle of the computation itself, and
19:17that's that's vital if you're not if
19:19you're going to pay a large time penalty
19:21overheads.
19:23We were first to get 99.9%
19:26two-qubit gate fidelities,
19:28um in a commercially available quantum
19:30computer. I'm glad that point comes up
19:33because our machines are not just test
19:35machines. They are our R&D factory as
19:37well, but there are others that are
19:39using it. Our customers are using it.
19:42Um
19:43and and so we have to make them
19:46commercial in the sense that if you take
19:48money from someone and you're promising
19:50an SLA in return, these can't just be
19:53purely you know, research machines that
19:55aren't predictably up, that don't um
19:59resiliently have a set of features that
20:02stay for the customer to to be able to
20:04expect to use over time.
20:08Um and so in three These are three
20:10generations of commercial quantum
20:12computers
20:14uh that we've been able to improve in
20:17its lifetime and generate these results.
20:19The third one being quantum volume. 2 to
20:21the power 21 was what we demonstrated
20:24last. Uh
20:26and this is if you think in if you want
20:28to put it in context,
20:30uh that's the second and I won't name
20:33the company,
20:34uh quantum volume is at
20:362 to the power of nine.
20:39Uh now obviously you could say, "Well,
20:41you've got 56 qubits, why aren't you
20:43generating a higher quantum volume?" As
20:45you know, to generate quantum volume or
20:47to calculate quantum volume, uh you you
20:50need classical compute to verify a
20:52portion of the solution. We've now got
20:54H2, this machine,
20:56uh that would use so many so much
20:58classical compute just to generate the
21:00end of the the value of this benchmark,
21:03but that it becomes quite prohibitive to
21:05do so.
21:07Um
21:08but even at a time when it was possible
21:11for us and others to do so without
21:13breaking the bank
21:15and paying a lot of supercomputing time,
21:17uh we were well comfortably ahead. And
21:20that all got back to not just number of
21:22qubits, but as you know, in the quantum
21:25volume uh benchmark, or if you want to
21:27call it that, uh the measurement, it it
21:30captures the contributions of fidelity
21:32front and center.
21:37More recently, while we've been
21:40improving the the the scale of qubits,
21:43we've also demonstrated some fundamental
21:47leap forwards in the ability to to
21:50improve fault tolerance. Um I won't go
21:53into this in much detail because it's
21:55really out in in in literature right now
21:58uh and being talked about quite a bit,
22:00but the 12 logical qubits that we
22:02demonstrated with Microsoft or Microsoft
22:04demonstrated with the error correction
22:06code on our high-fidelity hardware is
22:08the largest number of entangled logical
22:11qubits with the highest fidelity on
22:13record uh on a commercial system. We
22:16were able to get 800x improvement um
22:19over
22:20uh in terms of of circuit resiliency
22:23over what would have could have been
22:24done just with the physical qubits that
22:26were
22:27uh that were used to create these 12
22:29logical qubits. I have to tell you that
22:31the rate and pace of innovation is
22:33something we will remember till the end
22:35of the stock
22:36has been dramatic. In 6 months, we went
22:39from four logical qubits
22:42to then eight, and then 12 on the same
22:45number of physical qubits. So, that is
22:48the rate and pace
22:50at which we can improve resiliency at
22:52scale even today before we further
22:55advance the state of the art in the area
22:58of both high-fidelity physical qubits
23:01and error correction.
23:02The other one is we're looking
23:04long-term, too.
23:05It's not just about what you could do in
23:07the next 2 years, 3 years, but we're
23:09looking at uh this whole area of
23:11topological qubits uh for which for not
23:15two to three decades have been kind of
23:17the holy grail promise um around uh
23:20around creating real fault tolerance um
23:24and getting beyond the NISQ era in uh in
23:27May of last year uh we demonstrated on
23:30our H2 machine the ability to create
23:33this new state of matter. Uh we call it
23:35a new state of matter. It's a
23:37non-Abelian topological order um
23:40vibrating.
23:41And and so if you're interested in that
23:43that paper, we can certainly make that
23:44available. That gives you a glimpse
23:47that we aren't just thinking NISQ is
23:48forever, that we are exploring ways to
23:51get these very high
23:53fidelity qubits and this particular
23:56architecture that we have uh in in our
23:59in our trapped ion system to be able to
24:01generate these next generation or next
24:04frontier of
24:06error resiliency on these machines.
24:12We So, we are
24:14building the world's most powerful and
24:16reliable quantum computers.
24:18And we will continue to build that.
24:22We believe as we move forward down this
24:24path what our customers, what our users
24:27are asking us is to lead
24:31is to build with leading scale
24:33and the lowest error rates
24:36and offer unparalleled customization
24:40because not all workloads look the same
24:42or there isn't just one workload.
24:45That is a pretty demanding ask, but that
24:48is
24:50the responsibilities of leadership is
24:52you get asked to do the impossible.
24:56And we've very gladly and happily taken
24:59on that challenge.
25:01So, in my remaining time
25:03I will talk about what are we doing from
25:06here onwards, as I've spent now the
25:08first part talking about why we dare to
25:11dream and why we dare to take on that
25:14challenge and say that we will have a
25:16very significant meaningful step forward
25:19in fully fault-tolerant universal
25:21quantum compute by the end of this
25:23decade.
25:27So, let's start with
25:29what do we mean by a universally fully
25:32fault-tolerant quantum computer?
25:34This may seem like an academic thing or
25:36may remind the students in the audience
25:38of of kind of your term papers or uh
25:41definitional foundational questions.
25:44But, in these early days of the
25:46industry, there is a lot of confusion
25:49and noise about what these terms even
25:51mean.
25:52And I think it's important for us from
25:54an intellectual integrity perspective,
25:56from a technical integrity perspective,
25:59and from a market clarity perspective,
26:02to have clarity and to maintain that
26:04clarity and measure ourselves to clear
26:08definitions of what we say we are
26:10building.
26:12So, what we mean by universal
26:15is it can perform any computation
26:18allowed by quantum mechanics, practical
26:20computation. Uh and if it's not because
26:23if it's not universal, it can be
26:25simulated with some amount of a
26:27classical computing infrastructure. And
26:30that boils down to you must have both
26:32Clifford and non-Clifford gates
26:34supported
26:35in in the machine that you offer to the
26:38algorithm.
26:40So, we believe universal is important.
26:43It's not just important, it's critical
26:45because what from a from a industry
26:48perspective,
26:50if we don't and we get to
26:52domain-specific quantum computers that
26:54are only good for certain algorithms, it
26:56makes the business case. Any of you here
26:59also thinking of getting your MBA, you
27:00could do a paper on
27:02if you don't achieve general purpose and
27:05everything becomes very very
27:07algorithm-specific, what are some of the
27:09challenges in the economics of
27:11continuing to invest in that?
27:15Then the next question is, okay, so we
27:17understand by universal, what do we mean
27:19by fully fault-tolerant?
27:22Well, first, this one I don't think
27:23there's any controversy, but fault
27:25tolerance enables errors
27:27to be corrected to achieve a targeted
27:29accuracy. It's not about getting a
27:32fault-free
27:34computer to begin with.
27:36Right? So, let's let's agree that that
27:38is the case. It's a specified set. If
27:41you target a particular accuracy
27:43level,
27:44which you could say in terms of circuit
27:46errors, circuit depth, whatever your
27:48metric is, you achieve fault tolerance
27:50to that level. And then you raise the
27:54bar.
27:56In practice, fault tolerance means you
27:58must have real-time syndrome extraction
28:01and decoding, so you can actually say
28:03this is the when the error happened,
28:04this is what the error is, and this is
28:06how I correct it
28:08in in situ, that is in the flow of the
28:10computation. And you must have a
28:12complete fault-tolerant gate set enabled
28:15by magic states.
28:17So, all your operations, whether it's
28:19encoding gates, measurement, syndrome
28:21extraction, if they are fault-tolerant,
28:24now you have a fully fault-tolerant
28:28computer, quantum computer,
28:31and if you have maintained
28:33the Clifford and non-Clifford
28:35capabilities or states or gates, then
28:39you have a universal fully
28:41fault-tolerant quantum computer.
28:43That is what we are going to build.
28:53So, as I lay that out,
28:56um
28:58let's talk about what is it going to
29:00take to do that, right?
29:04And I'll
29:06we are putting the stake in the ground
29:08saying we will build such a system
29:11on our road and as part of our road map
29:15called Apollo. It will be a fully
29:17fault-tolerant and universal quantum
29:19computer.
29:20It will maintain the critical features
29:23of our QCCD trapped ion architecture,
29:27all-to-all connectivity, which is
29:28critical critical for
29:30some many of the well-known and up and
29:33even upcoming error correcting codes,
29:36mid-circuit measurement, and the ability
29:37to reuse the qubit after that, and
29:40conditional logic. We will build it and
29:42make it available in 2029.
29:46And in particular, what we will target
29:48for this machine
29:50is thousands of physical qubits,
29:54hundreds of logical qubits, and the
29:57interesting part is the the range
30:00of logical error rates or error rates.
30:04And this is where that customization
30:06becomes very very important.
30:09The customization becomes very important
30:11because a number of you get a trade
30:13space a trade-off space between the
30:17number of logical qubits you want and
30:19the error rates
30:22you want to target.
30:25And that we think is important because
30:27this quantum computer will run
30:29a wide number of algorithms,
30:33um and not just be good for one or two
30:35hero algorithms that demonstrate some
30:37form of quantum advantage or even
30:40supremacy.
30:42This is going to be a practical, fully
30:45fault-tolerant, and universal quantum
30:47computer.
30:50So, how are we going to get there?
30:55Um
30:56at
30:57When I look at it, I I literally do
30:59post-its in my office.
31:01Um and and our
31:04our technical leaders tell me
31:07um that there are two sets of things
31:11that need to progress
31:12on a certain clip and cadence. Um I
31:15write them down and I check on how we're
31:18doing against them.
31:20The first one is how do we get to larger
31:22number of physical qubits?
31:25And there's three things that need to be
31:27done
31:28in order for us to get there
31:30to this thousands of physical qubits in
31:32Apollo and even Apollo just being the
31:35start of of a generation, how do you get
31:39to millions of qubits after that?
31:41The second is scaling traps to hold more
31:43qubits. Just giving you more qubit
31:45density because you don't want that trap
31:47that I just showed you to become
31:50at least more than a wafer in size.
31:53Um
31:55The third is
31:56yes, we've got qubits, but the way we
31:58control them, interact with them is with
32:01lasers. And how do you get laser beams
32:04delivered into this highly dense
32:06population of qubits in a very small
32:09space uh
32:10for gating and cooling operations.
32:14So, that's
32:15just getting to larger number of
32:17physical qubits. How do you build fault
32:19tolerance? You got to improve
32:22overheads and improve the effectiveness
32:24of real-time syndrome extraction and
32:26decoding. And that's a done.
32:29The high rate of error getting high rate
32:31error correcting codes so that we can
32:33actually use the syndrome extraction
32:36while we push on the number of qubits
32:38that many would agree
32:41um do strain the notion of fidelities,
32:43right? Crosstalk increases, some of the
32:45other error drivers increase.
32:48How do we still have the error
32:50correcting codes be effective in
32:52providing those logical error rates?
32:56And then, how do we build a fault-tol-
32:58How do we build all of this error
33:00correction detection and correction and
33:03maintain a universal gate set, Clifford
33:05gates and non-Clifford gates? What do
33:07you see are my check marks and whip?
33:10The things that we believe we've
33:13demonstrated, not just think we have a
33:14solution
33:16get check marks.
33:17The things that are whip
33:19I check on.
33:21And and we collectively work as a team
33:24to de-risk those through experiments and
33:26then finally demonstrate them in test
33:28beds.
33:30So, this is kind of the to-do list of
33:33getting to Apollo, if you will, on
33:35post-its.
33:37Highly oversimplifies the tremendously
33:39complex tasks that underlie these, but
33:43it's a good way to create a milepost on
33:47this journey to Apollo.
33:50So, let's look at a larger number of
33:52physical qubits.
33:54Um the first one is the wiring problem.
33:57If you see, I'll I'll decode the the
33:59marks for you. Uh check means done. Done
34:03as in high confidence demonstrations,
34:06including even test chips of where we
34:08think we can get it we have the solution
34:10that will not only do Apollo, but scale
34:12beyond Apollo.
34:14Number one Number second is scaling
34:16traps to hold more qubits, getting
34:18maturity
34:19uh into the trap architecture, as well
34:22as into some of the manufacturing
34:24technologies needed for those traps.
34:27Uh we partner on those uh
34:30with folks that build uh traps.
34:33Um, and that is
34:36the the horizontal arrow means that's
34:38progressing on schedule and on target.
34:41And so is the laser beam delivery for
34:44individually gating and and cooling um
34:48qubits.
34:50When we get to that second list list of
34:52things is real-time syndrome extraction
34:54and decoding.
34:55Act- actually I'll go back. I think I
34:57did disservice to this slide.
35:00Um, so if I can just build it back.
35:05Um
35:06the wiring problem for individually
35:08controlling qubits I kind of glossed
35:10over as and it I may give you the
35:12impression that's easy to do. But it is
35:15one of the harder tasks
35:17uh in creating scale without giving up
35:21uh fidelity and just pumping a lot of
35:23useless qubits
35:25uh out of the trap or into the trap
35:27actually in this case.
35:28Uh because the the concept is simple. We
35:32We use
35:33electrical voltages to to to move ions.
35:37Thankfully they're charged so you can do
35:38that. Um
35:40and and at you know, things like 20 or
35:4256 qubits the design of the control
35:45mechanism is very very ru- you know, you
35:49can see it literally with your eyes. But
35:51when you have hundred and thousands of
35:53qubits in there you can't have that
35:55one-to-one. So you must find a way where
35:58you can use broadcast method- ologies
36:01on voltages to say I'm moving all not
36:04one but I'm using one signal to move
36:06multiple. But then have a way so that
36:09all those multiple can do different
36:11things and that's the sort part and
36:13that's what you see on the right here is
36:15in in a eight-site two-grid trap
36:19we've got a sorting operation happening
36:22after they've been moved interacted with
36:26one common what we called an analog
36:28voltage. Uh Um
36:31with digital signals. So, we This is the
36:34wiring and sorting problem that
36:36otherwise would mean how do you get to
36:3710,000 qubits? How do you get to a
36:39million qubits and have that many a
36:40million voltages going into the trap,
36:42right? And this is the solution. The
36:45broadcast solution gets that overhead
36:47down uh to multiple qubits, a large
36:51number of multiple qubits being handled
36:53by one
36:54not each individually through voltages,
36:57right? Uh we have a paper on this. So,
36:59for those that are interested, again, we
37:01the team here can can um give you the
37:04paper and you can have a look at it.
37:07Um but this was a very fundamental
37:09breakthrough that we had to come
37:11through. And frankly, I will tell you as
37:13I go around and talk to people, one of
37:15the the one of the questions posed to me
37:17is can QCCD iron trap scale?
37:20Right? And and before we did this, we
37:23asked the same question of ourselves.
37:26Having done this, we are very confident
37:28the answer is yes, and that's why we put
37:30it as plan of record for Apollo. And we
37:33And And Apollo again isn't an endpoint.
37:36Apollo is the start
37:38of the of the practical fully
37:41fault-tolerant universal quantum
37:43computer generation.
37:47So, I'm glad I didn't gloss over that.
37:50Um
37:51On the paving the path to Apollo on the
37:54fault-tolerant scale um
37:57you know, we've made really good
37:58progress nailing down the uh the
38:01real-time syndrome extraction and
38:02decoding part. We've obviously have
38:05multiple ways uh to do error correction.
38:09We have our own one of the examples of a
38:11recent paper
38:13uh which is um which you see on the
38:15right here uh is our own work. We work
38:18with partners like Microsoft uh on their
38:21qubit virtualization, which is uh
38:23their name for essentially our detection
38:25and correction. Um
38:28And and we've been able to achieve
38:30what's uh something that is really
38:32really significant.
38:34We've been able to achieve
38:36um
38:37the ability to say logical to physical
38:41overhead, that is the number of physical
38:42qubits you need to build a a uh a
38:46logical qubit is not hundreds or
38:48thousands,
38:50but tens or maybe even single digits.
38:54I mean, if you just do the math and say
38:56at 56 qubits, if you can generate 12
38:59logical qubits, which is what Microsoft
39:01did, uh then that's very different from
39:04even what was published a year or two
39:06ago saying you need hundreds uh physical
39:08qubits to generate a logical qubit.
39:12So, we are making very good progress.
39:15The one that's on the end, it's also on
39:17my posted note, if you notice, is a
39:20fault-tolerant universal gate set. And
39:23we've got the Clifford gates done, the
39:25non-Clifford coming soon, watch that
39:28space, uh and and we will we will give
39:31you what we believe is the right answer
39:34there.
39:37So,
39:38that's Apollo. That's our path to
39:40Apollo. That's knocking down the
39:42challenges to building something like
39:44Apollo.
39:46It's not just wait till 2029
39:49and you will see this machine.
39:52It has never worked in classical
39:53computing to do that.
39:55Uh you have to mature the technologies,
39:57you have to mature uh the integration of
40:00multiple hard technologies, and you have
40:03to grow the software ecosystem uh to
40:05use, for instance, logical qubits,
40:08um to be able to write algorithms that
40:10are
40:10what we call fault-tolerant algorithms,
40:13um and be ready for something like a
40:15machine like Apollo.
40:17So, we have two more generations of
40:20systems we are going to build and and
40:22make commercially available so that the
40:24experimentation on what is working, what
40:26is working well, what needs to be
40:28improved is not just our judgment, but
40:30is also the feedback from the community
40:33that we serve.
40:34And that's Helios, which is going to be
40:36out first half 2025, so it's just around
40:40the corner, and Sol in 2020
40:43uh
40:457.
40:47Again, you'll see the logical error
40:49rates and you will see the number of
40:51logical qubits that we think we're going
40:53to achieve
40:54uh or we plan to achieve at those error
40:56rates, right? Um
40:58so again, this is then a roadmap that
41:01gets us curing those technologies,
41:04curing the the ability to integrate
41:07those technologies at increasing scale,
41:09and deliver these machines that are
41:11reliable uh and can be used on a
41:14commercial basis.
41:16I have been asked a few times
41:18if we have taken a random walk through
41:20the trap space. If you look at the top
41:23line, it's like
41:25someone asked me yesterday uh
41:27in each generation
41:29do you just realize you you made a
41:31mistake and you have to go to a new
41:32trap? Because that's what it could look
41:34like
41:35to a quick glance of every generation's
41:38trap looks kind of different except
41:40between Sol and Apollo. That's not the
41:42case.
41:44We've always had an Apollo-like
41:48QCCD scaling architecture in mind.
41:51Except you can't get there in one step.
41:56And so what we've done, if you look
41:58through H1, H2, a- and Helios in
42:01particular
42:03is
42:04matured some of the key challenges that
42:06you have to do to be be able to build a
42:08two-dimensional highly dense trap like
42:11Apollo's.
42:12Um you have to do things in transport,
42:14you have to do things in beam
42:16delivery, you have to do things in
42:17cooling, you have to do multiple things.
42:20So, H1 is a single track, H2 is two
42:22single tracks side by side. Helios
42:25actually provides a trap crossover, I
42:27call it. I'm sure that's the wrong
42:28technical term, but it's a junction. And
42:31junctions are important because then
42:32when you get to solve, you have what we
42:35call a a two-dimensional
42:38it's no longer a race tracks, it's a C
42:40block, right? You actually have to make
42:41cubits uh move across junctions and turn
42:44corners. And then, once you can do that
42:47reliably, then you apply things like the
42:48wiring and sorting problem solutions and
42:51scale that.
42:52Right? So, this the trap designs are a
42:56part of a long-term road map plan uh to
42:59de-risk critical technologies and
43:01challenges and not just, you know, hey,
43:04what do we do next time, right? Not like
43:06one of those uh
43:08GRE questions that says, if you give you
43:10a sequence, can you predict the next
43:12one? On this one, you can predict the
43:14next one. And so, let's talk about that.
43:19Apollo represents
43:21a point in time, but it also represents
43:24the start of a new era for us.
43:26And what it does is is
43:29the first of machines we build
43:33very soon thereafter that with an
43:35architecture that can scale
43:37to millions of cubits.
43:41We're
43:42and we do that in two ways. One is
43:44increasing the density, the ability to
43:47have 50,000 cubits on a single square
43:50inch die,
43:51um
43:52that then can go into millions, into
43:56high levels of scale by tiling them
43:58together and and not having to use very
44:02complex photonics other forms of
44:04transport between tiles, uh
44:07but actually uh
44:09transport them across
44:11using voltages.
44:12So that is one of our plan.
44:16It's kind of eerily what the classical
44:18computing world has also realized with
44:21the notion of chiplets. Is you don't
44:23have to build everything on one
44:25die. You can have heterogeneous die or
44:28heterogeneous functions, but you can
44:30then connect them in a chiplet
44:32architecture. This is the quantum
44:33version of the the chiplet architecture.
44:36For that we are looking at the key
44:38enabling technologies and have them in
44:41R&D.
44:42Integrated photonics to deliver all the
44:44laser beams that we will need. We will
44:46no longer be to get to these levels of
44:49scale. You will have to have integrated
44:52optics.
44:54We will leverage chip tiling as I just
44:55said. And we will have to increase the
44:58rate and pace of broadcast signals and
45:01sort protocols to actually continue to
45:04not have the density be a headwind
45:07against
45:08the ability to scale and maintain those
45:10fidelities.
45:12So we have a clear plan with technical
45:16conviction
45:17on
45:19how we get Apollo to be the start of a
45:23journey to millions of qubits.
45:27Physical qubits, 100,000s to millions of
45:31logical qubits and very very
45:35low error rates.
45:40We at Quantinuum, if you go to our
45:42website, you'll see it says our mission
45:45is accelerating quantum computing.
45:48I'll add a little bit of a
45:49clarification. Our mission now with the
45:52clear line of sight to Apollo and what
45:56is needed to get us from a
45:59to beyond Apollo is accelerating
46:02universal fully fault tolerant quantum
46:04computing.
46:06And that is the mission that 500 of us
46:08within the company, working with
46:10thousands and now rapidly growing
46:13numbers outside the company, uh
46:15are on
46:18the path of. And that's our invitation
46:21to the rest of the community to join us.
46:23We are highly collaborative, um
46:25to work with us, to work collectively on
46:27technologies, to work collectively on
46:30benchmarks and specifications,
46:33and even policy and regulations that
46:36keeps the ability and the the the
46:39promise of quantum computing
46:41uh
46:41from being delayed in getting delivered
46:43to humanity.
46:45I will end with this one quote, and it's
46:47kind of um
46:48interesting that given where I stand
46:50today, uh it's Justin Trudeau who in
46:532019 at Davos
46:56said that line, "The pace of change has
46:57never been this fast." And if you really
46:59look around and look at all the papers,
47:02um
47:03and all the progress being reported on
47:05quantum computing, it does feel like
47:07man, the last 6 months have been like
47:09very different than the previous six.
47:10And I can guarantee you the next six
47:12um will feel will make this six feel
47:17old.
47:18Uh
47:19but
47:21he's exactly right. I don't know about
47:23I'm not a politician, so I don't know
47:25whether that's true for society and
47:26human humanity or or governmental
47:28change, but I will tell you he's right
47:30about it in quantum computing. The pace
47:32of change has never been this fast, yet
47:35it will never be this slow again.
47:37And we, collectively, can not only make
47:41it happen, but make it our destiny.
47:43Thank you.
47:54Thank you, Raj, for that energizing
47:56presentation and and sharing your vision
47:58about the donor what will be an amazing
48:00paradigm in the world of technology. So,
48:01thank you.
48:03We are going to open up for some
48:04questions. You see two microphones over
48:06here. And please do line up and state
48:10your name, your affiliation, and
48:13do definitely recommend questions around
48:15the presentation talking about a
48:16performance fidelity scaling. There may
48:19be questions that we may not be able to
48:20get into confidentiality reasons and
48:22that's why I'm here to help moderate.
48:23Thank you.
48:25Please join in for questions.
48:36Hey, my name is Vincent Michaud Hu. I'm
48:38with Xanadu Quantum Tech.
48:41I'm really unfamiliar with quantum error
48:43correction and I was wondering like how
48:46should I think of this logical qubit? Is
48:48it like a partitioning of the physical
48:50qubit or like there's many physical
48:52qubit
48:53contributing to each logical qubit?
48:58Great question.
49:00Um
49:01we could go hours on this and there are
49:04people on the team here that can give
49:05you whatever level of detail you want,
49:07but the way you should look at it is a
49:09collection of physical qubits
49:12that represent
49:14the state that you would have expected
49:17in that physical qubit and make it
49:20available to you more resiliently than
49:22you would on the physical qubit. So,
49:23I'll take the point of view of a
49:25software guy that's trying to write code
49:28that uses that physical qubit or the
49:30qubits in that. I look at it as an
49:32abstraction.
49:34I look at it as I'm still going to
49:36subject that logical qubit.
49:39I'm going to code for it like it would
49:40have been a physical qubit.
49:43Except it's going to be it's going to do
49:45its internal management and look at the
49:46redundancies of of carrying that state
49:49resiliently across a number of physical
49:52qubits
49:53and give me the right answer.
49:55Right? So, if you think about it and and
49:58I think you're building one of these
50:00demonstration because the frequently
50:02asked question
50:03uh
50:04if you think of it like an orchestra,
50:06right? So, I'm the conductor, I'm the
50:09software guy. I want a piece of music to
50:12be played as it was written, not errors.
50:15And I ask the violin section to play it,
50:18right?
50:19And there's five violinists and they
50:21decide amongst themselves how they're
50:22going to play it.
50:24Are they going to all play the same
50:25thing so if one stops playing or their
50:26string breaks
50:28you know, all other four or they're
50:30going to do play different parts of it
50:31because they're better and in different
50:33parts their their skill sets are
50:35different. That's their their business,
50:37right?
50:38But I get as a conductor
50:40a true
50:41faultless rendition by multiple
50:44violinists cooperating
50:46to play that music.
50:48Right? That is
50:50the most musical analogy I can give you
50:53for what a logical qubit is from
50:55physical.
50:57Thank you. Should sound like music to
50:59your ears.
51:01Thank you. We have We have Anthony here.
51:04I uh Tony Pribis, Northrop Grumman and
51:07the Council on Superconductivity.
51:09Uh
51:11when I saw your plan for scaling to
51:13millions of qubits,
51:15the chiplet architecture is very
51:18interesting. I'm I'm curious about how
51:23the the the chips will interact. Is that
51:26is that a movement of ions between
51:28chips? Is that a you know, couple of
51:31ions come up to the edge of the chip and
51:33do and and interact? What's the what's
51:37what's going on there and and and are
51:39you you know, where does that sit on
51:41your list of checks and work in progress
51:43and that sort of thing?
51:45So, if I understood your question
51:48correctly, the question is around
51:50Chip-to-chip interaction.
51:51to have a lot of zones. That's what's
51:54called a QCCD where
51:55um your qubits are, how you're going to
51:57move them between chiplets to keep the
51:59all-to-all connectivity, right? Is that
52:01your question?
52:02they going to move between chips?
52:03Yeah, so the way we
52:06you know, again, millions of qubits is
52:08very different from 1 million qubits to
52:10100 million qubits. So, I can't tell you
52:12what I tell you now will scale across
52:14that range. But, what we see is given
52:19um how we do transport of our qubits and
52:22and the nearness before we get out of
52:24two-dimensional and have to go
52:25three-dimensional other spaces, that we
52:27can do that electrostatically, just
52:29using potential wells to make
52:31qubits jump between um chiplets or
52:34chiplet boundaries. Okay, thank you.
52:38Thank you. Right here.
52:39Hi, I'm Manus from Technical University
52:41of Munich. Let me comment first that it
52:43is very interesting that your next QPU's
52:46are the
52:47um
52:47sun in Greek, sun in Latin, and then the
52:50god of sun, uh Apollo.
52:52But, uh my question is that after
52:53Apollo, if I understood correctly, you
52:56are going to still be uh contained in a
52:58single chip, more or less. I I'm not
53:00sure if I understood if there is some
53:01kind of modularity and some kind of uh
53:03building blocks and building larger
53:06uh QPU's, distributed QPU's based on
53:08that.
53:10So, I'm not sure I understood the
53:12question 100% or didn't hear
53:14Can you Can you reframe but I think you
53:16may be going into territory where we
53:17cannot go into right now, which is going
53:20I didn't even hear the question, so Why
53:22don't you repeat the question?
53:23It's a bit similar to the previous
53:24question, but the question is um
53:26do you have any plans for distributed
53:28quantum computing and building modular
53:31um
53:31quantum computers based on some building
53:33blocks like Apollo and then
53:35interconnecting them
53:36to build something bigger?
53:39Are we thinking about it? If If I get
53:40your question, yes, are be thinking
53:42about
53:43uh in the notion of how classical
53:45computer did scale up and then scale
53:47out. Are we thinking of scale out? Yes,
53:49absolutely we're thinking about it,
53:50right? We're looking at potential ways
53:53to do it. Uh I think our interest is
53:56ensuring
53:57uh our
53:58our more near-term interest in seeing
53:59how far you can push without you have
54:02without having to go to to a scale-out
54:04distributed architecture. And we believe
54:06there's lots more room in just scale up,
54:09right? Which is building more and more
54:10dense single
54:12uh
54:13not maybe single chip, but single
54:15package
54:17uh traps that can be controlled with
54:19with lasers.
54:21Um so, at some scale, do we have to get
54:23to a distributed architecture? The
54:25answer is probably yes. Um
54:28do we think that that is the first
54:30problem we have to solve to get to
54:31Apollo in in a few generations be beyond
54:34it? No.
54:35Okay, thank you. Thank you.
54:37Candace.
54:39Hi. I'm Candace Cahain from Los Alamos
54:42National Laboratory. And um you know,
54:45exciting visionary talk. Thank you very
54:47much for that. And so, my question is a
54:50a little more philosophical, I guess. Um
54:53one of the great things about being here
54:55at Quantum Week is hearing the buzz in
54:58the rooms and the chatter over coffee.
55:00And I've heard various comments this
55:02week, and I'd like to get your
55:03perspective. I've heard some people say,
55:06"We're in the '80s, you know, it's like
55:08when massive parallel processing
55:10completely disrupted the mainframes, you
55:12know, it's this is where we are with
55:14quantum." I've heard other people
55:15saying, "Are you kidding? No, no, no,
55:17we're in the '40s. These are bespoke
55:18machines, you know, it's like having a
55:20couple of ENIACs on the planet." So,
55:23that's a pretty wide range, and I was
55:26curious from where you're standing
55:28and using an analogy to our our heritage
55:31of um binary digital computing, where do
55:34you think we are?
55:37Great question. Thanks, Candy. It's
55:38always good to
55:40I knew it would be a tough question when
55:42Candy stepped up, so
55:44um
55:45Where Where do I think we are? Uh I
55:48think we are in
55:51um the early stages.
55:54Uh if I was to use a classical computing
55:56analogy, we are more like mainframes
55:59right now.
56:00Right?
56:02But it's what happens after that
56:05is not to me an exact parallel.
56:08Right? You know, what mainframes did was
56:10mainframe the mainframe economics
56:13became tougher and tougher, other
56:15technologies developed, x86,
56:18software more importantly when open
56:19source and Linux developed, so
56:22um things became more democratized as
56:24technology layers became more
56:27stable, they got APIs and people could
56:30depend on one layer looking like that to
56:33the layer below. So, delamination of the
56:36layers happened, people would operate in
56:37systems without knowing what the machine
56:39was, but
56:41so specifications happened. I think
56:43that's a little ways off, frankly. I
56:44think still some of the hard challenges
56:47that have to be solved
56:49can only be done in co-design, and that
56:51does make it look like a mainframe for a
56:53while, right?
56:55Um
56:56at some point the delamination will
56:59happen very deep, right? You should be
57:01able to just get gate-based, you know,
57:03for instance,
57:05uh quantum computers and even today
57:07there's early signs like in ticket ti-
57:10ticket. I sometimes can't pronounce the
57:12names of my own
57:13uh software modules, but we already say
57:18you know, we can handle multiple kinds
57:19of gate-based quantum computers, but the
57:21question is can we handle them with the
57:22best performance that they can deliver,
57:24right? That still requires co-design and
57:26understanding what microarchitectural
57:29features like all-to-all our
57:30or mid-circuit measurement is present.
57:32So, I think I look at it more like the
57:34mainframe today,
57:36but I don't see it staying like the
57:39mainframe or suffering like the
57:40mainframe. Uh
57:43uh I think what will happen is
57:46it it will delaminate. There will be
57:48hardware, there will be middleware which
57:51where error correction falls, because
57:53today it's both
57:55related to the application or the
57:57algorithm, and it's also part of what
58:00hardware has to support. Uh whether that
58:02gets cleanly delaminated or not or just
58:04becomes part of the application. I mean,
58:06you submit the application, you submit
58:08the error correction that you want that
58:10is best for that application. Those are
58:11all models that will play out. The one
58:14thing I will take the opportunity, even
58:15though may not be part of your question,
58:17is
58:19this discussion and notion that it's
58:22going to disrupt as in obsolete the
58:26previous generation of technology
58:29to me is hogwash.
58:31All right? And I think No, we do this to
58:33ourselves as technologists. As we we
58:35say, "Well, we have CPUs, then we have
58:37GPUs, and GPUs are going to eat up
58:39CPUs." And to a certain extent, that it
58:40does happen.
58:42But I think the power and beauty of
58:43quantum computing is it extends
58:46human insight generation
58:50in a way that you could only do because
58:52you have classical computing also
58:54improving and the advent of technologies
58:56like AI or or generative AI or machine
58:59learning.
59:00And I think the beauty of quantum
59:02computing is
59:03going to be fully expressed when quantum
59:06computers are used
59:08to actually make things happen that you
59:12couldn't do without quantum computing
59:13being in the mix, not replacing
59:18uh classical computers. Uh a very good
59:21example is what Microsoft and us are
59:23working on, which is not saying, you
59:25know, you were doing uh mo- moleculars,
59:28you know, design by looking at the
59:30search space and saying,
59:32you know, running out of, you know,
59:34trained models to essentially go look at
59:36certain properties and and predict what
59:38kind of molecular properties would best
59:40serve them.
59:41Um
59:42and you say, oh, I'm just going to not
59:44run it on a classical, you know,
59:47DGX box or whatever it's called today.
59:50I'm just going to replace that with a
59:51quantum computer. No. The beauty of that
59:53is you use the quantum computer to
59:55actually generate unique quantum data
59:58that can train these AI models to be
1:00:01more accurate. And then the actual
1:00:03process still runs, but now you have a
1:00:05better outcome because your model is
1:00:07better trained with data that you
1:00:09couldn't have had before
1:00:10because you didn't have a quantum
1:00:11computer.
1:00:12Those are the kinds of things that I
1:00:14think are going to be big breakthroughs.
1:00:16So, when I hear this,
1:00:18oh, you're working on quantum supremacy
1:00:20or quantum advantage. True, certainly
1:00:22there will be some algorithms where
1:00:23we'll show power benefits. We've already
1:00:25done that or just performance time
1:00:27benefits. But that to me isn't the
1:00:30casting it in that light where it's A
1:00:32versus B, not A and B, is actually
1:00:35detrimental in the long run
1:00:38to
1:00:38to a a computing ecosystem that will
1:00:41need both quantum and classical.
1:00:43Sorry, long answer to your question, but
1:00:45Thank you. Thank you, honey.
1:00:47We have about 2 minutes left. So, we're
1:00:50going to run through a few questions
1:00:51that I hear at the three bucks. Thank
1:00:52you.
1:00:54Thank you. I'm Roger de Souza from from
1:00:56University of Victoria, British
1:00:58Columbia.
1:00:59So, my question is this.
1:01:02You're claiming that you're going to be
1:01:04able to move the ion trap system from
1:01:07one dimensional to two dimensions,
1:01:09right?
1:01:10Now, I don't think
1:01:12I as I understand this is a very
1:01:14non-trivial step. And one of the
1:01:16problems is the way the qubits interact
1:01:19with each other.
1:01:20So, the CNOT gate between two qubits
1:01:24uses so-called quantum bus, which is a
1:01:26vibration along the line of ions.
1:01:29So, I'm wondering how is that going to
1:01:31work in the two-dimensional
1:01:33architecture? In particular,
1:01:35the question is, are you going to retain
1:01:38all-to-all coupling in the
1:01:40two-dimensional architecture?
1:01:42And question number two, aren't you
1:01:43going to have a lot of crosstalk when
1:01:46you move to two dimensions? Okay,
1:01:48great questions. They would take two
1:01:50more than two minutes to answer. I'll
1:01:52answer part of it and then I'll make
1:01:53sure you get the answer for someone
1:01:55who's far more knowledgeable about about
1:01:57this, right? The answer is yes, we do
1:02:00foresee
1:02:02up to many levels of scale maintaining
1:02:04all-to-all connectivity.
1:02:06Now, the the answer to how we do that is
1:02:08something that I I'm happy to get you in
1:02:10touch with
1:02:12someone on our technical team who's
1:02:13actually already here.
1:02:15Um and and have you sit down and
1:02:17understand what we've done to actually
1:02:19demonstrate that that is possible.
1:02:22Right? Um so,
1:02:24the answer is yes, in order to go to two
1:02:26dimensions, we're not giving up on
1:02:27all-to-all.
1:02:29In fact, we've demonstrated that you can
1:02:30keep all-to-all.
1:02:32The question is at very large scales,
1:02:34what does all-to-all mean?
1:02:37Right? Okay.
1:02:38Thank you.
1:02:39I'm just informed that the
1:02:40conversation's going very lively, so my
1:02:42clock got upgraded by another two
1:02:44minutes. If there are more questions,
1:02:45please join in. Thank you.
1:02:47Uh right here.
1:02:48Uh
1:02:49Dan Gonzales from Rand. Uh thank you for
1:02:52a a very interesting uh and visionary
1:02:55talk.
1:02:56Uh the question I have is
1:02:59uh there are some that are concerned
1:03:01what uh that fault-tolerant quantum
1:03:03computers
1:03:05are going to require a lot of power
1:03:08uh and
1:03:10uh are are going to the the economics of
1:03:13building them and then operating them
1:03:15will be difficult.
1:03:17Um I'm
1:03:18Are you going to require cryogenics?
1:03:21And And do you see the power
1:03:24requirements scaling uh in a way that's
1:03:26economical for Apollo? Thank you.
1:03:31Um
1:03:33you know, I I was I worked hard not to
1:03:35interrupt you during the question
1:03:37because when you first asked about
1:03:38power, I said, "Man, you got to be
1:03:40kidding. Have you seen what classical
1:03:42data centers take?
1:03:44Right?" So,
1:03:46today the H2 system that takes about 20
1:03:49kW. I'm not talking megawatts. I'm
1:03:51talking kilowatts.
1:03:53Um
1:03:54to do for the basic computation, add a
1:03:57few kilowatts for the cooling system or
1:03:59cryogenics. And then I'll tell you in a
1:04:00second how that scales or why it
1:04:03is kilowatts. And it cannot be simulated
1:04:07by one of the DOE um
1:04:11you know, supercomputers that takes
1:04:12megawatts.
1:04:15Right? So, that So, I we're not even in
1:04:17the same freaking ballpark as as far as
1:04:19I'm concerned. Now, but you asked a
1:04:21great question on scaling. That's true
1:04:23today. What happens in the future,
1:04:24right?
1:04:26The power of quantum computing isn't by
1:04:29It's not the old C squared we thing,
1:04:32right? Where you're given more voltage
1:04:34and suddenly it's
1:04:35things run faster, right?
1:04:38The power We will scale power as we get
1:04:41to a larger trap, but that's to maintain
1:04:43the cryogenics to maintain vacuum, not
1:04:45to actually drive the computational
1:04:48speed, if you will, right?
1:04:50Um so, yes, when we get to, you know,
1:04:54uh
1:04:55Apollo, will we be much higher or
1:04:57somewhat higher than where we are today?
1:04:59Yes, but we won't be in the megawatts
1:05:01range.
1:05:03Uh because again, our scalability of
1:05:05performance is not tied to voltages
1:05:08scaling and and drawing more power per
1:05:10qubit. It's actually maintain vacuum
1:05:12across a broader set of of cubit
1:05:14numbers.
1:05:15Um and so I think this power
1:05:18consideration is
1:05:21the only time this should come up is if
1:05:23you're going to spend 20 megawatts of
1:05:25power to chain go chain this large
1:05:26language model, could you do it for
1:05:291/3 of that or 1/100 of that uh and do
1:05:33parts of it on a quantum computer.
1:05:36Power shouldn't isn't your biggest
1:05:37worry.
1:05:40Thank you.
1:05:41Thank you.
1:05:42That's right.
1:05:43Uh
1:05:43I'm Kenta Kondo from Kenaz Assistant
1:05:45Startup Company in Japan. So really
1:05:47exciting and encouraging roadmap. So my
1:05:50question is do you have any plan to make
1:05:54the error corrected logical cubit
1:05:55available to users before Apollo or do
1:05:59we have to wait until Apollo? Because
1:06:01the 12 cubits now logical cubits not
1:06:03available to users, right? So can we use
1:06:06as a general users the error corrected
1:06:09logical cubits? Do you have any plan
1:06:11that you can share now?
1:06:13So whatever we do, we will have it in
1:06:15our commercial systems.
1:06:17Right? So
1:06:18uh there's there's no secret enclave
1:06:21where we only have it amongst us and we
1:06:23don't offer it
1:06:24in the systems we build.
1:06:26It might be true for a while as we are
1:06:28curing the technology but it'll show up
1:06:29in one of the systems I mentioned.
1:06:32Uh we are looking at ways to make it
1:06:34more generally available beyond our
1:06:36customers. Like what do we do for an
1:06:38academic ecosystem for instance? Um
1:06:41but it isn't that we found it we'll keep
1:06:43it in the labs. It will we we only think
1:06:46it works when you put it out in a
1:06:48commercial system which is banged on by
1:06:51users, not just us.
1:06:55Thank you.
1:06:57Great. I think one last question. Uh
1:06:59hello. Uh my name is Yuri Alexy from
1:07:01Nvidia Corporation.
1:07:03So I have a question about classical
1:07:06computing
1:07:07because classical computing is still
1:07:09required, right, to operate quantum
1:07:10devices.
1:07:12As we scale up to millions of qubits,
1:07:15do you think ultimately quantum
1:07:16computing may be bound actually by
1:07:18classical computing?
1:07:19Because Jay kind of like maybe in our
1:07:21presentation, he mentioned he actually
1:07:23saw early signs that classical computing
1:07:27actually become might become actually
1:07:29limit. Do you do share this vision or
1:07:31not?
1:07:32I'm not sure.
1:07:34Can you I think Yeah, I think can you
1:07:35repeat the last part again with the
1:07:36vision with Yeah, so question is, as you
1:07:39scale up to millions of qubits,
1:07:41uh
1:07:42classical computing, I'm talking about
1:07:44let's say decoding, compiling, these
1:07:46stages, will become a bottleneck to
1:07:49capabilities of quantum computing.
1:07:53You know,
1:07:54will it become more complex? Yes.
1:07:57Right? I mean, if you're generating, you
1:07:59know,
1:08:01um not million gate circuits, deep uh
1:08:04circuits, but you're generating
1:08:0710 million,
1:08:08you know, as you get to those error
1:08:10rates, they will become more complex to
1:08:13optimize.
1:08:15Um but they will also there will also be
1:08:18automation tools like there are today
1:08:19called ticket, right? So, we we say
1:08:21focus on
1:08:23writing the code correctly.
1:08:25Mhm. We take on, because we know the
1:08:28architecture below,
1:08:30the job of optimizing the gates or the
1:08:32circuits to that. There will always be
1:08:35tools. Will the time for that increase?
1:08:37Probably. I mean,
1:08:39um
1:08:40at what stage does it become a
1:08:42bottleneck? That's actually a great
1:08:43question. But bottleneck to what, right?
1:08:45I mean, if it was going to take me,
1:08:48you know, in classical compute, you
1:08:50know, 6 months of running a model to
1:08:52figure out
1:08:54or to to reduce the search state space
1:08:57for a search, and I'm spending a day
1:09:00longer to do actual compu- compilation
1:09:02of the quantum algorithm and be able to
1:09:04run it in 30 days. They all sound very
1:09:08long time frames.
1:09:09Um
1:09:11but that's a huge win, right? So, it's
1:09:14time to solution. That's the context of
1:09:17whether something is the overhead or
1:09:18not. Not just absolute time, which could
1:09:21go up and will likely go up.
1:09:25Thank you.
1:09:26Yeah.
1:09:27I can stay and answer all kinds of
1:09:29questions.
1:09:30Thank you. One last question over there.
1:09:31Thank you.
1:09:33Thank you for the very nice talk. Uh my
1:09:35name is Peter Gresskowski. I'm at the
1:09:36Oak Ridge National Lab. So, my question
1:09:38was about your road map. Uh it looked
1:09:39like your next iteration, your next
1:09:41machine will have 96 physical qubits,
1:09:44but yet 450 logical qubits. Could you
1:09:47share what kind of error correcting
1:09:49codes you'll use to do this? Uh I'm I'm
1:09:51not going to put that out publicly. We
1:09:53are not talking about it, but I'll just
1:09:54tell you if you can get to
1:09:57um
1:09:58if you can get
1:10:01when we brought the order down to 12
1:10:03logical qubits
1:10:05uh from 56, right? We have other tricks
1:10:08in the book.
1:10:09Um there's also the fully corrected
1:10:11versus where you do post processing and
1:10:14selection. That's another uh
1:10:16avenue, although that's not in our mind,
1:10:19that's uh not true error correction, but
1:10:21there are other ways to get there.
1:10:23Uh but no, I'm not going to tell you
1:10:24today Thank you. which code, whether
1:10:26it's a steam code or something else, a
1:10:28color code that we're going to use.
1:10:30Thank you.
1:10:31For that, you'll have to wait and we'll
1:10:32write it in a paper and then then you
1:10:34can read it.
1:10:36All right. Thank you, Raj. Appreciate
1:10:37it. Thank you. Thank you. I think we
1:10:42we have meant to say, I think a lot of
1:10:44lot of the uh detailed notes, uh
1:10:46including information about the road map
1:10:48and the video that Raj mentioned about
1:10:50with the music metaphor, are available
1:10:52in our blog post on our website.
1:10:54Uh we shared that last week. Uh you can
1:10:56also, I think if you go to the blog with
1:10:58recent news with
1:10:59how we are using the combination of AI
1:11:01and quantum computing. We released a new
1:11:03paper yesterday. And this morning we
1:11:06went out with a new
1:11:08announcement with HSBC on gold
1:11:09tokenization. So
1:11:11please do enjoy reading those on our
1:11:13blog as well as our marketing
1:11:16collateral. Thank you again everybody.
1:11:17And the next session here will start at
1:11:2010:00 a.m.
1:11:21Until then, appreciate it. Thank you
1:11:22again.