Rigetti Computing, Inc. (RGTI) Earnings Call Transcript & Summary

Good morning, everyone. My name is Quinn Bolton. I'm the semiconductor and quantum computing analyst for Needham & Company. Thank you for joining us here on the first day of Needham's 27th Annual Growth Conference. It's my pleasure to host this fireside chat with Rigetti Computing. Rigetti became the second publicly traded quantum computing company in March of 2022. The company is a pioneer in full stack quantum computing and is a leading supplier of superconducting quantum processors and computers that leverage its semiconductor design expertise and vertically integrated manufacturing capabilities. Joining me from the company is Dr. Subodh Kulkarni, President and CEO. Subodh, thank you for joining us.

Thanks, Quinn.

Before I get to my questions, the company has asked me to read the safe harbor language. Certain statements in this fireside chat may be considered forward-looking statements within the meaning of the federal securities laws. Forward-looking statements generally relate to future events and can be identified by terminology such as may, should, could, might, plan, possible, expect, believe, anticipate or the negative of these terms or variations of them or similar terminology and are subject to risks and uncertainties. Factors that may cause actual results to differ materially from current expectations are set forth in Rigetti's annual report on Form 10-K for the year ended December 31, 2023, quarterly reports on Form 10-Q and other documents filed with -- by the company from time to time with the SEC. We urge you to review these risk factors and cautionary statements contained in the filings. Thank you for bearing with me for that.

Subodh, maybe just since the Quantum sector has certainly caught the attention of many investors over the last 3 to 4 months. We may have some folks that are newer to the company. Can you give us a quick overview of Rigetti, what differentiates Rigetti from other quantum computing hardware vendors?

So thank you for having me, Quinn, and I'm glad you had to do the safe harbor statement and not me. So Rigetti is a public company based in Berkeley, California. We are at about 140 employees. We call ourselves a full stack quantum computing company. That means we do everything from designing the chip to building a chip. Our fab is in Fremont, California and all the layers of the stack, the control systems, the dilution refrigeration, the firmware, software and so on. What is unique about Rigetti, we are about 140 employees right now. What is unique about Rigetti is, we started about 10 years ago, founded by Dr. Chad Rigetti, have evolved over the time period. So we are one of the early entrants. And what we have designed today is an open modular stack. So unlike some other companies, like big tech companies, like IBM and Google who play in this space, who deal with a quantum computer like a mainframe computer, where it's all integrated. We have intentionally [ carry ] our stack open-end modular so we can slide in components that may be more creative coming from outside than what we can do internally. So we can do everything internally. But when, for instance, NVIDIA comes up with a CUDA-Q, quantum version of CUDA, we could easily integrate that into our stack, whereas if you do it in a mainframe approach is difficult, same with error correction companies like Riverlane and so on. So right now, we are developing, making quantum computers. They are already there. We have sold a few and we'll get into that. So they work, they are not vaporware anymore. We continuously get those questions. Is it real? Is it happening? It's already happening in that sense? We are developing, making quantum computers. Customers are using them. All for research purposes, we are clearly still in the R&D mode right now.

Perfect. The company has chosen the superconducting qubit modality for its quantum computers. Can you talk about the advantages of superconducting qubits versus the other modalities, ion-trap, neutral-ion photonic. And are there any disadvantages today for superconducting?

So certainly. So within quantum computing, before we get into superconducting, there are 2 different approaches. One is annealing or entropy-based approach, which is not quite computing per se, but it is quantum effects. We don't deal with that. We are more in the gate-based computing approaches, which is the much larger general purpose computing market. Within the gate-based computing approaches, there are several modalities, as you pointed out. Most of the attention right now is on superconducting, which is where we are. But trapped ion gets a lot of attention, pure items, photonics, spin and a few other modalities like that. We obviously belong to superconducting cam. We believe that's the cam that's most likely to win, along with us, you have IBM, Google, Amazon, the government of China and a few other companies investing in superconducting modality. The big advantage of superconducting modality are scalability and gate speed, scalability because we are essentially dealing with semiconductor chip like technology. And just like semiconductor chips, once you have the fundamental architecture defined and the transistors in our case qubits working properly, you can scale up fairly easily using all the known tricks in semiconductor industry, lithography, deposition, etching all the standard stuff we have done for the last 50 years in semiconductor industry are leverageable here. So we feel we have a huge advantage in terms of scalability and going up in in superconducting side. Equally important is gate speed, that's the clock speed, that's the speed at which our computing happens. We are fundamentally dealing with electron speed just like CPUs and GPUs today with CMOS technology. So our gate speeds are in tens of nanoseconds. Like right now, our gate speed is about 15 nano seconds. And that's typically 4 orders of magnitude, so 10,000x faster than trapped ion or pure atom type modalities. So those are big strengths we have scalability and gate space. On the flip side, historically, superconducting camps, biggest challenge was fidelity because we deal with engineered devices. So we used to get more errors than typically trapped ion or pure atoms where they are dealing with pure ions or pure atoms. But as you can see recently from Google Willow announcement or our own press release at the end of December, we are basically caught up. So if anything, Google's Willow, our Ankaa-3 is ahead of companies like IonQ's fidelity right now. IonQ at 99.4% fidelity. We are at 99.5% fidelity. Google is at 99.7% fidelity. So the historic disadvantage is superconducting, we believe we have more than quarter and actually in some cases. But certainly, our advantages in terms of scalability and gate speed continue to be the case. So we feel very strongly that of all the different modalities, superconducting has the most likely chances to win. And look no further, right? Why would a company like Amazon and Government of China invest in superconducting if they didn't think this was a winning modality. They could have chosen any different modalities to invest in. The fact that they choose to invest in superconducting and not trapped ion or pure atoms of spend that should tell you something right there.

Right. You started touching on it a little bit earlier, but just Rigetti approach to hardware sales is a little different than some of your competitors in that you'll oftentimes sell just the quantum processor chip to QPU rather than the complete system with the dilution refrigerator. Can you talk about sort of why you pursued that strategy in terms of processor sales, not necessarily entire systems?

Those are all good questions. Fundamentally, our view about quantum computing is a very different than some of our peer companies. We believe quantum computing is going to exist in the ecosystem of hybrid computing. So we'll always be in parallel with CPUs and GPUs. We'll never be in an isolated situation. And we'll always interface with CPUs and GPUs, just like the CPU and GPU interest each other right now. Secondly, I mean, you asked about sales. We still view sales is a bit further out than where we are, and that's why we have talked about it before. And I keep saying we are very much in tech dev mode right now and sales don't matter at this point. What's far more important is the technology milestones, the fidelity, milestones, the qubit count, the gate speed, those are far more important than the few -- all the sales that we have reported at most of it anyway, have been government contracts. And I believe that's the same case for our peer companies as well. So anytime we talk about sales and one contract that certainly spikes up the sales. I don't think those really matter in the big long term here. It's the technology milestones that are important. That's why we don't like to talk too much about sales at this point. Long term, of course, that becomes important. So getting to your point, we believe quantum computing in the hybrid computing ecosystem has to evolve. There has to be an ecosystem that comes along. It cannot be just one company trying to dominate the whole system. So we opted to -- and because of our open modular stack, we opted to allow our customers and the government national labs, the acedemic researchers even corporations. So as of today, I don't believe any corporation has bought a quantum computer yet. It's mostly been government labs. But they are experts in different areas. So one of our customers, for example, is Fermilab in Chicago. And Fermilab is one the inventors of cryogenic systems in the world. So they obviously understand dilution refrigeration probably more than anyone else around this planet. So it kind of is a little idiotic for us to try to sell them a dilution refrigeration technology. So what we told them is that you keep your dilution refrigeration technology will give you the QPU portion that you can interface with your dilutions refrigeration technology. So we opened up our stack. We are modular. So if a customer wants the full stack from us, we are happy to give the full stack. But if they just want a QPU portion, we are happy to do that. In some cases, like Air Force Research Lab, they have expertise and they want to design their own chips, and we will allow them to do that. So they actually give us a design for the chip and we will build a chip in our fab and give it back to them. So we are open to working with creative arrangements with our customers, and we want to keep it that way at this stage in the game because we are still in R&D. Once it becomes a commercial business, then the discussion should evolve to what is the right business for the long term.

Got it. Got it. The quantum computing industry certainly caught the eye of investors over the last 3 to 4 months. What do you think some of the most important developments at the industry level have been over the last -- over that period of time. And specifically, you mentioned Google's Willow and their 105-qubit processor. Can you share sort of your thoughts as to the importance of this announcement and how it might affect we're getting?

Well, I believe several announcements happened in the last few months that led to the significant increase in investor interest in the quantum computing sector. It probably started with government of China announcing that they have started breaking AES encryption with quantum computing, then, of course, there was the Trump victory that led to more interesting small camps and emerging technologies. Then Amazon made an announcement that they are increasing their investment in quantum computing. And then, of course, Google's Willow announcement was a big one. To us -- I mean, we had a couple of important press releases along the way, just before Google's Willow announcement, we announce that we have accomplished real-time error correction code with very low agency with Riverlane in U.K. Although our announcements don't get as much notice as the Google's Willow announcement, but that's a very important technical milestone that we demonstrated. And then after Google Willow announcement, we announced our 84-qubit chip at 99.5% that has been deployed already on AWS and Azure now. So many factors have led up to the run up in stocks of pure play companies like us. Without a doubt, the biggest of them was the Google Willow announcement. What Google did is twofold things. although the attention goes on the Willow chip, and as you said, it's 105-qubit, 99.7%, awesome accomplishment. But equally, if not more important, is the concept of using surface error correction code. And what they showed in that paper is that as you increase the length of the surface error correction code. In other words, as you increase the qubit count, your errors actually come down. And that's counterintuitive because normally, as you add qubits, your errors go up. So Google's demonstration was a very powerful demonstration that once you start deploying surface error codes, you can actually bring it down. Now surface error codes have not been invented by Google. They have been around, but it was -- as far as we know, it was the first deployment at that scale where Google demonstrated something like that. Now obviously, we have been working with Riverlane and we'll continue to do so. So expect us to be doing surface error codes and similar demonstrations in the next few months here. But that was a very important announcement and that led to the interest.

I guess with Willow and the surface codes, I mean, isn't that the whole concept of quantum error correction as you deploy it, the logical -- the fidelity of the logical qubit should be much higher or error rate much lower than the underlying physical qubits.

So -- but it was all on paper, now it's all real.

No, it's okay. Got it. The last question sort of at a high level -- excuse me, I was hoping to get your comments in response to comments made last week by Jensen that it could take 15 to 30 years before we have commercially available quantum computers. And I guess Zuckerberg made some comments either over the weekend...

Couple of days ago.

Yesterday as well, which I think sort of implied the same sort of it may be a while before we get a usable quantum computer. How do you respond to those comments?

Well, it's tough to respond to vague words like useful, right? What does useful mean? Somebody may find something useful today may not find it up even 5 years from now, that kind of -- so I hate to get into those subjective vague words particularly going with someone like Jensen or Mark Zuckerberg whose microphone is a lot bigger than any one of us here in the room here. So rather than getting to that I prefer to stay with the technical milestones, which we know. And so we can talk fidelity, we can talk qubit count, we can talk gate speed and other metrics. I know what our road map is, I know -- I have a good feel forward IBM, Google's road map and some other company's road maps are. What I can say is that a couple of years, we already had at 84-qubit, 99.5%, 15 nano second gate speed, I mentioned that. In a couple of years, I'm pretty sure we'll be at a few hundred qubits at 99.7%, 99.8% fidelity, maybe 30, 40 nano second gauge speed. We think that would be good enough to start demonstrating what we call narrow quantum advantage, which is demonstrating superiority with the quantum computing over classical computing for some select applications. Another a couple of years later, so maybe 3 to 5 years from now, when we start approaching 1,000 qubit at 99.8%, with surfaced error codes that we just talked about earlier. We think we will be able to demonstrate quantum advantage, QA over -- so quantum computing superiority or classical computing for most practical applications. And that's the time when we think commercial customers will start showing interest in quantum computing in their data centers, whether quantum computing as a service or on-premise quantum computing. So somewhere 4 to 5 years from now is our view and commercial interest. Now if you want to define usefulness as that, that's what I would pick as a number about 4 to 5 years from now. But academic researchers are using our quantum computers today and our peers, companies quantum computers and they're finding value in that. So hopefully, that answers your question.

Yes. No, that's great. You sort of mentioned the Ankaa system, so maybe we'll skip over that. But looking to the 2025, you mentioned technical milestones, probably the key that investors should be looking for? Maybe talk about your '25 road map. You've recently shifted your road map now to focus on tiling QPUs together. And so walk us through why tiling and what you hope to accomplish this year?

So very important question, and that's really where we are starting to differentiate ourselves from our peer companies like IBM and Google in the superconducting camp. Our view is that to reach thousands and tens of thousands of qubits, we will need to tie the chips together, just like the CMOS industry. And the reason is simply the same reason why I see most industry does chiplets right now. If you look at the most advanced nodes, 3-nanometer, 5-nanometer nodes, so take your latest iPhone or latest gadget, chances are very high that you are using chiplet right now. And the reason that happens is because fundamentally, it's very relatively easy to get a high yield smaller size chip than a larger size chip just because uniformities and everything is difficult to control on a larger area than a smaller area. And we find the same in the quantum world. So our 9-qubit chip, if you will, is a 6-millimeter by 6-millimeter chip. Our 84-qubit Ankaa chip is 1 centimeter by 1.5 centimeter. So we find that our yield on the 9-qubit chip is far higher than our 84-qubit chip for the same reasons why CMOS industry. So our view is that this is going to continue. I mean, CMOS industry has tried for decades to try to get a single chip to a highly high yield. There is a reason why they ended up with a chiplet format. And we see the same way. So I try to reinvent a deal when the CMOS industry has already paved the path here. So we are jumping on to the chiplet approach. We already demonstrated that you can target 2 chips without losing any performance. We did that once with 20-qubit chips, we did that under the time that 9-qubit ships at very high fidelity. That gives us the confidence that it's time to embark on a tiling approach with [ 4 9 ] qubits, that's the intermediate milestone we want to demonstrate middle of this year. So even though qubit count will go down, so we'll go from 84% to 36, but instead of a single monolithic chip, now you have [ 4 9 ] chips. And it's important to demonstrate that, and we want to get futility up to 99.5% for a generic gates 99.7% with the FSM get. So we'll bump up the fail from where we are today, but more importantly, we'll demonstrate 4 chiplets, if you will. Assuming that we succeed with that, then we will try to continue that. And we think that's the right way to go to several thousands and several hundred thousand qubits. IBM has said something similar in their disclosures that they are thinking about a chiplet approach. But as of today, we haven't seen anything, and we haven't heard anything from Google about that or any other companies. So we believe we are the first ones to embark on this milestone.

Okay. Got it. In August, you released a paper introducing a new chip fabrication process called Alternating-Bias Assisted Annealing, ABAA, which allows you to sort of tune the frequency of the qubits. Talk about the advantages of that new process and sort of how that helps you increase the fidelity of your qubits.

So fundamentally, when we build a qubit, I mean, essentially, in physics words, it's called a Joseph Injunction. It's a sandwich structure of aluminum, aluminum oxide and aluminum, and we are sending a microwave pulse to one aluminum, it shoots a pair of electron and coal to the other side of aluminum. And in the insulating aluminum oxide layer is when all these quantum effects are being observed and that's what we are using to do the computation. That's the simple lay man's description of what's going on in a fairly complex quantum physics world. So this micro frequencies that we send to the qubit to get it initiated in the quantum state, in the ideal world, we would get a perfect frequency for every qubit, but we don't because there's fabrication methods, there's all kinds of things. So we get a frequency spread of qubits. The tighter the frequency spread, the better control we have, the better fidelity, the better everything. So we try to tighten up the frequency spread. Historically, we and others like IBM, what we have done is we did laser-annealing approaches. So what we do is once we build a qubit, we actually go around the qubit with a high-power laser, and that heats the perimeter of the qubit and then the heat permeates towards the center of the qubit. So you're essentially using the heat of the laser to adjust the morphology, if you will, of the aluminum or momoxide interfaces. And that change of interface enables us to get a tightening of the frequency distribution, which also means the irregularities in the interface is the root cause of the frequency spread of the qubit. That's been done for a few years now. It's not that it's a bad approach. It's been working, and that's how we and IBM and Google have managed to get the systems to where they are right now. What we did like about laser annealing approach is it's slow because it's individual going around laser. It's not a scalable process. You can do it for 100, maybe even 500 qubits. But imagine doing 100,000 qubit chip or chiplets and going around each qubit one at a time, all the alignment issues, the cost, the pain index associated with it, it's just not a manufacturable process. So we have been experimenting with many approaches and one of the approaches that seems to work really well is this ABAA, we call it Alternate-Bias Assisted Annealing. Essentially, it's DC pulses, plus/minus DC pulses through the entire surface area of the qubit. So essentially, once the chip is built, we are proving the chip to test it for all the room temperature properties. So it's already a probe sitting on top of every qubit. We are using the same book to drive DC pulses plus or minus through the qubit. So it's literally a part of the qubit manufacturing processes. But we found that with the right sequence of DC pulsing you can alter the morphology if anything better than the laser annealing approach. And it makes sense because we are sending the DC flux right to the surface area of the qubit instead of from a perimeter heating. So basically, a flux through the entire area of qubit is far superior than perimeter heating control morphological adjustment. And we find frequency spreading to be remarkably better. So our frequency spreading is significantly better, almost 10x better than what we were getting with the laser-annealing approach. So certainly, it's a key part of our go-forward process is the ABAA annealing. And of course, there's a lot of IP that surrounds it.

And was that sort of used the ABAA to get to the 99.5%?

Absolutely. Yes. But in all fairness, we could have gotten 99.5% or better without ABAA. Case in point, Google Willow. I don't believe Google Willow is using ABAA. I believe they are using something like laser-annealing. They obviously managed to get it to 99.7%. But I would dare say that to get to 10,000 qubit chiplet or chip, we will need an ABAA like manufacturing process.

Got it. Okay. So as you scale the qubits, it becomes more and more...

Yes, much more important.

Yes. Okay. You've mentioned that the quantum error correction work you've done with Riverlane, maybe spend a minute, what you're able to show in your work with Riverlane?

So first, I mean, we do our own -- we did our own are correction. We continue to do it. But as I mentioned earlier, because of our open modular stack, if a company X comes up with a creative solution and is willing to partner with us, we will do that. And Riverlane is a good example. Riverlane is a small company in Cambridge U.K., about the same size as we are, 140, 150 people, very creative. They do fantastic work in error correction. So when we saw what they were doing, we partnered, we started incorporating their error correction code in our stack. And a couple of months ago now, we've published some papers with them and demonstrated real-time error correction with very, very low latency. It's a very powerful demonstration. So all of the other corrections are done in the backdrop. So you generate errors, you do it offline and then you bring back the results. But this time, literally, you're doing real time as you are running the computing and as you're going down the depth of circuit, as you discover errors, you fix them on the fly, which is very important in practical computing world. So if you're just doing an experiment, it's fine, you can do off-line. But if you want to do a real computing, you need real-time error correction. So that was a very important demonstration and the latency was extremely small. I mean, all of this is in the papers. The next challenge we have, obviously, is do what Google Willow did, which is surface error correction code and demonstrate that you can reduce the errors with longer surface error correction code. No reason it won't happen. We just had to demonstrate it with Riverlane now.

And to demonstrate the surface code, do you need to get to 100-plus qubits sort of like Willow did? Or I mean, I guess, it depends on the length of the surface?