Which quantum computer will actually scale? In Part 2 of our quantum computing deep dive, Lester Nare and Krishna Choudhary move from theory to hardware - comparing superconducting qubits, trapped ions, neutral atoms, and silicon spin qubits before going inside the new Nature cover paper Krishna co-authored with the HRL Quantum Team and collaborators. The episode begins with a simple question: what makes a good quantum computer? We evaluate each architecture using three criteria: qubit quality, qubit control, and scalability and economics. Superconducting qubits offer extremely fast operations, but scaling them introduces challenges involving microwave control, frequency crowding, cryogenic wiring, physical size, and cooling. Trapped ions preserve quantum information for extraordinary lengths of time, but their slower gates and increasingly complex optical systems introduce a different set of tradeoffs. Neutral atoms can be arranged in dense, reconfigurable arrays using optical tweezers and entangled through Rydberg interactions, while raising questions involving atom loss, correlated noise, readout, and execution time. Then we get to silicon. Beginning with the Loss-DiVincenzo proposal, Krishna explains how individual electron spins can be confined inside semiconductor quantum dots, manipulated through exchange interactions, and measured using single-electron transistors. We then explore exchange-only qubits, where three electron spins encode a single qubit and quantum gates can be performed using electrical control. That leads to the Nature cover paper, A digitally controlled silicon quantum processing unit. The HRL system integrates 18 encoded qubits built from 54 quantum dots with cryogenic control electronics, a superconducting interconnect, automated calibration, and an engineered silicon-germanium heterostructure. Krishna also explains his own work using machine learning to automate quantum-device tuning - an essential problem if spin-qubit systems are ever going to grow from dozens of components to millions. The larger thesis is about manufacturing. The semiconductor industry has spent decades learning how to fabricate silicon devices at enormous scale. If quantum processors can inherit that infrastructure, the architecture that ultimately wins may not be the one that reaches the finish line first - but the one humanity already knows how to manufacture. Nature paper:A digitally controlled silicon quantum processing unitDOI: 10.1038/s41586-026-10754-7https://www.nature.com/articles/s41586-026-10754-7 Explore the FFP Science Transfer Portal:ffppod.com/transfers Support the show:ffppod.com/donate Follow:@FFPPod on X / Instagram / TikTok / Facebook
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
If you want to build a 100,000-cubit-trapped ion machine, you're going to have to invent a 100,000 laser optical miracle. Yeah. All right? But if you want a million superconducting qubits, you're going to have to build a warehouse-sized priostat. But if you want a million spin qubits, you just put it on a standard 300-millimeter silicon wafer. You run it through the exact same photolithography machines at TSMC. asml intel scaling is the key yeah and scaling is the future hello internet this is your captain speaking lester nare joined as always by my co-host and our resident phd krishna chowdhury this is part two
of our two-part deep dive on quantum computing. We are covering the recent Nature Cover Story, Volume 655, Issue 8125, released on July 30th, 2026, and featuring our resident PhD, Krishna, as one of the co-authors. For those of you joining us from Part 1 in the last episode, we did a deep dive into Now we get to the beef.
We talked about what is it? Why would we want to build quantum computer? In this episode, we're going to take a deep dive into how to actually make one. And Krishna is going to make the case that there is only one way. Make it silicon. As a brief note, the opinions expressed in this episode are those of us personally here as the hosts and do not represent the views or opinions or positions of any entities that may be mentioned previously.
throughout this episode as we talked about in the last episode. This is a burgeoning trillion dollar industry. We are talking about this from our personal viewpoints and the opinions expressed in this episode are our own. As always, we are going to talk about the science from the ground up today because this is from First Principles. It's change only coobits, they feel so right. Scalable quantum shining in the night. It's change only coobits.
No noise scene. Exchange only cool bits. Do it with ease. Scalable bottle up. Longtime listeners of this podcast might recognize that the intro music today is different from our usual music. And that's because today is a very special episode. We're going to repeat the song at the very end of the episode in the credits, and you can take a listen afterwards, and hopefully you'll understand the lyrics. The reason why this is a special episode is because, you know, we've covered probably a hundred scientific papers on this podcast over the past year.
a scientific paper where I am co-author. I'm one of 250. It's a very large list of co-authors because it took a lot of human power, a lot of resources to do what we've done. But it's a special day. It's a special episode. I'm holding it right here. Nature's Volume 655, Issue 8125. Quantum Silicon. All right. Look at it. Look at it.
Shout out to John Carpenter for the visuals on the cover. So this is a paper out of HRL Labs in Malibu, which was my former workplace. And it was recently announced that IBM has acquired the place. So hopefully... By the time this episode ends, it'll give you a little bit of context about why that is a very interesting move. Okay? Now, this is not the first nature cover story to have quantum computing or adjacent things. Okay? We've got several. Mighty Atoms is a trapped ion story.
Quantum supremacy, that's the Google one about quantum supremacy. There's the weird, like, holographic wormhole thing. I don't know if you remember, but, like... Some people did some quantum experiment on a computer that suggested that they've finally realized the Einstein-Rosen bridge and like wormholes exist. We tapped into it. And it's like, no, you know, that's not what happened. But in any case, the current cover that I'm holding here is in a long line. It's kind of the debut of a type of MVP, you know, that they say in VC circles.
Minimum viable product. That's right. This is an MVP for a type of quantum computer that I think will be the future quantum computer. Okay. Why? Well, because it's in the name right here. Quantum Silicon. It's made out of silicon, which means that it can be made using the same machines and the same infrastructure that makes all of the classical computers that are in my phone, that are in this laptop, and in everything that we know and love. whenever we have some new technology or some new paradigm, the question becomes, okay, you can prove it in the lab, but then how do you scale it in a production environment to actually make it viable for business use cases?
Graphing. And to that point, you know, there's a, there's a, gap between the theoretical application and the production scale application. That's right. And so part of where I think you're saying we are is there are different approaches to how can we make this work at scale. Yes. And you're going to make the argument today that this debut is the way to make it work at scale because there's been promise around quantum computers for some time and it's
we will have a fundamental understanding of why this is the way forward. Exactly, yeah. And to be clear, you've already given the disclaimer, but I'll just say it again. This is not a normal episode in that we will be covering a lot of cool science. I mean, it is kind of a normal episode because we will be covering a lot of cool science and science history from First Principles, but it will be unlike other episodes. um because it will not be unbiased science journalism now i don't know if the other episodes were unbiased science journalism to to be prefer perfectly honest but um here i definitely have an agenda and the agenda will become pretty obvious as the episode goes on but i want to make it explicitly clear here so there's no confusion and debate and people are in the comments being like you're biased you're drinking the kool-aid yeah it's my kool-aid
I made it. And I'm drinking it. Yeah. Purple. Red. Exactly. So this will be an opinionated episode and the opinions are entirely our own. I do want to also just note for those who are audio listeners, as we were setting up the production for this episode, this episode has the most visual overlays of any episode we ever, we've ever done. Yeah. It's quite, there's a lot of complex ideas that are only best expressed by the by looking at it visually. And so to the extent that you can catch it on YouTube or on Spotify, where we'll have the full video available, I do encourage folks after maybe a first listen via audio to do so because it is going to be very visually heavy.
Yeah, yeah, yeah. And it's going to be a lot of fun. I mean, I had a lot of fun planning out the notes for this episode. This is going to be fantastic. I mean, it's a paper that I'm on. And we're actually going to go into a deep dive, not just about silicon quantum computers, which is the subject of the paper, but quantum computing hardware in general. And I want to answer the question, how do you make a good computer? quantum computer and what does that word good mean in this context right um there's going to be a shallow deep dive on other quantum technologies you know we've got superconducting circuits trapped ions neutral atoms and then i'm going to make the case for why this technology quantum is going to win there's the agenda yes that's that's the agenda the agenda yeah and um i used to work at hrl which is where the paper comes from but now i work at a company called dirac
Great name, by the way. And Dirac also works on spin quantum computers in silicon. So it's always been spins for me. That's fair. There's these other options, but you've been in the spin universe in terms of this is the way. Yeah, this is the way. This is the way. Because it's still early in this race to make the first quantum computer. Other players might want you to not think that, right? Quantum supremacy on the cover of Nature. They did it. Yeah. Well, okay, you simulated a probability distribution that is designed for a quantum computer to simulate because it was a quantum probability distribution.
Okay, all right, cool. There are different strategies. We've got superconducting circuits, as I said, trapped ions, neutral atoms. And for a very long time, actually, those technologies have been members of the Premier League system. of quantum the premier league of quantum yeah yeah get in yeah yeah yeah and silicon has been like the underdog like what's the what's the championship the champ under the premier league the championship okay so what's under championship oh uh uh we'll just say third division yeah yeah yeah i would say silicon has been like third division it hasn't even been in the championship yeah i mean the premier league the championship and then there's one more and everyone's gonna hate me for not remembering yeah we've been we've been like wrexham you know you know the ryan reynolds
Yeah, it's been like Manchester United, Arsenal, and Chelsea. That's been superconducting circuits, trapped ions, and neutral atoms. And then Silicon has kind of been like Wrexham. Okay. No, that's really good. Who's basically every season... Yeah. Gone up. It's gone up from being in non-league play, which is like below six. Yeah. Yeah. And to really demonstrate that worldview, let me share a story with you. Okay. Okay. So I'm going to bring you back to March meeting, 2026.
We're going to go back to this event a lot, but the American Physical Society, APS, has a March meeting every year. Oh, I remember this. And this 2026, I got to present on some of the work that actually contributed to my getting on this paper. Yeah. So I was giving a talk on behalf of HRL. There I am. And I'm actually wearing the FFP pod. That was, I think, the first sweater that we had made to try out whether merch would work. And so I had the platonic solids and I was wearing it.
I was representing. This was a meeting where this technology had its big debut to the outside world. All of the stuff that you see on this paper, for the first time, the outside world got to see it at this meeting. Okay. there's a juicy little backstory that we'll get into, but that night. Okay. So after that, after that presentation, I'm feeling good. A lot of people said it was a great presentation. You know, I'm having, I'm, I'm, I'm riding high. Right. So I snag an invite to the Alison Bob party bar in Denver.
Alison Bob is a quantum computing company out of Europe. Yes. And they like, you know, I guess rented out a bar and then you, you got to get, get like mail in invites. And then one of my friends, um, from PhD days, Elliot boar, shout out to Elliot. Shout out Elliot. He got me a fan, long time listener. Yeah. And he, he got me the, he forwarded me the email invite. So then I got to go. They had some great drinks at the bar, by the way. And they were all quantum themed. So I took a photo of the cocktail menu. You got mango collider, strawberry field theory, Bellini barrier, like a tunneling barrier, uncertainty principle.
That was kind of lazy. Dark matter, mocking Bellini. I didn't understand that. If someone has an idea of why that's quantum related. Anyways, I wanted to be a good guest. Right. You got to be a good guest. So you got to try everything. Yes. They put a lot of thought into the cocktail menu. So I got to try every single cocktail. I'm not going to disrespect as a guest. I try every single cocktail. I'm feeling good. Just had a great talk.
Been getting compliments all day. And this is a nice like day ender. Me and my friends, we see the traditional group of Europeans outside the bar in the smoking section. Yes. This is true for any international physics, even just science conference in general. There's going to be a gaggle of Europeans outside the bar just chain smoking. Okay? And I wanted to enjoy some fresh air in Denver. Yes. We head over there to socialize. Yes, with our international collaborators. Yes, exactly.
For our colleagues to get some fresh air straight into my lungs. And again, I've had a couple of beers. So I'm away from home. So some fresh air in my lungs is not – it's permissible. And I had a great talk. You know, whatever. We joined them. We get to socializing, find out that they're all from Munich, Germany. Yes. Okay. The conversation turns to work. Okay. And one of the scientists, yeah, one of the scientists asked me, you know, what do you work on? And I said, automation in spin qubits. Yeah. And they go, spin qubits?
What's that? And it's as if I had said, like, qubits made out of cheese. Like, they just had no idea. They'd never even heard of spin qubits. I'm like, okay. I try to keep the conversation going. I return the courtesy. I'm like, all right, what do you work on? They say neutral atoms. And I say, oh, yeah, okay. Okay. And then the killer, this German person, they say, yeah, see, you know about my technology, but I don't know about yours. So maybe that says something.
There is a reason, yeah? And everyone starts laughing and I'm like an idiot. You know, I can't. I'm caught off guard so much that I'm just laughing at myself. I was so ill prepared for this ambush. For the quip. Yeah, for the quip. I thought we were having a good time. It was a cool quip. Yeah. Like you could, you could have just been nice. Right. Like if somebody, if somebody comes up to me at APS and they're like, I work at Microsoft on topological qubits, I'm not going to be like, what topological qubits?
Right. I'm not going to say, I'm not going to say, are the topological qubits in the room with us? Right. Like, I mean, it's true. I could think it, but I'm not going to say it. And the background around that joke is Microsoft has... These things that they call topological qubits. Not everyone in the industry is convinced that it's even a qubit. It's a two-level system, but they might have something. They just haven't shown it yet. And there's a running joke that the only people who believe that Microsoft has a qubit is Microsoft. Okay, so that's a joke. But if I'm at a bar, I'm not going to say that.
Yes. Right? But you also don't have a German sense of humor. Yeah, yeah, yeah. And these Germans, right? It's also hard to have a comeback to, I haven't even heard of that. Yeah, yeah, yeah. Right? Like... He doesn't even go here. Yeah, yeah. It's like, remember in Pirates of the Caribbean when Jack Sparrow gets captured by the British dude and the British guy's like, you must be the worst pirate I have ever heard of. And then Jack Sparrow goes, but you have heard of him.
do that that's good never even heard of me yeah right right right right and so what do you and when so what do you there there's no clap back there's no like possibility even even in the moment i couldn't think of anything so i go back to my hotel room and i don't know if you do this but whenever like things like that happen you take a long shower oh yeah and you just start like fantasizing yeah and rehearsing like all the different ways that could have gone down like like dr strange yes All the multiverses. All the many worlds. Yeah, all the many worlds. That we don't exist in.
Yeah, yeah, yeah. Of like, where I'm like, I look like Shah Rukh Khan. And I'm like, I'm just, you know, there's a comeback after comeback. I started thinking about like, stuff I could have said. You know, one of them could have been like, you know, OK, so you haven't heard me. You haven't heard of me. Maybe I know more than you. That was the that was the like the layup. Yeah. Like maybe like maybe read the literature. Oh, I'm sorry that you're not well versed. Yeah. Yeah. Yeah. Yeah, exactly. It's like it's like I thought like not knowing stopped being cool in like middle school.
Remember those guys would just like love to not know like, shit, I don't know. Like, how is that something that we're still doing at APS March meeting? Sorry, I'm getting a little... No, it's okay. You know, the PTSD is palpable. Bro, it's ridiculous. But guess what? Who's on the cover of Nature? That's right. Do you like apples? Well, I got a Nature cover. How about Dem apples? I guess you should have known. Right? Yeah. I mean... So what I would like the audience to do right now is try to figure out just using neutral atoms, right?
Like something I could have said. It can be about Germany and how they lost the war. It can be just anything. I want to read these comments because I don't pay attention to my mental health, and this is going to be my therapy to get over this. Okay? So in the comments, drop something about neutrality, neutral atoms, Switzerland, anything. I want to read all about it. You don't have to know about the physics, although later I'll get into the physics and maybe— You can work that in. But this is an ask for the audience for help in my PTSD recovery.
And I think part of what you're sort of setting up here is... The methodology that you all were working on was below zero, even in just like awareness, the idea that it was possible. Yeah. You were not even in the boat that was in the race. Yeah. There's really like people say there's three. Maybe photonics can be the fourth one, but for a long time, it's just been those four. And those are the four that are taken seriously. Now, the serious practitioners in those four know what a dark horse quantum silicon is, okay?
And what silicon qubits can do. I think this person was just like maybe an early grad student, you know, just starting out, trying to be funny in front of his lab, whatever, you know? Yeah, yeah. Whatever. I'm over it. No, I'm not. But, you know, it's... I've been thinking about that event ever since. Because I think it sort of reflects the idea of the state of play in the race. Yeah. Or what is the best approach. Yeah. Yeah. And that person clearly thinks it's neutral Adams, right?
It's been in the back of my mind ever since that day because I knew that I'd be doing this episode. I knew that the work was going to come out. I knew that it was big enough that it was definitely going to get into one of the top journals. It's always a crapshoot to get on the cover, but there was a chance that we would be on the cover. And I knew that at some point I would cover it on the podcast. And I was just trying to think like... How the clapback episode. Yes. Like, how was I going to structure this in that one hour long shower? You know, I was thinking about all the different ways that I could clap back.
And so this is that clapback. Yeah. This episode is that clapback. You know, I think that story, right. It gives a good sense of like where the field was. And what people thought about spins. But now I get to talk about it on my show. Yes. So with that, let's actually get started. I'm going to repeat the agenda for the rest of the episode. The agenda is the following. First, we're going to talk about what makes a good quantum computer. What are the criteria that a good quantum computer needs to satisfy?
Next, we'll go over the other computing technologies, superconducting circuits, trapped ions, and, of course, neutral atoms from our German boy. And then finally, we'll get into quantum silicon, this paper that's out in nature. We'll go over the history of the field briefly and where the recent paper fits in. And we're going to make sense of this cover photo because this cover photo has a lot of really cool tidbits that I think will be cool to understand, like why certain parts are highlighted, things like that. It's quite nice. Yeah.
If you notice, there's three parts highlighted here. There's this square up here. Then there's this like sort of... Half pipe, a quarter pipe. Quarter pipe looking thing. It's like a ribbon. Yeah. And then there's this tiny little square down here. So those are the three that are the big three that came out of this paper. And we're going to talk about all those. And what this paper means for the future of quantum computing in general. Yeah. So that's going to be the episode. And before that, let's do some housekeeping. We'll do some brief housekeeping. For those joining us for the first time, welcome.
This is the best science show on the planet. For those who really want to understand the fundamentals of the latest breaking frontier science research, as always, the best way that you can support the pod is a like, a share, a comment. Put it in the group chat, five star if you're listening on any of the podcast networks. Really helps us get this show to more people. We are available in video on Spotify and YouTube, as we mentioned. Again, we are waiting for Spotify creators to do video distribution to Apple Podcasts.
For those commenting on it, we have not forgotten. It is just in process by late 2026. And so we should see that hopefully before our Christmastime wishes get done. I'll try to keep this tight today. So a couple of quick other points of order for our winners of our one-year merch anniversary episode. We have just sent out those emails to you. And so you should have a link to claim the merch from our shop. We appreciate, again, those of you who responded and have been here with us for some time.
And it was really incredible to hear a little bit of feedback, again, for what folks want to see from us in year two. We also have just put out our transfer portal. Again, it is the transfer season in science, just like it is in sports. And after our last episode, many of you said you'd be interested in seeing this evolve. So we've put out a first test at ffppod.com backslash transfers. You can see some of the latest high-level transfers happening all across the world into the U.S., out of the U.S., and where are the best and brightest going to build the future of science.
If you would like to donate, you can also donate at the website ffppod.com backslash donate. And so with that, we are going to get back to the quantum, no quantum mania, bad Marvel movie, but to this idea, spin cubits. That's right. And before we get started dragging all the neutral atoms people and everyone else through the mud, which I promise we are going to get to.
Don't you worry. We do need to establish some ground rules, right? Because if we're going to evaluate whether a quantum computing architecture is a legitimate computational platform that can scale, or if it's just a multi-million dollar physics art installation, we need a rigorous scorecard. All right? We need a scorecard. So first, let's kill the hype. And I'm going to just briefly review last episode to catch everyone up on what is a quantum computer and what is it not. A quantum computer is not... A machine that tries every single answer at the same time because a qubit is a zero and a one simultaneously.
No. Okay. A quantum computer is a giant interference machine. There's more details in the last podcast, but effectively the entire game with quantum algorithms, whether it's Shor's algorithm to make Bitcoin go to zero or Grover's or simulating some electronic ground state of some new compound that you've made, you're applying unitary transformations. Okay. So that the wrong answers undergo destructive interference and the right answers undergo constructive interference. And that interference effect is going to tell you, one way or the other, what the answer is.
And a good quantum computer is something that can do all of this. It can store quantum information, one, and then it can manipulate quantum information. It can read it out so I can get an answer out. And it can be very large. It's something that is small now, but might be easy to scale up. Okay? The fundamental unit of a quantum computer is a qubit. We're used to classical bits, which is just a bit. That can be either a 0 or a 1. Usually it's the transistor.
It'll either let current through or it won't. A qubit, you can imagine, is in a superposition of 0 and 1. And really, mathematically... what you can do is say a bit is basically whether I'm on the north pole or the south pole of a sphere. And a qubit is anywhere, any location, latitude and longitude on the sphere. The sphere is called a block sphere. And this represents all of the different states that a qubit can be in. Okay, so crucially, there's two numbers that we have to keep track of.
One is this way. This way is latitude. Yeah, this way is latitude, north-south. That's going to tell you how much it's aligned in zero or one. That's going to tell you if I were to measure the qubit, what's the probability that I'm going to get a zero and what's the probability that I'm going to get a one. If the qubit is at the north pole, then I'm going to get a zero all the time. If the qubit is on the south pole, I'm going to get a one all the time. And then there's also a phase. Right. There's the longitude. And that phase... is how the qubits interact with one another.
It might not have to do with the probability at the very end, but it does have to do with how things get entangled, how they interfere, so on and so forth. And the phase is, even though we're looking at this visual as 0 or 1, basically northern hemisphere, southern hemisphere, the phase still matters for other types of interactions that are going to impact the system, even if it's not determinative of our 0 or 1. Exactly, yeah. So the qubit is that quantum version of a bit. It's some thingy. That's in your quantum computer that can do this.
That can be a two-state system with quantum mechanics governing it. Yep. Okay? And a quantum computer effectively is a set of qubits that can be manipulated using quantum gates. Now, quantum gates we talked about in the last episode. These are unitary transformations that will rotate the qubits and put them together and things like that. They do not destroy information. Yes. Right? They are reversible. Yes. So they're not like the conventional AND gate. You need to have other strings attached in order to get this thing to work. You can go from either the question to the answer or the answer to the question in both ways.
In both ways. A lot of times we talk about AND gates. You can only go from the question to the answer. Exactly. But you can't go back to the question. Exactly. Very good. And for a single qubit... We've got a lot of gates. You can rotate the qubit, as you showed in the block sphere. And here, in this case, this is called a Hadamard gate. This takes a single state, a quantum state, from a 0 or a 1. So it's in a pure state, either in the North Pole or the South Pole. And it puts that block sphere on the equator. Okay? So either you've got the 0 goes to the sum of 0 plus 1, and the 1 goes to 0 minus 1.
Okay? Now... That's for single qubits. You know, that's the Hadamard gate. You can also think of these things as rotations on the Blocksphere, as I was saying, right? In this case, the Hadamard gate rotates along some direction in the Blocksphere. The Z gate rotates along the north-south axis. And then the Hadamard gate, again, rotates along one of the off-axes. So in order to get an X gate For example, an X gate is something where you rotate along the X axis of your sphere.
You can put a bunch of gates together to get an X gate, for example. This is like one of those desk tchotchkes where you have the little sphere thingy and it has the three axes of rotation and you move it around. Yeah, like the toy gyroscopes, if you've ever seen. This is as a visual representation of what the math is doing or what we're talking about as we make these... transformations. Exactly, yeah. And that's what these single qubit gates do. Okay. You also need the qubits to interact, though, right? So you can have two qubit gates.
And you can put a bunch of these two qubit gates together to create a quantum circuit. And the circuit is going to implement a quantum algorithm, like your Shores, like your Grovers, like your Deutsch-Jose, or whatever you want. And here you can see, like, you know, based on the state of one qubit, the block sphere of another qubit will rotate in one direction, you know, and they can get entangled and so on and so forth. All of the quantum magic that happens. The idea is the single gate is like a single instruction in a recipe.
And then when we get to this idea of the entanglement display we just showed, it's like now we're stacking multiple individual instructions. Yeah. To be able to go. And they're interacting. And they're interacting with each other. Exactly. Yeah. And so now let's get into like what the classical qubits are today. Sorry, the classical bits, I should say. Transistors are the classical bits of today. Transistors are the stuff that's in our computer. Right. On our iPhone, in our da-da-da. Da-da-da. The car. Yes. This is why everyone loves Taiwan. Yeah. Right, right. TSMC, shout out. Yeah. Shout out to TSMC.
And this is why geopolitically it's such an important thing for us because TSMC makes all of our transistors, all of our chips. Now, they were not the only classical bits out there, though. Historically... There's been a lot of different types of computers. There's been the imitation game with Alan Turing. He created a computer out of electronic relay switches. It does the same thing that a transistor does. It holds a zero and a one. And based on that, it can make logic. And he defeated the Nazis' code with his giant computer.
Was it Enigma? Yeah, the Enigma code, exactly. And on the right-hand side, we've got John von Neumann, standing in front of his giant computer that's made out of vacuum tubes he's standing next to robert oppenheimer again another type of computer these things were giant though if you remember from the movie they took up an entire warehouse the vacuum tube stuff also took up an entire warehouse and it wasn't even like a megabyte worth of memory right or maybe it was like a megabyte but like a few the basically they were they worked but were inefficient because they required such size and scale that we're not going to be practical to be able to create iphones exactly
Then we got the transistor out of Bell Labs, won the Nobel Prize. And lo and behold, all of these other technologies went into the museums. And what is a transistor made out of again? Oh, silicon. Oh, right. That's right. That's interesting. And I do want to make the, I think I'm seeing the parallel you're drawing here, which is, again, when you have a frontier technology, there are a lot of ways that people think are the best way to go about it. But there's a difference between what works and what works at scale. Yes. Yes. And we had vacuum tubes and electric relays that both did work.
Yeah. No one's saying they didn't work. However... They would not have birthed the internet. They would not have birthed DoorDash, Uber, little robots you have for your kids every day because it could not scale, which is both a performance problem and a size problem, among other things. And so I'm just trying to re-bring up this parallel that I think you're trying to build here, which is, yes, there can be multiple ways to do some functional goal.
Yeah. but it might not be the most efficient or best way to accomplish it if you ultimately want to make this a productized Something. Something, right? Something. You said it, not me. This wasn't even my opinion. I guess we're both drinking the Kool-Aid. The logic is falling in the gate. So with that in mind, right? With that background in mind about how classical computing, how the history of classical computing has unfolded, let's get into some of the history behind quantum computing.
So in the previous episode, we went into Feynman's lecture at MIT where he talked about quantum computing, quantum simulation using a quantum computer. And that sort of started this, it was in 1980, I think. 20 years after that, in those 20 years, people had started thinking about how to make a quantum computer. And there were all these different proposals out there. In 2000, David DiVincenzo, he published his famous five criteria. The paper was titled, The Physical Implementation of Quantum Computation.
He was actually at the IBM Watson Research Center. That'll come back in the future. That'll come back. So IBM has a rich legacy in quantum computation. So he wrote this paper, and he wanted to lay down the baseline physics for what you need to build one of these things. And it was brilliant for its time, mostly because it was designed to politely tell the liquid state, NMR crowd, the nuclear magnetic resonance crowd, hey let's not pretend that your little physics experiment is actually a big step towards quantum computation it was mostly like targeted specifically for these guys who were using NMR to say oh we've got like you know, we can use this as a qubit
In particular, he was skeptical of approaches of this liquid state NMR quantum computing, which in the late 90s, 1990s, it had achieved small logic gates, but highly mixed, not fully pure quantum states. So NMR could perform computation on, let's say, an ensemble of molecules, but it couldn't even initialize a zero or a one. Right? So DiVincenzo is writing there, and he... explicitly includes the terms scalable fiducial which means pure initial state um and these are warning phrases that were pointed at the nmr crowd yeah he's basically saying you don't you don't have the juice yeah yeah uh you can't like hand wave your way out of not having the ability to scale and not having pure initial states exactly
Yeah, and so over 20 years on, the DiVincenzo criteria is still very relevant, but it's 2026 now. And almost every modality can demonstrate a cubit in a vacuum. Not everything can do what DiVincenzo's criteria is doing, but in line of DiVincenzo targeting a specific modality... For the negative, we here at From First Principles have created an FFP audit that is going to also target a specific modality, but in the positive.
Yes. Right? Because we can do that. This is the FFP criteria. Yes. This is DiVincenzo 2.0. FFP criteria, trademark pending. Pending. So don't come at us. We're going to sue you. We've got the legal system on our back. So we're going to think about three criteria about what makes a good quantum computer. Okay? Qubit quality, qubit control, and the scalability and the economics of the thing.
All right. So let's get into what each of these things mean. Yep. So qubit quality. All right. So you've made a qubit. Question is, how hard is it to keep isolated? Yeah. How long does it last? If your qubit doesn't last very long and you need more time to poke it around with your quantum circuit... to get an answer, and by the time the circuit has ended, your qubit has failed, well, you're kind of done, right? Yes. So the gates that you saw in those rotations, those rotations need to happen much faster than the qubit can forget what it is.
The entanglement states that we had seen in the previous. Yeah. The idea again here is efficiency and data fidelity. Mm-hmm. That's exactly right. Yeah. The qubit quality is effectively how efficient is the circuit that you're going to implement in relation to how long your qubit can last, right? For example, if you take a bunch of transistors and you're trying to do AND gates on them, right? If the transistor forgets if it's a zero and a one before you can even implement the AND gate, then the thing you're going to get out of the AND gate is not really going to be the AND of the two.
Right. It's going to be some random zero or one. I'd like to actually do computation. And so the transistor better be stable. Well, in the quantum sense, it's very similar. The decoherence time is what we call it, of how long the qubit can remember itself and not get messed around with all of the outside noise. That needs to be much larger than the gate time. Right? And that brings us to another question, which is how does it compare to your, like, what's the noise sources? That has to do with the qubit quality. What are the noise sources and can you get rid of them?
Exactly. Okay. So that's qubit quality. Then there is qubit control, which is can you initialize, manipulate, and read out the state without your control rack setting on fire? Right? Initialize meaning like state preparation. Mm-hmm. So, you know, preparing all of your qubits in the zero state. Yeah. That's something Anamar couldn't do. Yeah. Which is why DiVincenzo wrote that thing in the first place. It's the pure initial state point. Exactly. Where we see the blue in the left where it's like we all have, we have to start everything at a known base.
Yeah. Like it better, like I better know the states of my system before I start doing quantum logic, right? Otherwise, like what are we doing? Right. Right. And then can I manipulate them with my gates? And then can I read it out? Can I measure what the thing actually is? Right? Now, when it comes to manipulation and gate fidelity, this middle part right here, here's where the noise directly translates into error rates. Because the question is... Can you reliably hit, let's say, two qubit gates, two qubit gates where you take one qubit and another qubit and let's say you want to entangle them or you want to rotate one based on the state of the other?
Like if the other one is near a one, then I rotate. If not, then that's called a controlled knot. Can I just take a moment here? Because I think this is a really key idea to make sure that people are getting right. So we talked about earlier when we have these cubic gates that there's two states. One of them is, again, to simplify it, North Pole, South Pole. And then the other one is like where in that north or it's northern or southern. Where in the longitude. East, west. East, west. How far? We can even just call it that.
Let's just call it that. North, south, east, west. North-south can be just a zero or one. It can just be a functional answer that we can then move on in whatever our process is. But when you just talked about, for example, when we start having two – like two of these states, two of these qubits interacting with each other. The phase is maybe what matters more. The east-west is maybe what matters more than the north-south. Yeah, I mean, sometimes, well, it depends on the gate, really.
But I think your point being that both of these directions matter. Right, right. In terms of what the next one is. Does or does not do. And the ability for those two things to interact successfully in this quantum state is meaningful. Like that's this orange band. Yeah. That's the computation part. Okay. And I just wanted to. That's exactly right. Okay. And the point that I'm trying to make over here is that when we do these gates, they're not going to be 100%.
It's not like classical computing where the transistor, if I wanted to flip that transistor, unless like a cosmic ray comes in or some nonsense, or unless I'm operating the computer at like 200 degrees Celsius, if I want to flip the bit from a 0 to a 1 or back and forth, that thing is going to flip, right? It's extremely reliable because it depends on like bulk electronics. Right? Here, it's depending on single qubits, right? These are very finicky things. And so there's a chance that you try to flip something or you try to rotate something and it just doesn't work because of whatever reason.
For sure. Right? You need to bring that chance down. You need to be able to do this correctly. reliably this is the qubit control point which is the criterion number two because this is a highly um volatile system or or a highly um uh maybe volatile is not the right word but no volatile yeah i would say and so but this is the and i'm trying to nail down this point which is by nature it's not like transistors no where you're always going to get a zero or you're always going to get a one based on some yeah and if you want to flip it it's going to flip
Because of that, the cubic control really does – is like a very important – in this computation layer especially is like a lot of – there's a lot of meat there to actually nail down. Yeah, and you better be good at it. And you better be good at it. Yeah. And you're not going to get perfect. You're never going to get perfect because quantum mechanics is a probabilistic thing. You're living not at absolute zero, blah, blah, blah. Yes. So one of the tricks that people use is they use –
redundancy, effectively. Okay. Where several qubits are performing a computation, and loosely speaking, you do a voting type thing. All of these different qubits are performing a computation, and then you vote on... what the actual thing is. That way, if some people failed, you can still have a sense of what the algorithm actually is, right? This is called error correction. Because there's going to be errors, but there's ways to mitigate for that error by assigning multiple physical qubits to a logical qubit.
A logical qubit is your block sphere. That's the thing that the algorithm is operating on, right? And if you've got multiple physical qubits that are kind of keeping track of this information, loosely speaking, you can form a redundancy. Now, it's a little bit trickier than that because there's this famous thing called the no-cloning theorem in quantum mechanics where you can't clone information, but you can get around it. You're not really cloning information, but you're still keeping track of it on multiple physical... Qubits, physical qubits, meaning whatever, your superconducting circuit, your trapped ion, whatever.
You've got a multiple of these to create a single logical qubit, and the logical qubit is the thing that is doing the algorithm. And so there's almost – now in a system-level case, and I know maybe we're getting a little hell of ourselves. We can come back. There's this idea that there are qubits that have different functional purposes. One might be tracking. Tracking. and one might be doing the actual calculation. Yeah. And you can have some combination of these different types of qubits to get to a system. Because in order to have a system, you need a goalkeeper, you need back four, you need your midfield, you need the guys who are going to score goals.
Exactly. And so qubits sort of take roles accordingly. Very good. Yeah, no, that's a really good analogy. Okay, okay, okay. Yeah. And so that's exactly right. And so when we think about, like, you know, noisy physical qubits versus – the logical qubits that we're trying to emulate in this like logical space. Yeah. The only thing that really matters is fault tolerant quantum computing. This is, this is kind of a big thing these days, which is, you know, suppose there's a, like a bit flip error.
Or there's a phase flip error where it's on east. It's like 90 degrees east, and then whoop, all of a sudden it's like 90 degrees west because, I don't know, some random photon came in from somewhere. Or there was like some charge, blah. Bit flip error, it was in the north axis, and then whoop, went to the south axis. No reason. Right. Just quantum stuff. Yes. Quantum stuff happening, right? If that happens, you better have these error-correcting codes... There's a fancy way of saying that there's some computational machinery that I can implement that is going to keep track of the lost information and keep track of where I'm trying to go with the algorithm.
And with that redundancy, I'm still able to compute whatever algorithm or circuit that I was trying to compute in the first place. Got it. This is not like a no-go deal breaker if one of these errors happens. Makes sense. There's the ability to detect and fix in post. We'll do it in post. Yeah, exactly. So I better be able to do this very well, right? Because every single physical qubit is going to have these types of problems. And this is this idea of fault-tolerant quantum computing. Yeah, so whenever you hear fault-tolerant quantum computing, the fault...
Is this. Is these errors. It's either going north when it should be south or east when it should be west. Yeah. Or it's, yeah. And it can be more complicated than this, right? For sure. It could be like multiple qubits are doing weird things. For sure. Right? For sure. To keep it simple. But to keep it simple, this is like one of the common ones, right? Yeah. It was north, now all of a sudden it's south. Fair enough. But if all of its neighbors are like, yo, you should... We're all north. It's a combinatorial thing. It's like, because, yeah. So it is both at an individual level and then you have to zoom out to then how does that impact the system?
Yeah. And so the complexity can get quite ridiculous. Exactly. Yeah. And one of the key things that matters with these types of fault tolerant computing is what is the connectivity of your qubits? For example, like if you're only, if you're all on a line. I can only talk to the qubit behind me and in front of me. Right. Um, if I'm in a square lattice, then I can talk to like four of my nearest neighbors. Right. But if I'm, all-to-all connected, like some of the ones that we're going to see claim, then I can talk to literally every single one somehow, right?
And then my error code can be that much more efficient. So maybe I don't need that many physical qubits to encode a logical qubit, right? Because I've got larger quorum sensing in some sense. A side note, in biology, we... like biological systems do this all the time. For example, when you're an embryo and like your cells are just a ball of stem cells, right? But then each cell has to decide, I'm going to go be the head. I'm going to go be the tail. I'm going to go be a hand.
I'm going to go be a liver cell, so on and so forth. How does the cell know where it is in the embryo, right? How come I don't have a liver in my skull? Right. Right. How come a cell near my skull didn't decide I'm going to go be a liver? It's because there's this idea called quorum sensing in biology, where in order to figure out where they are spatially in an embryo and who gets to do what, they start aggregating votes. from their neighbors to try to bring that noise down and figure out, okay, so I'm definitely liver, right, guys?
And then everyone around is like, yeah, because I'm going to be spleen. I'm going to be stomach, so you better be liver. It's kind of cool that that's the kind of stuff that I used to study in a little undergrad project, and now, you know... The, you know, physics is connected, even from biophysics to quantum physics. This is why when you're playing on a squad with 11 players, you have to communicate. Because the difference between a team that communicates and doesn't is players that are in there, the right and correct position to execute the tactical approach of the day versus a team that doesn't communicate and then players in the wrong spot.
And we saw that live with Belgium versus USA. We will not talk about it. It was depressing. That was pretty bad. But that's exactly what happened. But that's exactly what happened. And so this is actually very helpful because now we're kind of setting the basis of like, okay, we now – this criteria, we talked about two points so far. We talked about qubit quality, the ability to maintain itself, and then qubit control over those three stages, the initialization, the actual computation, and the measurement.
How well are you able to control? Even if you have good qubit quality, you can have bad qubit control. You can have good qubit control, but bad qubit quality. Exactly. You better be good at both. You better be good at both. And then that's leading us now to the final one, the final piece in our FFP criteria, trademark, which is scalability and economics. Okay. Okay. This is where the, the, this is where the startup pitch decks collide with reality. They collide with the laws of thermodynamics and the laws of economics, which are not real laws, which is why the economics Nobel Prize is not a real Nobel Prize.
Okay. Had to say it. But when someone says we've got 100 cubits today and our roadmap is that we're going to have a million cubits. In five years. Okay. Really? Are you? Yeah. Are you? I don't believe you. Yeah. Yeah. There should be some salt in that. Okay. So first of all, how much cooling are you going to need? That's a big question that you should ask. Okay. Because quantum... Things are finicky. Okay? And you have to talk about the biggest, loudest, and most obnoxious problem in the universe when it comes to maintaining quantum information, and that is heat.
So heat is the amount of jiggle and the amount of entropy—sorry, the amount of energy in any— degree of freedom of a system. For example, the heat in this room, right now the room is about 70 degrees Fahrenheit, so I don't know, let's say like 200, no, like 290 Kelvin above absolute zero. So 290 degrees Celsius above absolute zero. What that means is that the molecules in our room have an average amount of kinetic energy that is proportional to 270
the number multiplied by something called Boltzmann's constant. Okay. And the Boltzmann's constant is literally just, it's like a, a, a tally trick. Yeah. for us to convert from temperature to energy. Okay? If I were to, if a physicist were to invent a new society, we would be measuring temperature and energy with the same scale. There would not be a Kelvin and a Joule. There would just be FFP. That's what we would call it. Which represents it both across both systems.
Because both things are the same thing, really. Temperature is the amount of energy on average in every single degree of freedom. Now... In classical computers, this doesn't matter so much. Okay? Unless you get up to like 100 degrees Celsius, 400 Kelvin. I mean, you know, we did cover a recent paper where it got all the way up to 1,000 Kelvin. Yes. Right? And it was fine. It was fine. But that was a weird memristor made out of graphene and tungsten and all this other stuff. But like our normal computer, like, it's going to fail.
Right? Because there's going to be so much movement. of electrons and so much jiggling of the atoms that it's going to destroy whatever computation is happening. Yep. Right? Now, but that's at like 100 degrees Celsius, 400 Kelvin. A qubit is incredibly delicate, though. Okay? The energy difference between a qubit's zero and its one state is microscopic. We're measuring this stuff in terms of electron volts, which is the amount of energy that it takes to move an electron up one volt, a single electron. Yeah, which is... It's minuscule.
And if the ambient energy of the environment... is larger than the energy gap of your qubit, right? Imagine I've got two states, my zero and my one, which is the qubit. Whatever thingy, whether it's superconducting, and we'll get into what these states are. But imagine I've got a two-state system where the system can be in this spot or it can be in another spot, and the energy difference to jump from one to the other is some amount. But the amount of energy in the environment that's knocking you around is larger than that amount.
Well, then, if I prepared, right, I'd like to prepare my qubit in the zero state, well, pretty soon it's just going to go into a mix of the two. It's going to bounce around between the two, right? It reminds me of, like, imagine, for example, you're in a car. And you've got like one of those beach balls, Earth beach balls in the passenger seat. As opposed to those Mars beach balls. Yeah, yeah, yeah. No, I ain't going to Mars. I'm trying to get Earth beach ball and the North Pole is pointing north in your passenger seat and the South Pole is pointing south in your passenger seat.
If you're on a pristine road in, I don't know, like Switzerland… Right. And there's no curves and you're just going straight. That beach ball, once you prepare it in the zero and one in the North Pole spacing up, right, the Earth looks like the Earth. And I just keep driving because the road is so smooth. The beach ball is going to stay in that. But if I go into a New York City ridden pothole street, which Mom Donnie fixed, good for him, by the way. Good for him. Good for Mom Donnie that he fixed all those potholes.
But before the fixing of the potholes, if I was driving down a New York street, the road would be bumpy. And the bumps would impart an energy into my beach ball that would start turning it. And there's some amount of energy that's required to keep this thing upright. But if I'm bumping enough, then this thing can be in any which way, any which direction. That's the point. So I need to isolate thermal noise from my system. Right. And this goes back to the how good is your cooling?
Because basically you're saying how good is the suspension in your car when you hit a pothole? Exactly. Yeah. Is it really nice where you don't – and get rid of the jiggle for the beach ball? Or are you in a – Are you in a Jeep? No offense to Jeep. The way you feel, where you want to feel every... I mean, maybe that's part of the point of the Jeep. But a Jeep would not make a good qubit controller. Right. A Rolls Royce might or a Bentley might. Because that's what they optimize for, is a smooth luxury riding experience.
Yeah. And so for a lot of these solid state qubits, that means you got to operate at 10 millikelvin above absolute zero, which is colder than outer space. Outer space is at like three Kelvin. This is an order of magnitude, if not more, much colder than outside like outer space and this is where you have to operate the computer so it's like okay you're gonna create this environment at scale to get a million qubits and you're gonna do it at an order of magnitude cooler than deep space yes okay and then the other challenge is in order to operate my qubit i gotta send in electricity or stuff stuff right to like move it around like whether it's lasers or we'll get into that well that better not heat up
the computer. Right. Right. So that's a challenge that is going to affect how well you can scale stuff. Yes. Okay. And if you've ever seen a picture of a quantum computer, you've probably seen like those giant steampunk golden chandeliers. Yes. With the wires coming down and things like that. Yes. All of that is the, well, the infrastructure of a lot of it is the dilution refrigerator. Right. Where you're using helium, a mixture of helium-3 and helium-4, to get down to that base temperature of tens of millikelvin.
Okay? Who knew being a quantum computer architect was similar to being a butcher who needs to keep their beef frozen? Yeah, effectively. Managing refrigeration. Managing refrigeration, dude. I have so many horror stories about dilution refrigerators, and that is for another time, let's just say. But... The point is you got to keep it cold. Yeah. Okay. So that's why we got to freeze these things to that cold. And it better be frozen. Which is non-trivial. That is non-trivial.
Okay. And that's going to come in later. Okay. The other thing is how big is the thing going to be? Yeah. Right. The whole if I want a million cubits, how big is it going to be? That's a question that you need to ask. How much power are you going to need? If you've got a bunch of fridges, you're going to need a bunch of power. You're going to need a bunch of helium. There's not a lot of helium-3 out there. There's a lot of helium-4. There's not a lot of helium-3 out there in the world. And we're not yet bringing it back from the lunar surface to Earth, which is also expensive. Yeah. Yeah, that would also be it. Right? That's not a solution for scalability.
Right? At that point, just put a quantum computer on the moon. Are you serious? We're seeing the same issue with AI where it's like, oh, we want to do, you know, whatever. But it's like, okay, but the power is the limiting factor more than compute is. And it's the same issue. Exactly. And then finally, it would be like, how are you going to manufacture this thing at scale? Yeah. Okay? So those are the three criteria. Qubit quality, qubit control, and then scalability and economics. Okay, and with that, now let's get into our first leading modality.
Okay, this would be our Manchester City, let's say. Okay, so with that in mind, let's start with our first big modality, superconducting qubits. And let's analyze superconducting qubits using the FFP criteria. Superconducting qubits have made a name for themselves. These are the big players, Google, IBM, Rigetti Computing. They had the big quantum supremacy. This is the Willow chip that you see on there, on that hand. So what is the qubit itself? Okay.
A qubit has to be a two-state system that can be in a quantum thingy. Yes, yes. It's effectively a fancy LC circuit, an inductor and a capacitor. From classical electronics, we visited this a lot. An LC circuit is effectively an electronic pendulum of sorts. The inductor gets charged, then it gets discharged. During that time, the capacitor gets charged and then discharged. So you have this back and forth where the energy is moving from the inductor into the capacitor, into the inductor, into the capacitor. Um, and this becomes a harmonic oscillator is what we call it in physics.
When you've got like a oscillating system that just obeys, you know, a sine wave. Which is what we're seeing in this bottom left. Yeah, that's the current. The current is going one way, then it's going the other way, then it's going one way, then it's going the other way. You can also track the voltage. Whatever variable you want to track, it's going to look like a sine wave. It's going to have a harmonic. Okay. Now this is a classical harmonic oscillator. Yes. If I take this harmonic oscillator and I cool it down, I put it inside a dilution refrigerator, then everything becomes quantum.
Okay. Everything is actually quantum at the end of the day. It's just classical. There's enough temperature and there's enough modes that, you know, it obeys classical mechanics. But if you cool it down enough, you're going to start entering quantum mechanics level. And for those who have taken undergraduate quantum mechanics, there's a famous photo in the Griffiths textbook of a cat going up a ladder. These are the ladder states of a quantum harmonic oscillator, where each rung of the ladder is a different state that your quantum harmonic oscillator can be in. Crucially, the rungs of the ladder are equally spaced.
And the spacing is h-bar omega. H-bar is Planck's constant divided by 2 pi. Omega is the resonant frequency of your harmonic oscillator. The problem here is that all of the rungs are equally spaced. Okay. Okay? One of the questions with how good is your qubit, one of the questions that comes up with that question, how good is your qubit, is it really a two-state system that's isolated? Or can I accidentally go and access some other spot?
Is there noise in the system? Yeah, is there noise such that I leave my computational basis, is what they call it. I've got a 0 and a 1, and I want to stay within the 0 and 1. I don't want to go to 2 or 3, right? Because qubits, bit 2. But if there's equal rungs on the ladder, and let's say I poke it with enough energy to go from 0 to 1, I could also poke it with enough energy to go from zero to two because the spacing is equal. Or if I'm at one and I want to get down to zero, I could poke it.
It could go to two. And this comes to the idea of we have these systems at this very low temperature. If it was at a higher temperature, for example, would that be the equivalent of this poking where it could go from a zero to a two? Yeah. That's one way that it could do it. practically do this yes one way but actually what what's worse is even if that's even even if you're at a low enough temperature there's something called stimulated emission of radiation that's the ser in laser okay um and so you can literally go from like one to two even though you wanted to go from one to zero and there's no temperature effect okay if the if the states are equally uh
the rungs are equally spaced, right? The photon could just be like, oh, I'm just going to absorb this instead of absorbing and then emitting two. And now I'm at state two. So you got to do some finickiness to remove that equalness in the ladder. Because the point there is the equalness allows the easy transition from these states, and you basically want to make it so that the zero and one and one and zero is equal. But every other state transition is harder. Yeah, it's like not equal. It's like something different.
Yeah, something different. It's something different. So that I can very precisely control my transitions from 0 to 1 and back. But I can also very precisely say that I'm not going to go elsewhere. Yeah, that makes sense. So in order to fix that... They replace the inductor with a Josephson junction back to our Josephson junction. That's right. So last year we had a great episode. about the Nobel Prize winners in physics, Michel DeVere, John Clark, and John Martinez, they showed for the first time that you could have macroscopic quantum tunneling in a Josephson Junction at Berkeley.
It was a really good episode, and I encourage people to watch it. This sort of started that idea of using a Josephson Junction, cooling it down, and using that as part of your qubit. So now, if we replace a Josephson Junction in place of the inductor, what happens? Well, instead of a perfect quantum harmonic oscillator, which is on the left, that's a parabola. That's the potential of a parabola. And on the right-hand side, instead of a parabola, we introduce a cosine.
It's the bottom of a cosine. Now, the bottom of a cosine, crucially, kind of looks like a parabola, but the farther out you get, it diverges from a parabola. Right. What that means is your 0 to 1 has a certain spacing, but the other ones have different spacing. And it continues to diverge farther away from the parabola. Yeah, yeah, yeah. In a parabola, it would be exactly spaced. Right, right. Just because of how the math works out. Right, right, right, right. But with a cosine, now you've got different spacings, and now I can very exactly, hopefully...
toggle between 0 and 1, and I don't have to worry about going into 2 and things like that. So this would be a great qubit, right? If exactly I could always do 0 to 1 and so on and so forth. Fair enough. So that's what we're doing. We're going to replace the inductor with a Josephson junction. And this is what it looks like in practice. So this is called an X-mon. These qubits are called trans-mons, the ones that Google and IBM uses at least. And under a microscope, this is what it looks like. So on the left-hand side, we've got like a cross.
The big cross is the giant capacitor. Okay? And zooming in there, there's the Josephson junction. And effectively, this is your qubit. There's a little circuit that runs inside. It's superconducting, which means that if you let it run, it's just going to keep running, which is nice, because you want your qubit to sort of stay qubit-y. And... That interference device, the SQUID, that's called a superconducting quantum interference device, it's creating the 0 and 1.
This entire ensemble is creating your 0 and 1. Now, how do you talk to it? This is my qubit. This is the substrate. This can hold my 0, which is one state of the circuit, and the 1, which is another state of the circuit, the sort of a little bit higher frequency. Okay. Or, yeah, higher frequency. And would you say that this is speaking to the qubit quality variable in the criterion? Yeah, yeah. I'm trying to define the qubit itself. Right, right. So, like, by defining it, now we can then speak to this.
We can then judge it against the criterion. Yeah, yeah, yeah. We understand its structure and what it's actually doing. Made of. Made of. Yeah. And what the zero and the one state is. Yeah. So then we can begin to, as we go through this process, ask the questions about quality control. Exactly. And scalability. Yeah. And the one more thing before we start judging with the criteria is I'd like to talk about how do you actually talk to the thing? Yes. Right? Once we define the qubit and how we talk about it or how we talk to it, then we can get into the criteria.
So that's going to be the format of all of these sort of audits, so to speak. So how do we talk to the thing? We use microwaves in the gigahertz range. So here we've got a chip. That has four Transmons qubits. Those are on the bottom there, those four. They are connected to a driveline on the bottom. Those are little lines that send in microwave pulses to change your qubit from a zero to a one. And then you see the top squiggles.
Those are your readout resonators. Okay? And if the qubit is in one state or the other, it's going to resonate with a microwave that's inside that. You can imagine like, you know in fiber optics, fiber optics like carry light through it, right? This is a fancy mini fiber optic thingy. Okay? That's going to hold a microwave inside. And if the microwave is exactly the right frequency, it's going to resonate with each of these little squiggly waveguides. Okay. And so basically where we read is in these readout resonators.
Yeah. If the qubit is in one state or the other, the readout resonator is going to resonate. And then my line up top is going to go back up to my electronics and tell me. What state is each of the four in? Because we have four qubits. And so we want, and then we have each of these resonator readout, like basically lines that connect to our piece that's going to send it back up to us. So we can independently read each of the four qubits. And so this is a four qubit. It's a transmon computer, a four qubit superconducting circuits computer.
And the way they talk to each other is just through cross capacitance, meaning like if there's a circuit over here and a circuit over here, they're going to affect each other using electric fields. Just by proximity. Just by proximity, straight up. And I mean, so does that kind of make sense? The driveline sort of tells you how to poke it. The qubits are in the middle. They can be in either a zero or a one. And the resonator that's up top is going to let you read what the state it's in. So there's my, you know, initialize, manipulate, readout.
Yes. Okay? It's just going from bottom to top. Makes total sense. Okay. Now, we know about how the quantum computer works. Now let's do the audit, right? How does it hold up to the FFP criteria? Well, what are the strengths? The strengths are that... superconducting qubits have very fast operations, 10 to 30 nanoseconds. You know, those rotations on the block sphere that I was showing earlier, those gates, they can happen within 10 to 30 nanoseconds.
That's fast. Okay. That's very fast. And crucially, if you have two qubit gates, it's maybe a little bit longer, like 60 nanoseconds. But, The coherence time, how long a qubit remembers itself, is quite long. Yeah, that's nice. Okay? It's nice. We actually covered a paper by Princeton earlier in this podcast season where they described a qubit with one millisecond. of coherence time which is fantastic which is fantastic compared to nanoseconds yes if you're doing tens of nanoseconds to to like move stuff around if the thing can remember for a whole millisecond that's there's 10 to the 6 nanoseconds there's a million nanoseconds in a millisecond just to give you the like you've got a lot of time to poke around with it right and basically make sure you like can read what's happening right like the the coherence time effectively is how long do
And manipulate. And manipulate before the system collapses and you have to start again. Yes, exactly. And that Princeton paper used tantalum, which is – I always find it hilarious to like, you know, in high school when we learned about the periodic table – There were all these elements that we were just like, who uses tantalum? But yeah, in quantum computing industry, there's so many exotic materials like tantalum. We're going to get into ytterbium later, which is, you know, I didn't think, when I was in high school, I was like, why would I need to know about ytterbium?
There's good reasons for it, right? So, okay, that's a strength. The gate speeds are really fast. Yeah. The coherence time is pretty long compared to the gate speeds. So you can implement an algorithm pretty quickly. What are the negatives? Well, one, remember I told you about that that anharmonicity, meaning it's drifting away from harmonic. Harmonic meaning parabola. But I've introduced this cosine term that sort of gets rid of that degeneracy in the energy spacing.
So only 0 and 1 is a certain energy spacing. The other ones are not that energy spacing. So when I want to talk to 0 and 1, I send a microwave pulse that is exactly that energy, and I can toggle between 0 and 1. And this is the ladder rung distancing. And it's like you want to be able to know which rungs of the ladder you're on. And that's why you want there to be a difference between the different distance between 0 and 1 and others. And others. However. Yeah, there's a problem. Okay. There's a tiny problem, which is as I'm the gates, like how I manipulate this qubit.
Depends on microwaves getting sent in, right? Now, if I could send in a pure tone, right? Like, you know, those tuning forks that have a pure tone. If I could send in a pure tone at exactly that frequency, that's the difference between the zero and a one, then I'd be fine. But a pure tone necessarily means a very long time to send that frequency. Right? Yes. Now, as I start squishing the frequency, right, I start doing a beep or a boop or a boop.
That's a very... Mr. Acapella, everybody. But notice over there, right? What I did was I was trying to access different notes in sound, but I was trying to make it very short. Yeah. Now, if you were to take that microphone, that sound readout, and then you were to ask some computer algorithm, what are the frequencies in when Krishna did beep versus boop? There's going to be the main frequency, which is the note that I was trying to get to, but there's also going to be off frequencies.
Okay? There's going to be other frequencies in there because I'm trying to squish all of the notes into a very short time scale. This is actually straight up Heisenberg uncertainty principle. There is a tradeoff between your accuracy and frequency. and your accuracy in time. So if I make the time window smaller, the frequency bandwidth gets larger. That's a problem. Because now, if I'm trying to only toggle between 0 and 1, but I'm sending these really short pulses, there's a chance that I toggle the other frequency.
Like, there's a chance that some of the other frequencies have made it in into that short pulse. And so now I might be accessing the other states. Yes, this tracks. The idea is because we need to communicate at a very fast rate because of other limitations of the system, the accuracy by which we can read between the zero and one. it kind of gets fuzzy. It necessarily needs to get fuzzy because we're trying to communicate so quickly. Yes. So it's like, this is like, there's a trade-off. Right. It's one or the other.
Yeah. You can't have both. Yeah. You can't have very fast and then high fidelity, like understanding of the frequency that you're sitting. Like it's, if you do it very fast, Yeah. Then, yeah. The frequency is a little fuzzy. Yeah. Yeah. And so you need to control it really well. And just to show how much you want to control it, right? The thing, the difference between these states, the zero and the one, is at a gigahertz range. Okay. Okay. That's 10 to the nine. Yeah. But the difference between, so the difference between zero and one is a gigahertz. The difference between one and two is also in that gigahertz range.
It's different from the first one by only like hundreds of megahertz. Mm-hmm. Right? Yeah, yeah, yeah. So the two rungs of the ladder are not all that different. Yeah, yeah. In terms of how we're able to actually read the difference based on what we just talked about. Exactly. So that's a problem, right? And the other thing is something called microscopic defect coupling. Effectively, there are these two-level systems in any interface, and this is a problem in sort of any solid-state electronics. Whenever you have interfaces, like, for example, the Josephson Junction.
Let's say the Josephson Junction is made out of aluminum with some aluminum oxide in the middle and then aluminum. Now, the aluminum and oxygen, they form these bonds, but those bonds can be maybe in one of two states. Like, that's the red circle and the pink circle. If the energy between those two states... is about the same as your qubits, 0 and 1, then when I'm trying to talk to the qubit, sometimes instead of talking to the qubit, I will talk to this bond. And the bond will toggle between one bond and the other.
And then it's like, ah. So it's like you end up reading the substrate rather than the system. Yeah, yeah. I'm trying to talk to this one thing. But as I send my microwave, the microwave is going to spread out because microwaves have a large wavelength. They're going to spread out. And maybe it'll poke this other thing. And it just happens to be in the right state and phase to send a response back. And so the point here being, from cubic quality, we have gate speed is great, but the...
inability to distinguish between zero and one in other states. Yeah. And it might be interacting with your substrate. Yeah, yeah, some other stuff. It's problematic. It's problematic, right? Okay. So now that was cubic quality. Now let's talk about control. Now the strength is that you've got direct microwave interfacing. Okay, the energy splittings are firmly in the microwave domain. And that means that control and readout can leverage modern telecom and radio frequency equipment. Right? We're really good at radar.
We're really good at radio and things like that. The negatives, though... One is planar connectivity. The transmons rely on nearest neighbor 2D coupling. So the types of error codes that you can kind of implement here are limited by the connectivity of your chip. Meaning only certain types of connectivity can we really effectively use this for. And this is honestly like, I mean, it kind of makes sense. This is something that a lot of things have to deal with. That's totally fine.
The one that I want to kind of focus on is crosstalk and frequency crowding. Okay, here's the photo again of our four-qubit computer. I want you to notice something. You see those resonant readout cavities, the squiggles? They're all different. You see them? Yeah. The one on the left is like taller, like it's like thinner. Yeah. And the one on the right is larger.
Yeah, yeah, yeah. Right? Yeah. Okay. There's a very good reason for that. The reason is, suppose I send in a microwave at a certain frequency, at a certain energy difference to the one on the left. Yeah. To make sure that that microwave doesn't bleed out and talk to the other ones, each of the microwaves, each of the transmon qubits need to have a unique resonant frequency.
Does that make sense? It does. No, it does. Right? Because if I want to talk to this guy with one language, the other ones better not be able to understand me. It's like a walkie-talkie with different channels. Exactly. You need different channels, otherwise everyone's going to... Yeah. Or straight up the radio, right? There's a reason why 89.9 is KPCC and 91.5 is KUSC for classical. Right. For those who live in Los Angeles, right? And I think there's a 0.2 megahertz... like gap between all of our radio stations right because when i tune to one i better hear the one that i want to listen to yeah right and not the other ones and so you need them to have they're not trying to all listen to yeah and there's the problem with like am because am can bounce from the atmosphere sometimes when you like drive out you know you'll get like these you'll get like the the talk radio from sacramento and also and so and so
Depending on where you are in the mountain, you'll switch between someone talking in Sacramento or someone talking in Los Angeles, right? Yes. But that's the idea. That makes sense. Now, this creates a challenge because you've got a certain bandwidth where you can put all of your unique frequencies, right? Like for radio, for example, I think it goes from what? Let's say 87 to 106, right? That's a numbers game. Yeah.
That I can put. I can't put more. Because you need at least that two-meganets gap. Yeah. So this idea is called frequency crowding. Crowding and crosstalk. I get you. Now, there's ways to fix it. For example, you could have like a flux biasing that you pump some voltage into each of the qubits and then that raises or lowers the... the resonant frequency, but then that introduces like one over F noise and all sorts, because you're not introducing more electronics into the system. Right. So there's ways around it. I'm just saying this is like a kind of a mathematical thing that you need to worry about.
Yeah. Right. Okay. Which foreshadowing some other systems might not have to, might not have to. Right. So finally, let's get into the scalability and electronics. Yep. There's a wiring problem. This is the Google willow. computer, all of those lines that you're seeing, those are coaxial cables that bring the microwaves down to the qubit. Okay? That's not the refrigeration. The refrigeration is mostly in the metal. All of those lines are microwave lines. Oh, that's interesting.
Okay? Okay. That's the microwaves that are talking to your qubits. Yeah. And a lot of these lines are fat. They're semi-rigid coaxial cables made out of stainless steel or superconducting niobium. And they deliver some resonant frequency to these individual qubits. And this is the stuff that goes all the way down to the qubits. So there's a wiring problem in terms of there's a bunch of wires, and I don't know how many more wires I can fit into this thing. Yeah. Okay?
And we live in a wireless world. No, I'm kidding. Effectively. Now, the other problem is the qubit itself is at tens of millikelvin. Yeah. Yeah. Right? Now, if... And this is what a dilution refrigerator looks like. A dilution refrigerator you can think of as a Russian nesting doll of big, and then you use the big as a heat dump for... At the top. At the top. At the top, you have like a 55 Kelvin, right? I mean, at the top, I guess you have room temperature.
That's where the heat is getting dumped. You use the room temperature to dump heat and get down to 55 Kelvin. Right? And then you use another stage to get down to 4 Kelvin to dump heat to the 55 Kelvin, which dumps heat back to room. So you have this Russian nesting doll of like cooling stages, right? All the way down at the very end is the 10 millikelvin stage, which is where your qubits sit. But all of your control lines, the microwaves and everything, have to go all the way down to 10 millikelvin in order to talk to your qubits.
That means you're heating up that 10 millikelvin stage with a bunch of microwaves. Yeah. And there was a lot of them. And there was a lot of them. So the question is, how many can I get in this? There's some limiting capacity based on the need to be able to control the levels of heat at some very, very low level. Yes. And so there's some upper limit, theoretically. Exactly. Like how many microwave lines can you actually put there before it becomes... there's no room there's no room there's no room and there's no more cooling capacity right right um and the other thing is each of those transpons that you saw earlier they're at the scale of a millimeter one millimeter it's actually quite big it's quite large it's quite big um if you're trying to fit a million of these things it's not going to fit in a single dilution refrigerator yeah okay so one of the so there's there's there's a bunch of problems here um the cubit itself is too big there's too many wires that are going down um
And there's a frequency crowding, right? For a single sort of chip, there's not that many that I can do mathematically even. So in each dilution refrigerator, I can't actually fit that many qubits. What I'd have to do is get a bunch of dilution refrigerators and rig them up. Like a server rack where you have multiple individual servers into this rack system. Exactly. And actually, IBM is kind of working on that. Okay. They've made a modular dilution refrigerator.
This came out very, very recently. Okay. Okay. where they're showing that you can take a bunch of single dilution refrigerators, fit everything inside of it, and then these quantum fridges, they're 200 times colder than deep space, but that's just every dilution refrigerator, so that's part of that headline, and could pave way for fault-tolerant quantum computing, because you've got a bunch of these, you connect them together. Now the communication between one dilution refrigerator and the next better be very, very good. That's another layer. That's another layer that I have to worry about. Because now you're creating this individual component now in a larger system that was already its own system.
Yeah. And effectively, if I rig a bunch of these dilution refrigerators up, I'm going to need like a data center. type of warehouse. Yeah. Yeah. With a bunch of dilution refrigerators all rigged up. Yeah. And we already can't get data centers. Right. We already hate data centers. Right. Right. And we're using that as a, as a, as a structure and architectural and structural point. Yes. It's not literally a data center. No, no. It's a quantum, I guess it's a quantum. I don't know if you, but the point is you're talking about the scale and the size of power, water, water, physical space that's necessary to actually have this helium three where are you going to get the helium three i will say it looks really nice it does the marketing photo looks great right and like just for one of these there's going to be a lot of helium three right and now you want to scale this to multiple right where are you going there's questions there's resource constraints on the architecture there's questions that need to be answered yes okay and so now um and you can do these estimates other people have done these estimates for a thousand plus qubits um and
For a thousand plus logical qubits. Which is not the million we talked about earlier. No. It's just a thousand. A thousand logical qubits, you need ten million physical qubits of these transplants. Okay. And you need like ten million dollars, right? For a million, you're talking like billions of dollars for a single quantum computer. Yeah. Yeah, that's tough. It's tough, it's tough. Billions of dollars for a single computer? It's not gonna, it's not, right? And the footprint is the size of a warehouse, um...
Yeah. So let's look at the final verdict. Final verdict. The FFP criteria for superconducting circuits. Cupid quality, I'm going to say, is 5 out of 10. Cupid control is 5 out of 10. Scalability and economics, 1 out of 10. For a total of 11 out of 30. The quantitative criteria about which we judge this is, of course, proprietary. Proprietary. Proprietary FFP IP. If you claim that I made these numbers up, uh i will sue you i guess that's that's what uh you are making claims against our proprietary uh trade secrets yeah yeah yeah if you if you claim that i mean you'll see how these numbers get when we go to a neutral neutral atoms let's just say that um but 11 out of 30 not not great really struggled in the scalability and economics category which goes back to
This whole point we talked about earlier about electric relays and vacuum tubes, yes, they could operate as transistors. However, if you want an iPhone, you cannot have an iPhone made of vacuum tubes. And ultimately, what is going to matter in these spaces because of the way we engage with technology in general and the use cases for how people want to use this stuff is, is that the economics and the scalability matters just as much as the fundamentals, which is where a lot of the energy has been put so far.
And then the problem is you go down a route, and then you've invested so much in that route that then you get this inertia of like, well... we have to commit to this. So we have to make, we have to like make the economics fit into this path. We've already spent billions of dollars on. And this is where you get a lot of the tension in these corporate environments of like, well, we can't like pivot. Yeah. And like, to be clear, I mean, we're not trying to make an iPhone out of quantum computers, but we're trying to have a lot of quantum computers. And if a single quantum computer is billions of dollars, that's not great.
We, we, where's the helium three going to come from? Yeah. Yeah. Yeah. Yeah. Exactly. I mean, it's just like, there's just like functional. Yeah. Yeah. Even if you just had one. Yeah. And like you can't make this in a TSMC, for example. Right. And things like that. So, okay. So that was... Like it's been great for the quantum computing industry because it's been able to get to like this intermediate scale quantum computer where they can show quantum supremacy and things like that. I just don't think it's the future. That's fair. Superconducting qubits, 11 out of 30.
So next up, trapped ions. Quantinuum just had its IPO, $15 billion. Quantinuum? Yeah, Quantinuum. I love that. It debuted on the NASDAQ. Billion dollar IPO. Yeah. Another one is IonQ that's based out of Maryland. These two companies do trapped ions. Lots of UCLA physics people actually are involved with Quantinuum. Their chief quantum architect is Anthony Ransford, who told me himself that he's a fan of the podcast.
Shout out, Anthony. Shout out Bruins. Bruins Nation. Yeah, yeah. But you still got to go through the audit. Hopefully you still agree to call on the podcast. We shall see. So what is the qubit? Okay. Now, this is not a man-made circuit. They are using an actual literal atom. All right. Usually it's something like uterbium or barium. What they do, in this case, it's barium, right? Barium's got these two electrons on its outer shell. You remove one of them, then it becomes a positive ion because now there's more protons in the nucleus than there are electrons around it.
And now the whole thing has a positive electrical charge. Now, because of that charge, you can now move it around using electromagnetic fields. There's a key way to do this, though. You can't use a static electric field. You can't just put a bunch of, like, electrodes at certain voltages and trap them because the ion is going to figure out a way to just, like, shoot out. So instead, you have to oscillate the electric fields. You can find in one direction and these two other directions, you, like, you know, you go positive, negative, and then you switch from negative, positive.
And this back and forth... is sort of massaging the ions to stay in a certain spot. And that is how you confine a bunch of ions in a single location. This is called a pall trap. And this is basically how we're trying to create the qubit states. Yes. Well, first you've got to localize it. Right, right, right. With the transmons, it's like, oh, it's on a circuit. It's like right there. Right. Right. But with atoms now I got to put them, I got to keep them in a spot. And so the way you keep them in a spot is using this oscillating electromagnetic field that like oscillates at a certain frequency.
It's funny because before we're talking about oscillation from like the microwaves and how it's coming back to that same idea. You have to maintain the state via oscillating somehow. Somehow. Yeah. Harmonic oscillators are everywhere. In physics, to be honest. So that's how you keep the ion in a certain spot. The qubit states, the zero and the one, these are two very specific stable energy levels of the atom's outermost electron. We've seen this photo before of the hydrogen hyperfine transition.
This was sent on the Pioneer probes and the Voyager probes where the electron on the outside of the hydrogen is, is either parallel to the spin of the proton or it's anti-parallel to the spin of the proton. And the difference between those two is 1420 megahertz. And that set the time on the pioneer record that we sent out for the aliens if they wanted to decipher where we're from and how to locate Earth and things like that. This is called a hyperfine transition. It's when the electron is parallel to the nuclear spin versus opposite the nuclear spin.
The trapped ion people, at least in continuum, they're using these two as their zero and one. Okay. Okay? That's their zero and one. And the way you interact between them is using lasers. Laser beam. Right? And the energy scale between these guys is about 12 gigahertz. So it's in the microwave frequencies. Okay. And that temperature is equivalent to about 600 millikelvin. Okay. Okay. Okay, so the temperature difference or the energy difference between your zero and one in this case is less than a Kelvin.
Okay. It's larger than the superconducting but still less than a Kelvin. Yeah. My point is here, this stuff still operates at room temperature though. The question is why. It's because you suspend it in a vacuum. If you put it in a vacuum, then nothing is interacting with it. There's no atoms that are jiggling around that are poking it, hopefully, if the vacuum is good enough. And all of the electrons—I mean, sorry, all of the photons—because remember, at any given temperature— There's not just degrees of freedom with the atoms that are moving around, but also there's degrees of freedom with photons and the electromagnetic field.
And so in this room itself, for example, in our body, there are photons that are jiggling around at around infrared because we are at about 300 Kelvin. Now, there's going to be a bunch of photons that are poking this ion thing, right, at 300 Kelvin. It just so happens, though, that a lot of those photons are in the infrared. And those infrared photons are very different from the energy gap between the zero and the one. So they just sort of... Coast through. Coming back to our rungs of the ladder.
They're not at the range that matters for us in what we're looking at. Exactly. The rungs of the ladder. Exactly. Yeah. That was a question that I had when I was researching this. It's like, okay, fine. It's a vacuum, but you have an ambient heat bath of photons. How come that doesn't destroy you? Well, it's because these guys just aren't sensitive to that. The rungs of the ladder are very different from the poking that's happening, right? Mm-hmm. Okay, so how do you talk to them? Well, you use lasers. You line up a chain of these glowing ions in a trap so that you've got chains of ions.
Single qubit gates are done using precisely tuned laser beams on the individual ions. And then two qubit gates are where it gets kind of wild. Okay. Okay? So two qubit gates are done using effectively sound waves in your ion mesh. Okay. Okay? These things are charged, right? Which means if I move one guy, it's positively charged, so it's going to repel everything else. Because we've removed the two outer shell. One, just one of them. One of the outer shell electrons. So it's just a barium plus. Okay.
And we use the remaining outer shell as the hyperfine, right? Got it. But if I move one of these barium atoms, that's going to cause a Coulomb repulsion, an electrostatic repulsion on the other one. Because, you know, like charges... Do not attract. Repel. That's the word. And so if I move one of these guys, that's going to move one of these guys. That's going to move one of these guys, right? And those phonon modes, those sound waves, are how you start entangling and doing two-qubit gates. It's kind of interesting, right?
That's interesting. That's interesting. Yeah. Yeah. I thought it was pretty cool. Because now it's just another means of – is this now – I'm trying to think about it. This is how we measure. This is the measure. No, this is how we do two-cubit gates. So how we entangle them, how we get them to talk to one another is using sound waves. We read them out and measure them using the lasers again. Okay. Got it. It's effectively imaging. got it it's like you you poke it with and you try to see is it in a this state or this state that's interesting but the the the way in which we're having them entangle or interact is via sound with sound waves that's interesting that's cool yeah that's cool that's quite nice yeah it's called the cool points mulmur sorenson gate okay okay okay um and and that's it's the jiggling
of the sound waves between these ions that's actually doing this. So, okay. How does this hold up with the FFP criteria? And this is a chip that shows right in the center. You've got this line. Yeah, yeah, yeah. Right in the middle. Okay, okay, okay. Okay, so how does it hold up with the FFP criteria? The strength... This is mother nature's ultimate qubit in some sense, because every single qubit is identical. Before with the superconducting thing, you know, you got to manufacture it. Each thing is going to be different kind of by design because you want the resonant frequencies to be different.
Here, everything is exactly equal. And as long as I can point my laser accurately, I can be like, okay, talk to this guy. Now talk to this guy. and so on and so forth, right? So every single qubit is actually the same. There's no, like, two-level system causing headaches that are in the background. The other thing is the coherence time is pretty ridiculous on these guys. How long, again, coherence time being how long it remembers... the zero and one so you can then yeah chain manipulate or read yeah and i ultimately mess with mess with it ultimately we want as long of a coherence time as possible yeah this thing can get to 10 hours okay that's right so i i take an ion yeah i put it in the zero state i come back several hours later and it'll still be in the zero state that's that's quite nice that's quite nice that's quite nice right okay
No, that's quite nice. That's a big deal. Because that's a totally different category than what we were talking about with superconducting qubits. Yeah. In terms of coherence time. Yeah, yeah. They were happy with a millisecond. Right. We're just in... Right. Here we're doing hours. Okay. But... Okay. There's a caveat. Okay. With the superconducting, a millisecond was great because the gate times were nanoseconds. Yeah, right. Right? Here, the gates are really, really slow because we're using sound. Yeah, and sound waves are... Sound waves are kind of slow.
And there's actually an upper limit to how fast you can do these gates. And it has to do with how fast the pall traps are going. Like, you know the thing I was showing you earlier with the electromagnetic fields, like kind of maintaining these ions in a geographic position? Those things are creating like a sort of bulk That is like, it's really a saddle, you know, those saddles. It's like creating a saddle that is rotating around. The rotation rate of that saddle is kind of like an upper limit on how fast my gates can be.
Because the ions are moving around in that saddle, right? If you try to like, it's kind of like, imagine if you're like two people on a swing, right? And you're swinging back and forth. That swinging back and forth, let's say, is the electromagnetic saddle that is keeping you there. But you're trying to communicate using the beam that is connecting you in the playground. You're trying to communicate with the person next to you based on how you can vibrate the beam above you that is holding you together. There's going to be a limit to how fast you can make it.
And part of then what this means is because that gate time is now slower. Yeah. Yeah. Even though our coherence time is very long, you can just do less. Yeah. Right? Yeah. Yeah. Like the coherence time is way longer, but you can do less with the same amount of just physical time in the lab. Right. Right. Right. Right. Right. So just because you would ideally want a longer coherence time and a really short gate time. Yes. And so we're like, yeah, we have long coherence time, but we're now having a much longer from as compared to superconducting qubits, longer gate time.
So you can just basically cycle to do stuff less frequently. Is that... Yeah, yeah, yeah. So it's like the scale of the quantum circuit that you're trying to implement might actually be the same. Right, yeah. Because both of the things have scaled. Right, exactly, exactly, exactly. Okay. Yeah, yeah, yeah. So that's one of the caveats, right? Okay. Now, what about the control? So we did cubit quality and sort of gate quality. Now, the control, there is a strength here. You can get all-to-all connectivity. Okay? You don't have to do nearest neighbor.
Because what you can do is move around the ion so that any ion can talk to any other ion. Right? And for Helios from Quantinium, Helios is their latest processor. He's also out in Nature. They had a nice paper out in Nature. Not on the cover, though. That was me. That was Lester. That was me, Anthony. That was Lester. Anthony's first author on this. Congratulations. Congratulations. So anyways, this is what the cubit chip looks like. So on the right-hand side, it's printed out.
You can see there's like a ring. And then there's like a sort of two trains that are going into the ring. These are where all the ions sit. Okay, the ring is kind of like ring storage. And the... And that's your memory. The ring is your memory. And then on the right-hand side where you do the logic is your processor. It looks very von Neumann-y. Yeah, yeah. You know, von Neumann architecture. You've got the separation between memory and processing. Ring and processor. So I thought that was kind of cool. That is nice. I just think, yeah, the implementation of like...
there being a separate storage for processing and a separate storage for memory. And you can have this kind of memory because your coherence times are so long. You can mess with stuff and then just move it into the memory and it'll kind of just remember where it was. And what you can do is move stuff around so that there's an interaction zone Where you can make the two interact. You can put any two ions right next to each other. And that's where they get their all-to-all connectivity. Now, they out here saying that it's all-to-all connectivity, comma, two at a time.
Because it's really, you got to put two next to each other, and then you interact those two. It's just you can choose any two. It is all to all connectivity, but it's two at a time. But it's still pretty good. So I thought that was really cool. And then this is the quantum logic, the processing part of their paper. One other really cool thing that I thought of, actually, when I was visiting this. So they use barium as their computational qubit, right? The barium ion, the hyperfine, so the outer electron being either aligned or not aligned.
That's your zero and one. But if you want to cool this thing down, you still need to cool it down, right? Like you still need it to not jiggle around and access. Like the ion itself is going to have other states, just like how the superconducting qubit had all these twos, threes, and fours. The ion itself, the electron could do random other nonsense, right? So there's other states that you don't want it to get to, which means you've got to cool even the ion down. But you don't want to cool it down by poking it. Which is independent of it being in a vacuum. Yeah, yeah, yeah.
That's an independent thing. If it's in a vacuum, that means no atom is going to bump into it. But there's still, as I said, these photons and things going around, and you don't want it to access all these other spots. Yep. And the electromagnetic trap is also moving it around, right? The little rotating saddle is also moving it around. So you've got to be able to cool this thing. And when you're moving it around in this ring, in the processor, right, you're physically moving these ions around. That's going to introduce heat into the system. So you need a way to cool it down. Okay. But you don't want to poke the barium ion because that would destroy the quantum information.
So every single barium ion has a partner, uterbium ion, right next to it. And what you do is you cool the uterbium ion with a different laser at a different frequency, right? And then because the uterbium ion is cooling... And the barium ion is like interacting with the uterbium ion. They're all in the same trap. The barium ion is also going to get cooled, but it's not going to lose its quantum information. I thought that was kind of cool. And that's why you see the circle. There's like a small circle and a bigger circle. The bigger circle is the uterbium ion that we're using to cool down the barium.
I thought that was a cool trick. It's like a conduit. It's basically it becomes this companion that offloads the cooling function that gets passed on to the barium, in such a way that allows the barium to retain its quantum information. Yes. But it's always paired. The barium's always paired with the uterine in order to be able to, in the quantum logic part of the system, manage... Everywhere. Even in the memory. Even in the memory part. To basically manage its cooling...
aspect to it, which is quite cool. As a small side note, this is maybe unrelated. We've talked about the von Neumann bottleneck when you separate your memory from your processor. Is there a similar problem in this type of system because we're separating the two? Where you get some inefficiency because you have to translate. I mean, I think kind of, yes. Like, you know what I'm trying to ask? I think totally. Because, I mean, you have to move stuff around in order to get into the processing part, right?
And so this is the whole all-to-all connectivity two at a time. Two at a time. Right? Because in the processor part, you can only have these things two at a time. And then you're applying these logic gates using your lasers, right? Yeah, yeah. And so in part, the architect... But you can do this with trapped ions because the coherence time is long enough that you can move this stuff around and kind of retain that information. That makes sense. No, that makes sense. Okay. Okay. Yes. Yes. Yeah. Okay. So I thought that was kind of cool. We've talked about quality. uh very high yeah the quality is very high the gate times might be a little too long um control is pretty cool but again you have the slowness associated with moving stuff around but some cleverness in implementation scalability and economics scalability and economics is next this is where i think the architecture kind of hits the wall okay um
What they did with Helios, continuum, what they did with Helios, I think they know that they can't do this for larger systems. Okay. Because you can only really fit like 30 to 50 ions in that linear chain. I don't know how many they fit actually. No, with Helios, obviously, it was like 98. Right? So they had much more than, let's say like 200-ish. But if you want to get to many, many more, that thing is not going to work. That sort of architecture is not going to work. To scale up, you need this atomic highway to do all sorts of weird things um and it might be a real nightmare when it comes to optical engineering lasers are not great to work laser beam laser beams are not great to work with i'm gonna be honest okay um if you want to do arbitrary greats arbitrary gates on millions of ions you're gonna need like
a bunch of really perfectly aligned laser beams. Okay. There's going to be these like spatial light modulators. There's going to be massive lenses that are all going to be pointing inside this vacuum chamber. Now that's going to heat stuff up first. You're pumping in lasers into a thing that's supposed to be cold. You're heating stuff up. And it's like the more lasers you add, it changes the economics of the equation. Right. It's not like a linear where it's like, oh, one additional laser means just like one additional.
Yeah. It becomes more complex. Anyway. Exactly. Okay. And, you know, if let's say like some random thermal expansion changes the lens shape. Right. Right. Right. Then everything else. Then like the laser is not pointing somewhere. Right. Now you got to deal with that. Lasers are just tough. Yeah. Okay? Lasers are tough. The other thing about the economics part of things, right? With superconducting circuits, I told you that, you know, it's going to be big. These things are going to be massive. Yeah.
Well, with this, the algorithms are going to take a really, really long time. Because the gate times are so long. Yeah. Right? They're like a thousand times longer than the superconducting circuits. There have been estimates that show... And obviously these are biased estimates, but this is a biased show, so we're going to talk about the biased estimates. It shows that superconducting arrays, if they want to crack 2048-bit RSA encryption, it's going to take them about eight hours with a scalable, fault-tolerant quantum computer.
Here, even if we achieve scale, because the gate times are so slow, it might take on the order of a year. to decrypt 2048 RSA. Which defeats the purpose of why we want these systems. Yeah, yeah, yeah. I mean, you could say that... Yeah, yeah, it's one of the reasons, right? You could say there are other uses for quantum computing, but as I talked about last episode... The one that's clear to me is Shor's algorithm. Yeah. Right? It's like everyone wants to decrypt stuff.
And that's also the geopolitical incentive. Exactly. Which is a driver of the investment into the space. Yeah. I mean, so Continuum, they have a lot of private funding. I think Honeywell is one of the big computer chip technology firms, is I think one of the big sponsors. Also... JP Morgan, I think they want to do financial algorithms and portfolio management with quantum computers. More power to them.
I'm not convinced. So there are use cases for it. And they obviously have the backing, private backing, and now public backing with the IPO. And I'm actually excited to see what they're going to come up with for their next iteration after Helios. Because One of the things that we're going to get into later is the fact that algorithms are getting more and more efficient. So maybe you don't need that many qubits and that many gates to do stuff.
And it could be irrelevant. So it's an open problem, right? But in any case, as of now... With all of the optical engineering, like if it takes a year for you to do RSA, like in that year, your lasers have to be perfect, right? For a whole year. Like it's tough, right? So again, opinions are all my own. Final verdict. The same as superconducting. If you say that I copied the slide and just replaced it, that is actually exactly what I did.
But again, proprietary. The standards by which we judge things is proprietary. Cupid quality, 5 out of 10. Cupid control, 5 out of 10. scalability and economics 1 out of 10 to get 11 out of 30 what i would say just as a brief side note is that being said with the same score the lever in which superconducting qubits versus trapped ions could change uh the levers they would pull are different um yeah right like with with trapped ions if the algorithm problem goes away the scalability in economics very quickly um
is a solved problem as an example yeah and so while they're they're similarly scored the levers of changing are fundamentally different yes yeah and as as time goes on there's a there's an interesting plot oh i wish i'd had this overlay you know but maybe i'll put it up uh in the clips later go on go on instagram um there's a plot that shows um on the on the x-axis time and on the y-axis like the efficiency of the algorithm effectively. Like how many qubits do you need or how many gate, how many Clifford gates do you need to, or whatever gates that you need to implement.
And like, it's steadily decreasing as time goes on. People are finding more and more efficient ways to, to implement this kind of stuff. It's like the, not quite, but it's like the inverse of Moore's Law. Yeah. Not quite, because Moore's Law is a little bit, the numbers are a little different. Yeah, well, it depends. I mean, if you do Moore's Law size of transistor, that's going down, right? Oh, that's going down, yeah, okay. Moore's Law is traditionally number of transistors in a chip, and that goes up. But you could do it one way or the other. So that's trapped ions. So we've covered superconducting qubits. We've covered trapped ions.
And then the next thing we're going to look at. Neutral atoms. Neutral atoms. I didn't even want to do this. But we got it. For our audience, we're always going to give you the best you can get anywhere. This is the best. This is the best. So we're finally going to do neutral atoms. There's a bunch of companies that are actually trying to do neutral atoms. Qera, Pascal, Atom Computing. Or Atomic is a new one that came out of Caltech, actually. And it raised like $300 million. Yeah.
on low qubit quantum computing. Their idea is that they actually don't need that many qubits. Their physical to logical qubit ratio is very low. That's the claim they're making. But let's try to get behind the PR hype here. So what is the qubit? So like trapped ions, the qubit is a real pristine atom, but it's not ionized. Most of the time it's like a rubidium atom, or a cesium atom, these are coincidentally the same atoms that are used in atomic clocks, and for a very good reason, because they're highly tunable,
highly precise. The transitions are like highly precise and you can manage them very well with like lasers. There's no electric charge and because they're neutral, we can't confine them using that pall trap. Right, with the electrostatic fields. So instead you trap it in midair using optical tweezers. I shouldn't say midair, really it's mid-vacuum, right? But you trap it with an optical tweezer. It's a highly focused laser beam where... Because of the physics of the laser beam being focused, the neutral atom wants to live at the focus of that laser beam.
There's like a dipole force that happens that restores. Like every time the atom moves away, there's a dipole force that restores it back. The qubit states are 0 and 1. And again, it's the hyperfine clock states. So we basically can point at it and be like, stay right here. Yeah, there's like a laser. You concentrate it. You make it stay right there. And it's the same super stable spin transitions that are used in atomic clocks. It's like a tractor beam. It is kind of. Actually, exactly. A tractor beam is like, yeah, that's exactly right.
It's a tractor beam. Now, the problem is... Actually, before we get into the problem, how do we talk to it? Well, we use lasers, just like trapped ions. We use lasers. Single-cubit gates are driven by two-photon Raman laser pulses. We don't have to get into it. Two-cubit gates are a bit different. Now, instead of using sound waves like we did with trapped ions, we are going to use something called a Rydeberg state and the Rydeberg blockade. Here's the idea. You pump a laser into it.
So that the electron that's on the outside gets knocked to a really high energy level. Okay. Okay? Like, imagine you're taking an electron that's in the orbit of, like, Earth near the nucleus, and you're knocking it all the way out to Pluto. Oh, yes. That's quite some distance. You've increased the size of the atom. Okay. By a massive amount, like almost by a factor of a million. Because now the orbit of the electron... Is like out where Pluto is, kind of, right? Now, because this atom is now very big, and its neighboring atoms are still the same size, let's say, it's going to start having effects on the neighboring atom.
The neighboring atom is going to start getting nudged. That physical volume is going to start nudging this atom. And that massive shift... is going to make the other atom shift its energy levels. Yeah. Okay? And then whatever laser can no longer excite the other atom. This is how you do entangling, effectively. You're making them talk based on something called a Vanderwaal's force, which is like volume becoming fat and making the volume do the talking.
So, again, so previously we used sound as the basically the vector by which we, you know, had the communication between the qubits. Now we're seeing this like very like fast, almost instantaneous, massive change in volume. has these derivative impacts on the other qubits in the system. And that's what's communicating. So we're going to make it go from a current size to very, very big in a very particular way because we then know how that volume change is going to impact the other qubits.
And then we read – that's the way to communicate the zero-one – to communicate in this two qubit plus states. Plus states, yeah. It's how you make the qubits talk to one another is – blow it up, and then because this thing is blowed up, the other stuff is going to be like, whoa, and that's your entangling, and that's your two-qubit gates. Interesting. Interesting. I mean, I wouldn't do it that way, but whatever. Look, interesting. So one of the problems here, though, is that with the trapped ions, because it was an ion hyperfine frequency, and the difference between the zero and the one
is very far away from the infrared stuff that we see at room temperature. Here, though, the thing is not ionized. It's just at a really high Pluto orbit. And this goes back to the rungs of the ladder. Yeah, the rungs of the ladder. Before the rungs of the ladder were very big, so then If infrared came in, eh, it doesn't matter. Here, though, because you're now at Pluto's orbit, Pluto plus one is actually kind of nearby. Yeah. So the infrared bath that I'm in might actually start knocking you into these other registers.
In a way that entrapped ions, it did not. It did not, right? So that's a problem that they need to deal with. Okay. Oh, let's go into cubic quality now. Let's go into the FFP audit. The strength here... You can pack thousands of them into a tiny 2D or 3D array. Okay. Okay? You can have a bunch of these optical tweezers, and you can make a grid. Like, this is kind of cool. You've got a grid of neutral atoms that are all together. Mm-hmm. that are packed in. And each of those is a single atom that you're looking at.
That's suspended in a checkerboard pattern. In a sunbeam. It's kind of cool. And each of these is spaced just like four or five microns apart. Okay. So it's quite small. So it's quite small, right, yeah. And the physical density is quite large, right? That's actually pretty cool architecturally. Architecturally, right. And now a lot of these groups, they report fidelities that are really high, 99.5% gate fidelity. Like the way that they're moving these things around, it's like very, very accurate. But you read the fine print.
You read the fine print. What's actually happening is the following. during their experiment, they will lose atoms. Because the atoms are in this tractor beam. Sometimes they'll leave. Sometimes the photon will come in, they'll leave. Another atom from the vacuum will come in, they'll leave. They're going to lose atoms. Every time they lose atoms, they just throw away that experiment. That doesn't count. Yeah, they're just like, ah.
We have so many of them. Yeah, yeah, let's just do it again. And then so there's this post-selection, and many of the published fidelity figures calculate conditioned on Adam's survival. They throw away the trials where the Adams died. Which doesn't count. Which is like, I mean, there's a reason. Okay, there might be a good reason for you to do that. But then at the same time, I should be able to say, is that the same as the numbers that other people are reporting? If I'm running a hundred meter dash and I can run it a hundred times and just throw out the ones I don't like...
Does it still count? Yeah. I don't know. I don't know. I'm a layman. I'm just asking the question because that's what it sounds like to me. But please convince – I can be dissuaded from this perspective. Yeah. Anyway, so that's cubit quality, right? The quality of the cubit. All right. What about control? Control – like the fact that you can use these laser beams, you can bring any two atoms together. Mm-hmm. And you can make the tweezers sort of grab this atom and this atom and bring them together, grab this atom and this atom.
This is an experiment where they showed, like, using these tractor beams, They can create like a 3D sculpture of stuff, which is kind of cool. That's very cool. Right? And so this is to show you that you can physically move them across the array. You can park them right next to new orbits, mid-circuit. So you can have like this all-to-all connectivity type of thing. And because you have all-to-all connectivity, you can now... Not limited by two. Not limited by two, right? By trapped ion. Exactly, yeah. So now you can... you can do error codes that are like way more efficient, right?
Because you can harness this all-to-all connectivity. The readout, however, is kind of destructive because you do it by shining a laser and capturing the fluorescent photons, but the light pressure from this readout can heat up the atoms and then, again, they'll like leave. But then I guess you can just throw it out. Yeah. Would this then sort of mean like, the idea is that we're solving for lack of coherence time in this kind of system which was great in trapped ions but is not necessarily as great here by just saying we can scale this so much we can just have so many that coherence time is not no longer as important of a variable
Yeah, it's like, I mean, coherence time might not be the correct idea, but I think like survival. Survival. It's like cubit survival. Yeah, right. It's like, yeah, it might not be because maybe the gates are a little bit more faster. And I mean, the community is saying that they are working on it. Okay. Okay? They're working on this problem. And they've got these ideas where they've got a reservoir. where they can take the atom from the reservoir and put it in where the, where it was missing. The key thing is it's, it gets pretty obvious to know where the atom left. Right. Yeah. Yeah. Right. Because you've got a grid or whatever.
Right. And because you know exactly where the atom left, the error codes where, you know, where the error happened are really efficient. Yeah. And that makes sense. And again, these systems sort of have optimizations for different things or are successful in different arenas that, uh, Maybe other methodologies don't do as well. But the idea, again, going back to our original analogy of the transistor versus vacuum tube versus electric relays, it's still unclear right now, you'll make the argument otherwise, right now, which path is going to be.
Because there's going to be, maybe there are use cases for some niche areas for some of these, if there's not an outright winner, that's just good for everything. Maybe that's true. But it's also very possible that one of these is just going to solve each of these layers and just be the best for everything. And in this case, the architecture for these neutral atoms is really nice because you have a lot of fine grain control. Yeah. And that's why I think a lot of people think that this is the current sort of frontrunner.
Okay. Okay. Because it's out there, yeah. You've got all these nice little knobs that you can tune, right? Okay. And then to scale them, like how scalability? Well, to scale them, you've got, you know, how many atoms can I put in my array? Yeah. That's the idea, right? Yeah. You can take a singular electron beam and split it up into a bunch of optical tweezers. And then that's normally what we see when we saw that grid of atoms. That's really a single laser beam that you're splitting into a bunch of optical tweezers, a bunch of tractor beams, and then each of the atoms are in that.
One of the problems, though, there's two problems really with this. One is there's a laser power wall. Like how powerful can you make your laser? All right, you've got a thousand. Yeah. Right. Can you like 10x your laser power? If you 10x your laser power, can you not impart heat into the system? And two, if you 10x your laser power, but the laser has some noise because the laser is not magically tuned to some frequency. Right. There's a jitter in the frequency. And that frequency noise is going to be correlated across all of your beam splitters.
Each of the atoms is going to be receiving the same noise because it's coming from the same laser. And crucially, for a lot of the error detection algorithms for the fault-tolerant computing, that relies on uncorrelated noise. The assumption that the noise on all of these qubits is uncorrelated. That's a good point. So if it's all correlated because it's coming from the same source. Same source, then like, you know... That's good. Might might be a problem. Yeah. Right. There was also this really nice paper that came out in March 2026 that said that with just 10,000 quantum quantum bits, you might crack inherent encryption schemes.
This is out of Internet Internet encryption schemes. Yeah, internet encryption schemes, but also Bitcoin is going to go to zero. It's going to get your crypto wallets, everything. As headlines tend to. Yeah, this was out of Harvard, Caltech, and QARA. Naturally, the press lost its mind. But yet, mathematically, let's go into the paper. They've got a little figure that shows you the space-time trade-off. You're shuttling the atoms, and that's mechanically slow in order to do the error correction and things like that.
Even if you've got a 10,000 qubit array, it's going to take you a really, really long time. If you've got 100,000 physical qubits, even then... With mathematically optimal assumptions, it's going to take you about three months to crack RSA. Okay? Yeah. With 10,000, it goes into, like, years. Which is, again, not... Fine. Your computer isn't that big.
Right. Right? But it's taking a year now. Yeah. Yeah. And this goes back to the same issue we had with trapped ions. Yeah. Which is this cycle time is still... Yeah. In my head, again, as someone who doesn't live in this space, all I keep hearing is like this will allow us to do what you couldn't do in the entirety of the existence of the universe in the snap of Thanos' finger. So that's where my expectation level is.
Yeah, around time efficiency, specifically, independent of all the other details. And so when I hear it, it's still going to take, I thought this was going to be instant. No, no. As soon as it's done, RSA will just be over. Everyone can be hacked immediately. No, I mean, those headlines would have made you think that. Right. But you go into the figures and it's like, no, it'll still take a year. Yeah. Yeah. Which again, it's not a year. It's not a year, but also that means no one's targeting me. Right? No one has like the one quantum computer and they're targeting my wallet. It basically only gets critical infrastructure.
Yeah. It becomes like a war thing. Yeah. Yeah. Right. Yeah. And no one's going to use this to, to, to find the next room temperature superconductor. Right. Which is not, which is what we want to do. That's the whole point. Right. Yeah. And not just targeted for weapons systems. Please. So, so now we do the final verdict. On neutral atoms. Cubit quality, I give it a negative 1,000 out of 10. Yes. Cubit control, negative 1,000 out of 10. Yes, yes. And sustainability and economics, another negative 1,000 out of 10 for a total of negative 3,000 out of 30.
Yeah, yeah. I think this is fair. I think this is totally fair. Again, the metrics are proprietary, FFP, proprietary, IP. Yeah. for all the neutral atoms community i could never have thought um you know since that night at alice and bob that neutral atoms would end up at negative 3 000 out of 30 but i that's how the cookie crumbles so what can you do this is where we started and and it's just not it's not good yeah it's just not good and you can complain about in the comments
But there's just nothing you can do about it. And so we've now covered all of the top contenders but for the star of the show. So we started at super – no, we didn't start. We started, yeah, superconducting Qubits. Then we went to trapped ions. Then we went to neutral atoms. The worst. Yeah, the worst. Negative 3,000. Negative 3,000. Right. And all of them, we now have a very fundamental understanding of how they work across our three criteria.
Yeah. Which is qubit quality, qubit control, and scalability and economics. Which I think is a three-axis system that is really relevant to really try to understand what's going on. When we're talking about building a quantum computer, what actually matters? And from our part one, we have fundamentals of understanding all of the different weirdness that goes into the quantum mechanics and quantum algorithms and qubit state zero one, but with the phase and then the north south.
Something that people at the frontier didn't even know about. That's right. During your talk. Yeah, yeah. It's not made out of cheese. It's not made out of cheese. Made out of silicon. And it's also on the front page of nature. Unlike trapped ions. Yeah. Unlike... Neutral atoms. Trapped ions have been on the front. We're talking about it in 2026. 2026. Before. The before times. I don't know if neutral atoms have. I know superconducting has. Of course. I didn't see in my research if neutral atoms have.
If they have, definitely put it in the comments. Quantum supremacy, which is the superconducting. That's the superconducting. That's been on there. So we've seen everybody make their claim on the cover. Now it's time to see what the supreme qubit is. is going to look like I'm giving a lot of setup here because this has been a whole Christian. This has been a lot of hard work. Yeah. You've put a lot of blood, sweat and tears, not only into the work to be a part of this project, but to kind of put this series together to really give all of us as the audience, a real understanding of, of what it is this means and why in a biased way, you believe that this is the best option.
And so we are going to get now into this, understanding spin qubits. That's right. We're finally here. Spin qubits. All right. The nature cover... Quantum silicon. Which we will hang on the wall. I think we're going to now replace our downtown LA with a gallery wall. This will be our first addition to the gallery wall. Something that is very important. Yeah. We've covered the three big technologies. And now I want to get here. The story actually starts in 1998.
We're going to do a little bit of history. Okay. The year 2020 marked the Physical Review A's 50th anniversary, and as part of the celebrations, the journal presented a collection of milestone papers, 50 milestone papers. This was one of them. It was listed as a milestone paper. Quantum Computation with Quantum Dots by Daniel Loss, who was at UC Santa Barbara in Basel, and DiVincenzo. who was at IBM and UC Santa Barbara. So in this paper, Lawson DiVincenzo, they laid out a proposal for quantum computation based on quantum dots.
Quantum dots meaning little tiny localities of charge that are in some kind of solid state device. And it was a detailed investigation into how you can do one qubit gates, two qubit gates with these quantum dots, single electrons that are localized in some solid state device, in some like piece of metal. Now, the idea is you're going to use some piece of metal to confine the electrons into a geographically isolated place.
This thing is going to have to be small, something like tens of nanometers for each electron's room, so to speak, because the electron's wave function is that big. You don't want to fit too many. So this thing has to be really small. It's going to be confined in the z direction. by the material itself. And I think we've got a thing that shows that. Yes. So, you know, you can imagine you have a silicon substrate, like some kind of chip, and you do something to the silicon such that all of the electrons want to live in a plane.
So you get a two-dimensional electron gas, a 2-Deg is what they call it in the industry. This can be either because you do silicon and then you put some silicon oxide and then you put some more silicon. Or you put silicon and then germanium and then you put some more silicon. Some kind of heterostructure that makes the electrons want to live directly. in a plane everyone wants to live on the second floor of the apartment not the third fourth and fifth yeah not the first but for whatever reason yeah for whatever reason we all want to live on the second floor we all want to live on the second floor now crucially that means that one dimension of confinement has already been taken care of because of the material it at the architecture level we've already confined the system to some variable we know yeah right so
What that means is I only need to confine it now in these two dimensions into this spot and this spot and this spot and this spot, kind of like a chess checkerboard. As opposed to when we talked about trapped ions and neutral atoms, they're in a 3D, three-dimensional space. So the complexity of maintaining, you know, X, Y, and Z, pun intended, is significantly more difficult. Exactly. Specifically with trapped ions, right? Because they're kind of related to electrons. Electrons have charge, just like trapped ions.
And one of the fundamental theorems that you learn about in undergrad electromagnetism is this idea that a static electric... field cannot confine you in three dimensions. That's why the trapped ions needed that rotating saddle. But here, I've already got 1D confinement. And so in order to confine in 2D, I can use static electric fields. Right? I don't need to... oscillate stuff in order to localize the charge here. We don't need a harmonic oscillator?
No. That's quite nice. That's quite nice. No harmonic oscillator. Well, there's still going to be a harmonic oscillator. But like... You know what I mean. I know what you mean. I know what you mean. So what we can do there, and if you could bring up that... For 59 again? Yeah. If you could bring that up. So you're confining it into 2D. Okay. And now what you can do is in the 2D, you can create an egg carton potential landscape. Okay, for example, you can have an electrode, like a little wire that goes up top here, pops down.
On the left, you see an electron microscope of all of these wires that are going down from all these places, and that are going to plop down at a certain spot. I can have the wire go down, maintain that wire at plus 5 volts, have another wire right next to it that goes down, maintain that wire at negative 5 volts, right next to it, plus 5 volts, negative 5 volts, or millivolts or whatever. The idea is not the scale. The idea is that the sine I can switch such that I create a mountain and a valley and a mountain and a valley.
And all of the electrons are going to want to sit at the valley. And so now I've got a way to confine single electrons. Okay? That's the idea. Okay. Okay, so all of these electrons are now confined to... in these single electron wells is what we would call it. Because we've been able to confine everything to floor two. And then we have these electrons rooms that, that we can basically decide which has a two story on the second floor. And we can just turn off the two story second floor room or not.
Or not. That's very good. Yeah. And yeah, it's like we can open the door to the room or not based on like the, the voltage that's coming in. right? The qubit itself in the, in the original proposal, the lost even Chenzo proposal in that original proposal, the qubit itself was the spins of individual electrons. So you can split them up using a magnetic field. Like if the spin is this way and then the one neighboring, it is also spinning in one direction. You can split them up with a magnetic field. And how do you talk to the qubits? How do you talk to these like LD qubits?
You apply a microwave magnetic field. That's going to switch them from one spin to another. And you can make them talk to each other by simply lowering the voltage barrier between two adjacent spins. So you've got one electron, let's say, in this egg depression. And then the one right next to it, there's a barrier in between that's keeping them separate. If you want them to talk to each other, just lower the barrier. Mm-hmm. Right. There's a, there's a little electrode that is maintaining that barrier.
It goes negative five volts, positive five volts, negative five volts. Actually it's opposite because electrons have negative charge, but whatever, whatever that barrier is, just lower it. And now the electrons mush together and they can talk to one another. Would it be like in those hotels where they have the two doors on each side of the hotel room? And you just open that door. Yeah, you just open that door. To allow them. So it's like the first level is like opening the door to get into the room. The second level, which is creating the well. And the second level is the door in between the two wells, like the hotel room, to allow them to communicate. Exactly. And that's called the exchange interaction.
Okay. It was first proposed by Heisenberg when he was trying to talk about magnetic fields. But here's the idea with the exchange interaction and why that works so well. First... It is a DC signal. Okay. It's just DC voltage where I'm like setting the voltage to this and then I'm moving. There's no oscillating stuff that's happening, right? When I want the electrons to talk to one another. And that exchange interaction is really, um, the way it works is it sounds kind of like a magnet magnetism interaction.
Like, you know, if you have two bar magnets that are aligned the same way, they're going to kind of repel one another. Um, yeah. Quantum mechanically, it's not really a magnetic interaction, so much so as it's really electrostatics, like the electrons don't want to be next to each other because they're both negatively charged. And there's a Pauli exclusion principle that's happening where if the two electrons are spinning in the same direction... If they have the same spin, they're not going to want to be close to one another.
But if they're spinning in the opposite direction, they're okay being close to one another. Because they have opposite spins, and this is how the periodic table works. This is why there's two for every energy state, every angular momentum state. You populate it one this way, and then you put the next electron in the down state. So one up state, one down state, and then you populate the periodic table that way. The electrons want to be close to one another only if they have opposite spin. If you have this whole, in this hotel room analogy, I don't mean to keep making this dumb, but like if you have a soccer tournament and you have two teams, the door opens, you have red team and blue team, right?
If the door opens between red team and blue team, they're not going to want to talk to each other because they're on opposite teams. But if you have between red team and red team, again, in this analogy, they're going to want to talk to each other. It'll be somehow opposite because what I'm saying is the electrons are opposite. They want to talk to one another. So fair enough. So maybe they want to fight. The red team and the blue team. Everyone's agitated. They want to get. So and I know. They'll keep their space if they're the same team. I'm mixing metaphors here because I know your point that you're saying that the energy, it's the positive or negative or the spin one versus another direction.
And so if it's the same, they're not going to want to. Yeah. And so in the analogy I'm talking about, you know, red and red and blue and blue are not going to want to. But if they're opposite colors, they will want to exchange words. Banter. Banter. Yeah. Let's say it's the banter gate. And so if they're opposite colors, they're going to want to jaw at each other. But again, just to keep – try to simplify that. There's the complexities about how those spin states exist. But I think part of the point we're getting at is like we –
Just like every other system we've talked about, you have the potential well has... The entanglement state has two... The zero and one is very well defined by the dynamics of spin in these systems as has already existed and is well defined. Yes, yeah. We understand electrons and how they interact with one another very, very well. It's no problem. And so if we can create these electron wells where they're sitting and we can move them around... in these wells and make them talk to one another, we've got a very good handle on how these electrons will talk to one another, how those spins will change as they talk to one another, and so on and so forth.
How do we read them out? Okay, you're right. Right? Like, how do I tell if the electron is spinning one way or the other way? Because we know we understand how these things could, but then how do we see that they are? Exactly. Like, with the superconducting circuits, there was a resonator, that if it's in a zero or one, it would resonate with that microwave cavity, and then I'd be able to read it out, whether it's one way or the other. With ions and with neutral atoms, you basically just image them in some sense. You, like, use a laser and you try to image the light that comes out.
Now, with spins, it's a little bit tricky, right? Because I don't have direct access to which way they're spinning. Okay. But from the dynamics that I just told you... If they're both spinning in the same direction or if they're spinning in opposite directions, I'm using this very colloquially in some sense. Because usually it's like the spin states are not just straight this and this or this. There's combinations between them. There's something called the singlet state and the triplet state, which is the spins going this way and this way.
You either add them up or you subtract them. And those are the two states that are zeros and ones. But the dynamics that I told you earlier about how they want to be together or not is still the same. If they're in the singlet state, then they have no problem being in a single state. Spot. If they're in the triplet state, they have a problem being in the same spot. OK, because of the Pauli exclusion principle is part of what you're saying that what puts them in the singlet or triplet state is a lot of options. Yeah, there's it's not just up or down.
Yeah. But to simplify like two options, but like it doesn't matter for this discussion. For sure. What matters is that those are the two states. Yes. And. For the singlet state, the Pauli exclusion principle lets them hang out in the same spot, but the triplet state, it does not let them hang out in the same spot. So now what you can do is if I have two wells, and now I raise one of them so that I try to put both of the electrons in the same room, if they're in the singlet state, it's going to be way easier for them to get into the same room.
But if they're in the triplet state, they're not going to want to get into the same room. It's going to take a lot more time. And if I sense... whether there are two electrons or just one electron in that room that I tried to shove everyone in, then I've got a readout mechanism to tell whether I'm in the zero or the one, whether I'm in a singlet or a triplet. That's what's happening here. So on the top, you've got kind of a triplet state in the sense that the electrons are not aligned. And when I try to... They are not aligned, right. And so in the singlet state, when they're not aligned, if I try to shove them into one, it's allowed.
But if they're in the same... then when I try to shove them, it kind of gets blocked because the electrons don't want to talk to one another. And the point is, if I can tell what the charge is, If there's two electrons in that second room or just one, then I can tell whether it was in the zero or the one. And that's my readout mechanism. So you basically do something to change the environment. And then when you look at it, if the two electrons are together, then you know it's zero. If they're not, then you know it's a one. Then it's a no.
To keep my annoying analogy together, in the movie Inception, when the hotel room started rotating... If when it started rotating, the electron moved into the room, then you have your zero. If for whatever reason it was sticking to the wall and refusing to go into the room when the whole hallway was rotating, you're in your one state here. Exactly. Now, this introduces a caveat, though. It's kind of magical that we can even tell that there are two electrons floating. In the room in the first place. In the well.
Right? Like, how do I tell if there's two electrons there or one electron's there? Right? Like, imagine, again, this is a solid state, like, piece of metal. It's a piece of metal where we've confined it single electrons. I need to know where the electrons are. Yeah. In a metal. Yeah. Yeah. There's a lot of electrons everywhere. Yeah, I was going to say, which one? Right? Like, how do I tell where the electron is and how many there are and so on and so forth?
Okay. So to actually see the thing, I mean, we're talking about single electrons again, right? You can't use the resonant cavities and you can't use the lasers. Right. and all that other stuff. Oh, no lasers. Instead, we used something called a single electron transistor. This was one of the coolest things that I had to get used to once I was getting involved in this project. A single electron transistor? Yeah, it's such a cool-sounding thing, right? Okay. So here's how a normal transistor works. There's a source and a drain, and if there's current flowing between the source and the drain...
that's your one. And if the current is not flowing in your source and drain, that's a zero. And the way you make the current flow is you control the gate, the voltage on the gate, and you lower the barrier. If you lower the barrier enough, there's a bunch of current. If you don't lower the barrier enough, there's a barrier, and it stops the current. That's how transistors work. Yeah, yeah, yeah. Now, there's going to be a point... If you manufacture this stuff so cleanly and so closely, such that there's a source electrode and a gate electrode that are right next to each other, maybe a few nanometers, tens of nanometers apart.
You've also got a little gate in the middle that controls that voltage. If you lower the gate just enough, single electrons... are going to go through on the left-hand side. You slaver dog. I see it. Yep. Right? I get it. Yeah, I get it. Basically, you're just creating a gap that's only the size of literally the shell of an electron. Yeah, this is a review paper that was actually written by Michel DeVore, who is one of the Nobel Prizes. last year. Yeah, that's quite nice.
He's been working in quantum tech for a long time. And he's saying you can amplify quantum signals using the single electron transistor. Here's the idea. If I've now got this sensor, this is effectively a sensor. Right. I've got a little single electron transistor that is moving these electrons from one spot to another. If now I start changing the environment in my device, right? The rooms, let's say your hotel rooms, are all on the... Let's say there's a lot... You know where the elevators are on the second floor?
Yeah, yeah. Right? Let's say those elevators are where the single electron transistor is. Okay. Right? You've got a source elevator and a drain elevator. Yeah, yeah. And you've got this like shuttling that's happening, right? If the rooms start moving around with electrons... the electron that is doing this single transistor effect is going to feel the effect because this is negatively charged and all the other electrons are close by and negatively charged. And if I can sense that current very precisely, I can get little tiny deflections, little blips, whenever the states in the rooms change.
So by watching my single electron transistor very precisely, I can tell all of the stuff that is happening on the second floor. Would it be almost like hearing footsteps when you're in the elevator of someone running up and down the hallway? Kind of. Yeah, yeah. Exactly. Like the idea that you can sense at distance something that's happening from your single electron transistor, which is this elevator shaft. Again, trying to create a very crude analogy here. But effectively, that's what we're trying to say. You can sense what's happening in these wells, these potential wells.
Yeah. at distance because the charge of what's happening in the wells will impact the electron gate that you have yeah that you're sensitive that you have is your sensor yeah exactly so i've got a sensor right here and that's like got a little tiny bit of current that i'm sensitive to and as the electrons move around in my device This thing is going to start changing. It'll jiggle. It'll jiggle. And if I'm sensitive to that jiggle and I know my physics well enough about what each jiggle represents, I can tell what's happening everywhere.
You can translate it into position of all of the... Yeah. Okay. It's kind of... When I first got my hands on one of these devices and I was like, you know, I had a single... Like I made a... call it a charge sensor or a single SET. It's just, it's so cool to think that that's what you're doing. It was one of those moments, I think like in my life when I was like, this is like really cool. That's really, I'm watching like single electrons move around from one compartment to the another on this nano fabricated device.
That's been fabricated. Yeah. Yeah. It's, it's, it's really cool. I mean, it's like you're, you're sensing quantum stuff. Yeah. Right. That's happening. Yeah. Yeah. That's quite nice. It was really cool. I still, I will never forget the day that I had my first SET. I made my first SET and then I saw like my first tunneling event and things like that. It was, it was really cool. Um, So that's one way to implement it, which is, in this case, every single electron that's in my checkerboard, eggshell, whatever you want to call it, is a single qubit, right?
It's either spinning one way or it's spinning the other way based on a magnetic field. I talk to it based on microwave frequencies. Now, there's an argument to be made that, like, you don't want any microwaves whatsoever, right? Even you could have just a single microwave that's talking to these guys, and then you can tune all of them based on the single microwave. But what if you don't want any microwaves whatsoever? What if all I want to do is lower the barriers? Okay? That's all I want to do is exchange.
Open the doors between the hotel rooms. Yeah. Remember, before the exchange was for two qubit gates. Yes. But in order to do one qubit flipping, I still needed the microwave. Yeah. And I still needed the magnetic field. Yeah. What if I want to do everything based on just opening and closing doors? Mm-hmm. So DiVincenzo came up in 2000 with this Nature paper, along with a bunch of his colleagues, with the exchange-only qubit. Okay? This is a qubit where both single-qubit gates and two-qubit gates are mediated only by the exchange interaction.
Meaning we now no longer need two separate basically vehicles for interacting with this or controlling the system. Yeah. That are almost related but independent variables that when you come to like the experimental apparatus – like need to be controlled in different ways. Yeah, and maybe it's better. Maybe it's better if there's just only one thing we do, which is just exchange, it's all DC. It's one thing to handle. There's no microwave signal that's going in. It's only one thing that we gotta get really, really good at. Right. Which is ideal. Yeah. The caveat though, is that now your qubit is not a single electron.
You need three electrons in entangled states. And then you can now have a singlet and a triplet be your zero and one. Before it's like this was a zero and one. Now it's like this is your zero. And then another set of spins is your one. Okay. Okay. So the caveat is you've made your problem a bit bigger. Right now, instead of a single electron, you're worried about three electrons at a time. Because that's what's required to give you a zero. That's the trade-off.
That's the trade-off. And... This is where the exchange-only interaction comes in, and this exchange-only qubit. Three electrons now. Yeah. Okay? Okay. The first two, whether the first two are in a singlet or a triplet state, these two different spin states that we don't have to worry about for this episode, that's going to be your zero and one. Okay. Okay? And the third electron is there to give you full qubit control on the block sphere, meaning, like, you know, for a full qubit, I need to be able to rotate on two axes, right?
I need to access this, and I need to access every longitude. Okay. And every latitude. And so the exchange on the first two is going to let you rotate on the north-south. Yeah. Yep. And the exchange on the second two, two and three, is going to let you rotate somewhere near the equator. Not actually at the equator, but somewhere off the equator, just based on how the algebra of the space works out. This is interesting. And so it requires three electrons now to basically define the two states of zero and one.
Yep. Based on how the first two of those three interact versus the second two of those three interact. Yeah. And so we're basically deriving two states from a three-body system. Yes. Now, with the three-body system, right, there's actually one – there's another advantage here. So the first advantage I've already told you, which is that everything is DC. All you have to do is lower the barriers in between them. So everything is DC control. So you don't have to worry about like AC oscillating, electromagnetic fields, and all that other kind of crap going in.
The other thing is that because there's three, the algebra also works out such that you are insensitive to... to global magnetic fields. If there's a giant magnetic field that's going through this, you don't care because all of these spins are rotating in the same way. It's kind of like, remember in Interstellar, that scene where they tried to dock with a rotating... like spaceship yeah it's like tars we need to we need to match the rotation and then like tars makes the thing spin in a certain way and um whoever who's the actor matthew mcconaughey is like he's like he's like pushing his uh head in the other direction to counteract the g-forces and they're spinning in such a way that they match the rotation and then they can dock similarly in this case like the magnetic field a big magnetic field is going to come in as long as it's the
Everything is going to rotate in the same way, and your quantum information is going to be preserved. And now, because we have that third body in the system, it partly is why. That enables... The system is large enough locally that a larger external factor is going to impact, but the system still has enough... Yeah, there's like an algebra in here that is going to just be invariant to all of the big rotations. Yeah. Right? Crucially, if there's local magnetic fields, that's still a problem.
Right? Like if the third one is rotating in a different way than the first two. Still a problem. That's still a problem. But a global magnetic field, like the Earth's big magnetic field, you don't care about it. Not a problem. Because it's mostly going to be the same across these three, like a few tens of nanometers. So it's like gauge hacking. I call it gauge hacking. Okay. Because there's like a weird gauge theory that happens here. Oh, gauge theory made it in. Yeah, exactly. So that's kind of cool, right? Okay. Now... Why are people initially excited about just spins in general?
Okay, when DiVincenzo came out in 2000 with this paper, or in 1998 with the first paper, people were excited because silicon is nice, and you can do this in silicon. This was a proposal to make a quantum computer in silicon, right? Which means I can now leverage all of the silicon manufacturing that humanity has gotten really, really good at. Right, right. This is a single silicon wafer. It's a piece of, I don't know, 99.999999% pure silicon that then you plug through an ASML EUV lithography machine and then you print chips.
This is where your GPUs come from. This is where the chips in your laptop come from. Everything is made out of silicon in today's economy. And so the idea is this methodology of spin cubit, like, sorry, of what DiVincenzo had come up with at the time, the substrate you could build it on top of did not require lasers, did not require barium, it did not require, what was the other one? The Y-savoc.
the the um neutral atoms yes yes it didn't uterbium uterbium did not require any of these yeah tantalum tantalum yeah like all of that it's just silicon it's just we can and the part of the point here is our entire uh current economy uh yes uh manufacturing based as it relates to computing systems are well suited for scaling silicon that's convenient this was in 2000 which it wasn't even at the point where this is that big yet yeah but nowadays it's even more clear exactly or more robust yeah okay so in terms of scalability this seems like already like there's there's there's ways that this could work right sure now why have people been skeptical though yeah because
Because clearly that guy at Alice and Bob didn't even know about spin qubits. And even the big higher-ups who know about spin qubits, they're always like, eh, but probably not. There has been good reason for skepticism. For one, the fabrication requirement here is difficult because you need to use one of these major fabs that has this... So unlike, for example... trapped ions on neutral atoms, you can create small qubits, small numbers of qubits, using bespoke physics machinery that is found in the lab.
Lasers are very common in labs. Superconducting circuits, also you can kind of just, like PhD students can make their own because it's the size of a millimeter. You can use a light microscope and make stuff. With this, you need to fab it. Yeah. You need a fab that's, like, willing to, you know, be like, okay, let's print this thing. So that's difficult. You can't just make this in grad school. At least not the big, at least you can make, like, big ones in grad school, right?
And there's certain problems. There are like inherent problems that have to be solved. For one, I told you about the magnetic field problem, which is that a global magnetic field is fine, but local magnetic fields are a problem. Now, silicon, normal silicon, the stuff that's in your computer, has a lot of local magnetic fields because silicon comes in two isotopes. It comes in a 28 and a 29. That's the number of protons and neutrons in its nucleus. With 28, because of the Pauli exclusion principle...
All of the spins are going to cancel each other. That's how they're on top of one another. And so there's going to be no net magnetic field inside every single atom. But for 29, there's an odd number. So there's going to be a tiny magnetic field because of the spin of the nucleus. So we want 28. We want a lot of 28. We don't want a lot of 29. But naturally occurring silicon has about 92% 28. And for the 29, there's about 5%. Now, for normal silicon, that doesn't matter.
For normal classical computing, it doesn't matter. But here, your quantum information is going to be lost. There's too much. Yeah. Okay? Okay. So that's one reason. Charge noise is a thing, right? If you have, like, random stray charges... your fab isn't good enough or something, then you're going to have some random electromagnetic fields. That's going to cause a problem. And then finally, there's this demon called the valley splitting. Okay? This is something that I'm going to be honest, I do not understand. I'm also going to be honest, a lot of practitioners in spin qubits don't understand.
And a lot of the people outside of spin qubits who say that valley splitting is the problem also don't understand. It's kind of just this thing where the real theorists are like, yeah, it's clearly a problem. here's from what I gather from reading about it, okay? The idea is that silicon has this really weird band structure. Like, remember, in previous episodes, we talked about the valence band and the conduction band and how the difference between these two is what gives silicon all of its nice properties as a semiconductor.
Now, part of that band structure means that in bulk silicon... The electrons actually can choose to live in one of six different states that all have the same energy. For classical computing, this doesn't matter. But for quantum computing, we only want two states. Okay? Which means we only want one of the valley states where the electron can live in either a spin up or a spin down. If there's six and in each of them they can live in a spin up or spin down, now all of a sudden I've got 12 different things that I need to worry about.
I only want two. Mm-hmm. Okay. Yeah. This is for us living in the valley. There's six different ways to define what is the valley in Los Angeles. Yes. And we want one way to define what is the valley. The valley is Los Angeles. Yeah. But the problem is part of the valley is in Los Angeles, the city. Part of it's not in the city of Los Angeles. What we're kind of saying is we just want the part of the valley that's in the city of Los Angeles. Yes. Yes. Exactly. And not the other stuff. And not the other stuff. Not Burbank, Glendale.
Right, right. Not Thousand Oaks and all that. However, that's not how it physically is. Yeah. Physically, all of these, the electrons can't tell the difference between the cities. Right, right. It's like we're in the valley. Oh, it's we're in the valley. Exactly. Anyway. Now, that's in bulk silicon. You get six. Now, if you do the heterostructure that we've been talking about that confines the electrons into this 2D plane, you put, I don't know, a layer of germanium or something, then that lifts the degeneracy, and now you're only worried about two things, because the two dimensions in X and Y are in one state, the Z dimension, which is the one that has the different symmetries in the other state.
So now you're worried about two. But that's still two, and if the electron lives in a spin-up, spin-down in the top, or a spin-up, spin-down in the bottom... There's still a problem. Because now you've got four. Again, I only want two. Yeah. Yeah. Okay. And that's been kind of a problem. This is a limiting factor because the position, like this impacts everything else downstream from a qubit quality and qubit control perspective.
Both. Both. You're very correct. Both. Qubit quality in terms of I don't really have a well-defined zero and a one. And qubit control because if I try to kick my zero, 201, how do I know I didn't go into the other zero of the other valley state or some nonsense, right? And when I'm reading it out specifically, like, and I'm trying to do this polyspin blockade thing where I, like, shoved, like, what if the stuff goes into the other valley state? Right. Right. This is like kind of an existential – It's fundamental. It's a fundamental problem.
Because if you don't solve this, then you can't get to sustainability in cubic quality or cubic control. Yeah, yeah. And then scalability is a matter of thing. Right, right, right, right, right, right, right. Because we still have this – there's a physical – it's about the material. So this is a problem that arises out of the actual physical manufacturing of the material, both in like the quality of it and then – anyway. Exactly. This is a big deal. This is a big deal that needs to get solved. I can understand why people are like, yeah, yeah, because it seems like just like this magic trick that silicon does that, first of all, they don't even understand.
So then when someone tells them, oh, it's impossible, I'm going to go do my neutral atoms nonsense in Munich, Germany. Sorry. So anyways, there's that. And then finally, there's also a wiring problem because each of these gates, you know, as I told you, there's a wire that has to go in. and control the plus five, minus five, plus five, minus five. Each of these have to be independent wires that go in. And so you're sort of met with the same wiring problem that you had with superconducting qubits. Conducting qubits, yeah. Except it's a little bit better because with superconducting qubits, you had coax cables that were like semi-rigid.
Here, at least, you've got like just wires, right? So it's a little bit more tractable. Inherently, though, there is still a wiring problem. Okay. Okay. Now let's finally talk about HRL and this particular paper. Yes. Because that's sort of where we landed. Yes. Right? With why people think spin qubits ain't going to be it. Right. Okay? Now, HRL, for those who don't know, it started its life out as Hughes Research Labs after Howard Hughes, who is the Leonardo DiCaprio character in The Aviator. Yes. And if you've seen that movie, there's like this…
There's like this gaggle of engineers and scientists that are trying to make him this crazy airplane. And he goes to them and he's like, these rivets are on the airplane. They need to be completely flush. And they spend their time like... sanding down the rivets on the airplane and then he takes it for a ride and he like lands in a cotton field and crazy things like that those engineers and scientists are what became Hughes research labs now at the time of the movie there was no hrl but i feel like that's yeah that's hrl okay it was like howard hughes going and being like make me this and then they do it he got this nice little plot of land right above pepperdine university in malibu and he built that lab um it's
it invented the laser way back in the day. Didn't get the Nobel prize because the Maser won the Nobel prize for microwave amplification. The laser was the optical version of that. So the Maser had already won. I think that's why the laser didn't win. Kind of weird because like other places, other laser things have won for a lot less. So HRL has been working on encoded spin qubits for a while. They've had a lot of papers over the past decades. A big one came out in 2023. It was submitted, I think, in 2022.
this one was in 2023 universal logic with encoded spin qubits in silicon here they showed an instance of the first universal logic meaning universal computational logic like you can do anything you want with two qubits but remember two qubits with exchange only means six electrons. So there were six electrons, yeah. And we've got a little video that shows exactly how that works. So here you've got your six qubits the six sort of potential wells, and the blue are the exchange gates, the barriers in between.
Okay? This is the wiring. And here you can see sort of the wiring problem already. Even with those six cubits, there's all these wires that need to go in to control the voltages in that little environment. Yeah. Right? Yeah. Yeah. These are the gates that go through the silicon heterostructure, as you can see. And right in that layer there is where the electrons are confined. Three at a time for a single qubit. It's called a DFS qubit here because it's decoherence-free subspace.
That's the whole idea of I don't care about magnetic fields. So I'm free from decoherence due to magnetic fields. And now you can see the voltage pulses come in. They mix the electrons together. And that is everything that is needed. For all logical operations. And then what we're seeing with the squiggles, is it kind of similar to the resonance lines we saw in, I guess it would have been the superconducting qubits? Oh, these squiggles over here? Yeah. No, that's a representation of which two electrons are getting mixed at a time.
Okay. So it's basically giving us our zero and one. Yeah, yeah. It's giving us our, yeah, like the zero and one is the three electrons. The three here being in either a singlet or a triplet, the three here being in either a singlet or a triplet plus whatever. And the squiggles are telling you like if there are six independent lines, that tells you that you're doing nothing. If two of the lines come together, that's telling you that you're removing the barrier in between those two lines. And so this is a quantum sort of like a simple gate operation that you're performing.
I got you. It's a two-qubit gate operation. I don't actually know what exactly the gate is that is being represented in this video. Again, if there are former colleagues in the comments, let me know what exactly this gate is. But in any case, that's the idea, right? It's the golden gate. No, I'm playing that. Yeah, so the golden electrodes are what maintain the valleys, right? And then the little blue pulses come in to make the electrons mix together.
And whenever a blue pulse comes in, those two lines sort of come together. Basically, our hotel rooms are where the pink is. The door in between the hotel rooms, in our analogy from earlier, is when they start coming together. That's when the doors open and that's being triggered by the pulses that were coming. Exactly. Having down the gold. And then we were reading out again... state based on whether they're not coming together. And then these two on the left, that's the single electron transistor on the left here. That's the thing that is sensing. That's the M1, the Z, M1, Z2.
Those are the single electron transistors. Got it. Because they can sense what's happening. It's the elevator shaft. Exactly. And then they can sense, like, when we finally want to read it out, we'll shove the two into one. If they get shoved together, then you've got to um singlet if they don't get shoved together then you've got a triplet and you can you can figure that out makes sense okay so this is this is the quantum computer right okay like that's made in silicon okay yes right it's printed in silicon yes the gate electrodes are all like printed the same way that you sort of print a chip yep and it works and it works right for two cubic gates it totally works this was in 2022 and this is right around the time that i started my job yep at hrl
I got involved about a year in into my career at HRL. I was having a great time. More on what I did specifically later. And then came early 2026. when HRL lost significant program funding. I know I can say this because Thaddeus Ladd, who's one of the corresponding authors on this paper, he went on a podcast with Sebastian Hessinger called the New Quantum Era Podcast. If you want to hear it from one of the corresponding authors on this paper, go check out that podcast.
It's not that long. This is going to be a three-hour long podcast, but that one's only 40 minutes. Yeah. And it was announced to the world in February. There's a bunch of headlines from the great journalistic powerhouses of the Malibu Times and the Los Angeles Daily News talking about how HRL had to lay off 376 employees. Malibu's HRL Labs cuts that many jobs after losing government work. So HRL was kind of in a crisis mode.
And they transitioned from working on this stuff to now we need to commercialize this. We need to bring in some outside work to commercialize this. Maybe someone's going to buy us. And that's when IBM comes in later. I want to make a quick note here about when we talk about funding conversations, there's a difference between basic research and applied research or applied technologies. And this is kind of that line that is being straddled here because – And correct me if I'm wrong here, to some extent, in its initial conception, it was still – it was going to eventually be applied somehow.
But the context was let's just do this as a basic research function and figure out how to get to stability here. However, the government is cutting basic – generally, it's cutting basic research in favor for applied research. Yeah. Which basically means where can we go invest our money and make billions of dollars? Yeah. How can we productize stuff? How can we create this golden age of American technology? That is kind of the shtick right now. And as such, IBM in the Chips Act was given a prop around this idea of we need to create re-onshore chip manufacturing.
Intel too, I think. Intel and IBM are two great American chip makers, right? And how do we compete and both? This is the background context in which this is happening, where there's a huge federal policy funding shift to putting money where it is perceived that it can boost stock prices, basically. Yeah. I mean, I don't know much about that. But that's the context.
That's the context, yeah. And it's not – it's very explicit in the documentation coming out of the executive offices, the president of the White House in terms of how they've defined – you have this project with the Department of Energy where they're doing the same idea. The idea is we want to decompress the time from research to applications. That's how it's framed, right? And application means monetization, ultimately. And how do we make this something that actually can generate GDP value?
Because they want to mimic what's been done with AI in terms of boosting growth, GDP growth in the U.S., in every other industry much more quickly. Yeah. So that's the context. Again, a separate issue and a separate topic. And my viewpoint on this, not that of Krishna's for those who are listening, who are still listening, who are his teammates. But again, it's weird. People get weirdly emotional about this subject in ways that it doesn't make sense to me because it's just – it's literally what they're saying.
Yeah. I mean, the White House is very explicit in its executive orders and its website, whitehouse.gov. You can check it out, right? So in any case, HRL loses significant program funding. And so this is when HRL decides to sort of come out with this work. And during that reduction force, I had to move on to other places. So... APS March meeting is also coming up right after this event.
And so at APS March meeting, we're all giving our talks. And we had already sort of applied to give the talks, but now there's this kind of urgency to give these talks, right? And we're not presenting the stuff that's in this paper. We all have like our own talks. And I had the last talk before the big talk of Thaddeus Ladd, who's one of the corresponding authors on this paper, right? And we're presenting stuff that's adjacent to it. At March meeting that Thaddeus' talk is going to be this big talk. Okay.
It's going to, he's going to present some like really cool results. I had the last HRL talk before Thaddeus'. And in my last slide, I snuck in a photo of some of the stuff that's shown here. And it was really funny, actually, just to give you guys kind of an insider of like, you know, how conferences work and things like that. This was a talk in the automation section. And spin qubits is quite a big field. So it was quite a crowded event. people came and in the last slide I showed like a little sneak peek of this stuff and I dropped it like it was like you know in a Nickelodeon shows that they'd have those celebrity cameos and like they'd stop and then the audience would like go crazy it was like when that happened like everybody took out their phones and started taking photographs um in the audience because this was all new stuff and I said you know
More on this. Go to Thaddeus' talk. It's going to be a banger. And a lot of people showed up. This was in a big conference room, one of these big spin-cubit exposés where you had HRL giving a talk, Intel giving a talk, Dirac, which is where I'm currently at, giving a talk. And Thaddeus goes up on stage and he drops this image. This is one of his first slides that he drops. This is... The quantum silicon.
Okay. It's got 18 qubits now. Okay. So 18 times 3, there's 54 electrons that are moving around in this thing. And this thing has enough wiring that is comparable to Google's Willow. That's crazy. That's actually crazy. Yeah. Yeah. We don't need microwave coax. That's so crazy. Right? Right. This just looks like a PC. Yeah. Yeah. Right? I mean, it's a lot of metal instead of like plastic.
Sure. That's like in the gaming PCs. But it kind of just looked like there's a printed circuit board. There's like these like... The, you know, ribbon cables that are coming in, flex cables that like connect your GPU to your monitor and things like that. It looks like a custom PC. But it just looks kind of boring. Yeah. And that's kind of the point. Right. Boring is good because boring is scalable. Okay. Now, the next photo we have, it's labeled. Okay. Although I don't know if you can see.
So on the bottom, you've got the chip itself, right? Before you had just 3 and 3. Now you've got 54 down there and a bunch of single electron transistors on the bottom. So that's your qubit chip. They call it a daughter board? Yeah, that's the daughterboard at the very bottom. At the bottom. Okay, that is being held up like this from the two sides and then the bottom. Okay. That thing is at the millikelvin temperature. Right. Okay. Yeah. And only that needs to be maintained at that temperature.
At the millikelvin temperature. That's kind of nice. That's kind of nice. That is key. That's kind of nice. Okay, but the green part that you're seeing behind it that's coming down there, that thing is actually not connected to the side and the bottom. Okay. Okay? Except... For a little tiny cable at the back there. And I don't know if you could see this, but actually let's remove the thing so I can show you. Yes. So here we've got the daughterboard on the bottom.
Yeah. You see that square that's highlighted? Yep. That's the chip. That's the quantum chip with the 54 electrons, 18 cubits. Yep. This is the 4 Kelvin stage that has something called a cryo controller, which we're going to get to. But the key thing is this bottom part is at tens of millikelvin. This top part here is at four kelvin. And the only connection between these two is this highlighted part here. A little corner pipe. Yes. All of the electronics, if this electronics wants to talk to the daughter board,
It only has to go through the superconducting ribbon cable. That's actually really nice. That's really nice because there's no thermal loading into the daughterboard. That's the key. These are separated inside a vacuum chamber. And the dilution refrigerator, the helium that's pumping, is pumping on this bottom part that is going to create that tens of millikelvin. That's actually really clever. Yeah. You've basically created like a Lincoln Tunnel of where everything needs to go through to get to New York City. But the thing is, it's like a wide Lincoln Tunnel with a lot of lanes.
Yeah. And you don't have congestion problems or anything. Yeah. It's just like if Lincoln Tunnel was actually efficient. If it was actually efficient, right? And if you go back to that photo 73. Yep. Right? So on the bottom, that's the millikelvin stage. That's got your qubits. The cryocontroller is another big part of this setup. And that's at the top there. We're going to get into what that actually does. And the way that cryocontroller controls the qubits at the bottom is through... the ribbon cable which in this in this kind of diagram we can kind of see in that smaller golden box below in the white yeah yeah it's going like behind it you can't see it because it's behind it's like outlined with like a little it's the same thing as here like it literally it's like yeah no that that makes it
that's okay so i can already i can already see now where this is going because of what we've previously laid the table for why that architectural or structural design choice is going to be meaningful in terms of our criterion and now let's let's go to the um let's go let's take a look at the actual qubits okay So this is just a bigger version of that six electrons that we had seen earlier. Now we've got 54. Again, there's wiring. But all of this wiring is...
silicon yeah you can print it like a chip we know how to do this with the lithography stuff that asml does yes as an example as an example right this is not done with the same like euv lithography but there's no reason why it wouldn't be able to right sure for sure um and so here you've got 54 um electrons that are being controlled there's like two there's like two wires in each because there's a barrier and then there's a there's a spot where it sits there is the 2d confinement of the electrons quite nice right right
Each of the electrons are sitting under these plunger gates, we call them. Those are the purples. That's a single qubit. There's three. And control pulses come in in the barrier gates. to give you oh this is nice dude this is nice yeah this is nice we we already know how to do all of the engineering of the substrate yeah the what was really missing was the like what you needed it to look like and do architecturally and then obviously the larger system around it but yeah
But this, you can see that maybe this can probably scale. I can, one, already from just, again, the, how can we control and read makes sense based on what we've talked about this, like with the potential wells, why we're doing the singlet and triplet and like the lack of global interference, but there's still local interference, which can be handled with all the things we've talked about. Yeah. But I think the ability to do this in silicon is like a huge, again, not knowing really much otherwise, seems to me to be a huge enabling layer for all of the reasons we've kind of talked about.
Yeah. And the fact that the in silica like works, like still has its own unique benefits. For the quality and control aspects that are independent of the scalability. It's not just a scalability solution. There's the chip, and then you zoom out. That goes inside the daughterboard. Looks like a computer. Yeah. And you can handle the cooling separately. And there's going to be aspects of the cooling that I want to talk about that are really cool. The Cryo controller has a lot to do with that. But all of this sits inside of a dilution refrigerator.
And there's plenty of space. For more qubits. Yeah. Each of these things is tens of nanometers. Yeah. Unlike superconducting qubits. Right. That are millimeters. Right? And so. There's plenty of space on a wafer. For more stuff. We can fit a lot into our data center than your data center. Yeah. And all it requires is like a single dilution refrigerator, or maybe two. You don't need a whole data center worth of... A data center can now have multiple quantum computers, not a single one. Which is, again, the scalability. But we'll get back to that.
Yeah. I mean, there's a lot more to talk about, right? Right. Unfortunately. No, we're going to get it. No, we're doing well. And I mean, I love talking about this stuff. So, okay. How do you keep that qubit chip isolated thermally? Right. Right. And I kind of alluded to this with the ribbon cable and so on and so forth, but let's actually get into it. So over here, we've got, as I said, on the bottom is our qubit chip with the 54 quantum dots.
That's connected via ribbon cable to this cryo controller. Right. Here's what's actually happening. The cryocontroller from room temperature, room temperature is outside of the dilution refrigerator, right? I have a computer where I'm talking to all of the electronics that's going all the way down into this stuff. Now, I don't want my room temperature wires to go all the way down to middle Kelvin stage. Right. That would be really bad. Because, again, on one end is room temperature 300 Kelvin, and on the other end is something colder than outer space.
It's going to be really hard to maintain that thing at colder than outer space. So instead... You've got this intermediary called the cryo controller assembly where you feed the cryo controller all of these communications, the digital communications, the instructions on what to put the gate. Yeah. Biases to like, what are the algorithms? What are the circuits that I want to implement? And furthermore, you don't even have to. Furthermore, you don't even have to give it. instructions on exactly what the circuit needs to be the cryo controller itself is a chip that has memory and processing yeah so you can give it a instruction on implement this it's got some like memory and processing to be able to know what signals signals to then send through the ribbon cable to our daughter board
It acts like an air traffic controller to some extent, which has people in it. It's not just like routing without intuiting it on its own. It has the ability to route, but also be like, ah, that doesn't make sense. You can compute. There's some headlines that say that it's autonomously controlled. This is kind of what they mean by that. It's like the cryo controller is given some instructions, but then it's making its own decisions based on the context. Yeah. Yeah. Okay. I like that. And because the cryo controller is this middleman, you can isolate the room temperature electronics to go to that.
And then this thing, when it feeds it through the ribbon cable that is superconducting, the amount of thermal leakage is out of minimum. Yeah, that makes sense. So we're basically splitting out the – we're splitting the journey that the instruction goes to get to its destination – We're decoupling the thermal impact from the delivery of the information. Exactly. One of the reasons why this is important is if we go back to our discussion about superconducting qubits, all of those microwave lines had to go all the way down to the chip. Yep.
There wasn't a middleman. It had to go all the way down to the chip. And if it goes all the way down to the chip, you're heating up the most delicate part, the part that is the coldest. You don't have a lot of power to cool the part that's the coldest. It's much easier to cool something that's 4 Kelvin than it is to cool something that's at tens of millikelvin. I've got too many wires, man. I don't know what to tell you. So even though there is a wiring problem... It's kind of okay. Right. At least for this many. Right? There's arguments to be made about, okay, like, if we get to millions of qubits, the ribbon cable isn't really going to work.
And we can get into that later. Okay. Right? So it's an MVP for a reason. Okay. I see paths, but I understand. Yeah. So... Now, that's all fine, but it's not going to amount to anything if you don't get all of the problems that we talked about down. Yeah. The valleys. The valleys. The charge noise. Yeah. The magnetic noise. Yeah. All of that stuff. I mean, you're scaling, but if your qubits are not good. Yeah. If the qubits are trash, if the gates are trash, if there's so much noise, you're not going to get anywhere. Yeah, you can't actually do it.
Okay. So... The fidelity has to go up. The gates, like, when I say flip this thing, it better be flipping this thing 99.9% of the time. Okay? Okay. Now, if a gate has a fidelity of 99.9%, what that really means is the error rate is 0.1%, right? It's just one minus that. So one in 1,000 operations are going to result in the wrong state. And because quantum algorithms string together a bunch of gates, it compounds. And so you really need that fidelity to be really, really high.
That makes sense. Okay? Now, this is kind of an interesting little part of... this current paper, the fidelity has gone down, gone up by a lot. The error rate has gone from that previous 2022, 2023 performance, 3.7% error down to nice 0.09%. So just under that 0.1%, right? So the performance has really, really gone up. And this makes it possible to now really start doing error correction algorithms and things like that, because now I can reliably...
do gates. Yes. Yes. Okay. And also your pace of improvement if you were to extrapolate moving forward. We've not necessarily reached the ceiling yet on that performance improvement. Actually, it's kind of interesting that you bring that point up. Okay. So how did they reduce this error? Okay. Right? What is the ceiling, for example? Yes. Well, there could be two sources to error. There could be an intrinsic and an extrinsic source. Intrinsic means it's the stuff that we've talked about. It's the charge noise. It's the magnetic noise. It's this valley nonsense. This is stuff that is intrinsic to the physics.
Extrinsic noise is stuff from the engineering, right? This is stuff like, how good are my control pulses? Are they squares? Are they calibrated correctly? How good is it to go from room temperature down to there and then from the cryocontroller through the ribbon cable down to this thing? Am I losing electricity and stuff as I move through there? So that's the extrinsic part. Okay? Yes. Here's the kicker. This is a quote from the paper. It says, Air is resulting from the mean values of,
and OSC and T2 star. These are two different measurements that you can do to qualify how good your qubit chip is. And you can use those measurements to then qualify how much of the error is because of extrinsic sources and intrinsic sources. It shows that collectively, those things only contribute to 0.02% of the absolute CNOT error, absolute error in these gates. So... The intrinsic part of things only contributes to 0.02% of the error.
This is a subtle way of them saying the physics problems have been solved. Yeah. Kind of. Yeah. Okay? And then when you go into the... Into the methods where they talk about the qubit chip, right? Because anyone who's in spin qubits is going to look at this or anyone outside who's like, how did they do that? The last sentence just says the SIGI, the silicon germanium heterostructure. It was enriched with 28 silicon and depleted with 73 germanium. It was engineered to increase valley-splitting energy.
That's it. Okay? That's kind of like Lincoln in Lincoln movie. He was like, to my knowledge, there's no Confederate negotiators in Washington, D.C. Right? And then the Confederate sympathizer in Congress was like, that doesn't mean anything. That's a lawyer's dodge. It's kind of what this feels like. Okay. Doesn't really mean anything. Right. Right. Here's to here's to underscore that point about about how they're trying to dodge this.
You know, as you can see, I'm no longer employed at HRL. So this is just stuff that is public. Right. I just read off stuff that's in the paper. It's in the paper. It's in the paper. That's all. I just I just read it off. In 2021, they showed Valley splitting splitting results in this paper. On the left-hand side, on the red, the red shows the distribution of valley-splitting energies that go anywhere from 300 microelectron volts all the way down to, like, 10 to 30. Okay?
Now, crucially, valley-splitting is a difference between energies, so it's always going to be positive. What this should remind you of is a Gaussian that's centered at zero. Okay? That they've just, like, made positive. Right? So... What this is saying is sometimes we get really good energy splitting where it's really high. And so my electrons can be localized in one state. Sometimes it's really bad. Sometimes maybe good. Sometimes maybe. Gennaro Gattuso was, I think, the gaffer at Lazio now where the transfer when this is funny.
Mussolini's great grandson now plays for Lazio, which is the same club that Gattuso is the coach at. I love that video. Like that's such a good video. It's so applicable in so many life situations. Classic meme. And this is one of them, right? The Valley splitting traditionally, it's kind of a crap shoot. Yeah. Okay. Okay. Um, in this paper, the one that's in nature in the supplement, they have like a 70 page supplement that they've uploaded. Um, cause there was a lot of work that was done. Um, Let's look at this latest result. It shows a valley splitting of 1.8 millielectron volts.
The other one was 300, so that would be 0.3. We're going from 0.3, which was the best that they had, to now 1.8 and 1. They're saying typical values of the valley splitting. Crucially, they don't show a distribution. What they say is that given our error rate is so low, you can tell that it's mostly... going to be around these values. Okay. Okay. It is no longer a problem. Yeah. And this is highlighted by the reviewers. Again, this is a nature paper. Right.
Right. So you can read the reviewer report. Yeah. And the reviewer says that the author state in a certain line that the conventional concerns such as valley splitting are no longer limiting. Attributing this to quantum well, which is engineered to increase valley splitting. Okay. Okay. And then he says, if confirmed, this would represent a significant advance for electron spin qubits. Okay? If this is true, this is a big deal. Why is it shoved in the supplement? It's kind of what he's saying. Right? And then he says, the current evidence, he says, the current evidence is not enough.
And you need to show, like, you know, I would expect, for example, a statistical analysis of valley splitting across devices. Right. So it's like, maybe give me a distribution. Right. You know, he's basically saying, like, how do you do that? Right. And part of the argument might be, well, now that we're going private, it's partly proprietary. And that's exactly the response. The response from HRL is, I'm not telling you. Because that's kind of the juice. Yeah. I mean, that is the juice. Yeah. Yeah. It certainly shows that it's possible, which is great for the spin community in general.
Right. Right. Because it's not an insurmountable, like... 2 plus 2 equals 5 problem. It is certainly possible. And the response there to the reviewers is, we have instead substantiated our claim that value splitting is not an impediment with circumstantial evidence like the lack of multiple frequencies in your exchange oscillations. Basically, all of these other little things are saying that if value splitting was still a problem, I would notice it in all of these other plots. There's derivative measurements you make that identify it, whether it's there or not.
Yeah, and it's clearly not there. It's not there. And unfortunately, the proprietary nature of that information prevents a complete discussion of those aspects. We got them. We got them. It's kind of interesting, right? It's like another aspect. It sometimes goes under the radar. Just putting it in the supplement. But the reviewer noticed and he brought it up. And that's like a real – I mean based on the history we've talked about, right? That's been the historical reason why the neutral items people or whoever, the trapped items people have been like, this is not going to happen.
Because you have to tell me how when we have naturally occurring silicon between isotope 28 and 29 and they have these six valleys – which means that you are going to have this exponentially exploding calculation you have to do because you can't control where your thingy is the way you want it to. That's like you have to solve that problem before I'm going to pay attention. And they just kind of were like, yeah, we solved it. But that's in the supplement. It's in the supplement, right.
And it's funny because the isotopic enrichment part, that's an open secret. That's how you do it. Okay. Right? It's just like... Yeah, that's a fundamental physics argument. Sure. You remove a bunch of magnetic spins from your substrate, and then you're no longer going to have the valley splitting part. But that's different. Right? Yeah. It's a little different. It's a little different. Right. But that's proprietary. Right? But it certainly means that it's possible. There's a chance. And that, I think, spells great things for the spin qubits community in general.
Speaking of proprietary, we can get to our... Our audit. Our audit. Which is also proprietary from First Principles audit. So what are the strengths for silicon? Well, silicon is a miracle material, right? I mean, now that if you isotopically enrich it, you don't... Apparently, the valley splitting is no longer a thing. The negatives are no longer a thing. Charge noise is no longer a thing. If you fab it in a really nice way, like HRL has done, it's no longer a thing.
So seems to me like qubit quality... Pretty high. Quite nice. Quite nice. Let's go into control. You see how I'm just breezing through these? Yeah. Pretty nice. We've already explained it. I already explained it. In great three and a half hour long detail. So it's very well established. The cubic quality here is pretty great. Spin qubits, DFS in this case, even with Lost Vincenzo, where you have the magnetic field and a single microwave thing.
Still, It's just the spins going up and down. You can pack them together. You can use Exchange, which is DC only, to make them talk. It's fundamental. It's great. Back to basics. Now, for control. Control. The gate speeds are really fast. Exchange-only gates execute in 1 to 10 nanoseconds. So the same stuff that we liked about superconducting circuits, which is the fact that I can run shores pretty quickly in a matter of hours rather than a month or a year, I can still do that here, right? Beautiful. As long as I have enough qubits to go around.
They're actually faster in some cases with superconducting circuits. And they use only baseband circuits. The gates are all baseband. It's all DC. Yeah, right. There's no AC. So there's no AC leakage everywhere. And I can, you know, isolate it from the system. Although that'll get into scalability part. So let me not go there. Now, what are the negatives with control? The negatives, there is one. And that's where my job comes in. Aha! Okay? If you've made it this long, you now will know what Krishna does.
Finally, you'll know what I do in my life when I'm not in this podcast studio. So there's a skeleton in the closet here, which is I'm fabricating stuff at the limit of... fabrication. There's these gates, these wires are tens of nanometers. They have to be done with this isotopically rich silicon. And when I fabricate, you know, at the nanometer scale for transistors, If a single transistor has a few extra atoms, not a big deal.
No big deal. Not a big deal. Here, it is a big deal. Because if a single gate has a few extra atoms, is slightly bigger, slightly smaller, if there's a random charge noise hanging around, then the same settings that I used for one set... Let's say I plus 0.4, minus 0.5, plus 0.652, whatever. I can't use those same settings everywhere in my device because every single one of these electrodes, because of the fab disorder, there's going to be inherent noise in how that happens, right?
So you would have to tune up this device. For every single qubit, you'd have to figure out exactly the right settings to get three isolated, to understand how long I have to wait to make an exchange gate, to understand what is the voltage, what is the current when there are two. When I do that polyspin blockade, when I do the readout, where I'm shoving two electrons into a single thing, how do I actually do that?
How long do I have to wait? What corresponds to two electrons being in there versus one? Because I'm reading a current, right? And so, I mean, you could have a PhD physicist do this by hand, which, fine, for like small enough for the 2023 paper, where there's like six electrons, fine, you could do that. Pretty soon, if you want to scale this thing, it is going to get insane, right? It would take a human lifetime to tune up a million qubits by hand. Yeah. Right?
Yeah. And so you are the qubit tuner. Yes, I am part of the qubit tuning team. I used to be part of the qubit tuning team at HRL. And that's still something that I do at Duroc now. QTT. So here's how it actually works. I'll just show you some of the supplement of this paper that goes into that kind of stuff. The first thing you do is you make a... Dot charge sensor is what they call it. This is your single electron transistor. So you initiate the single electron transistor. That way you can sense what the hell is happening in my device.
Then I start loading my dots. Before, there's no electrons. And then now I want to load exactly one electron into each, or in some cases, an odd number of electrons into each. That way the electrons pair up, and only the thing that is missing is going to be talking and doing the stuff. In order to do that, what I do is I toggle the voltages. And when I toggle the voltages, the electrons are going to tunnel and go bloop into one, bloop into another. And every time that happens, my single electron transistor is going to move
And you see on the bottom, the C and D? Yes. Those are images that show these tunneling events. Like every time an electron is moving from one room to another, the charge sensor is going to move. And so my current is going to be a different. color it's going to be a different value that's going to manifest as a different color and so the two axes are my voltages like i'm scanning this voltage and i'm scanning this voltage and there's going to be an event that's going to show up as a streak and those streaks are going to tell me what the settings are that i need to implement
Finally, once I have the correct settings, then I go into creating my polysplenball blockade, my readout, where I shove two electrons into a single confinement well to see if there's two electrons, then I get one... value of current. That's the histogram on the right-hand side. If there's one electron, then I get another value of current. And so those two then tell me, hey, if I got this current, that means I'm a one. If I got this current, that means I'm a zero. And then finally, I then tune the other axes to get full blocks for your control. So this is the sort of
path that you would have to take for every single qubit in order to make it work. Right. And each of them have their own subsequent sub steps. Like, that's a very brief. That's a very brief. But this whole thing is going to take like two hours for like a human being. Okay. Okay. Ain't nobody got time for that. Right. Right. Right. For each qubit. Yeah. For each qubit. Right. Yeah. So you use AI. Yeah, of course. Of course. Of course. Of course. Especially because like the bottom part where you're like finding the edges and you're finding the like, it's just an image.
Right. That I'm looking for streaks. Right. Right. Object detection. The computers have vision now. Yeah. Yes, exactly. Computer. So that's kind of some of the stuff that I worked on with the automation team. Makes sense. This is the machine learning sort of roadmap on how to get from the scan, where you get these tunneling events, all the way to, okay, what are the voltages that create those tunneling events? And then what are the settings that I need to create? This is a DETR.
detection transformer. So it's a... hybrid between a convolutional neural network, like a ResNet that actually does most of the object detection and stuff. But then you feed that stuff into a transformer, then then uses attention to then create like relationships between all of these lines to tell me where the stuff is. God, this is so good. Okay. So now you actually finally have a good idea. This chart, I understand. This one makes sense.
But I think it's a very, what's really cool about, or important about, not important. Yeah. What's interesting about this is that of everything we talked about, To get to this point, every aspect of it is this complex. And we're just talking about a sliver, a slice. That's a very good point. Every single aspect, which is why there's 250 authors. Because everything is so – you have to be so deeply understanding of the physics aspects of it and then whatever the applied avenue that you're dealing with.
Are you a materials person? Are you working in algorithms and computation at the chip level? Are you working on those at the – the external are you are you building the software that lets me right like do this do this actually because it's just because you have the chip doesn't mean you can do anything with it there's so many and every one of them has a depth of knowledge and execution necessary that i think people really can underestimate when we don't work in these things and we sort of will go through a paper and we'll explain it and it's like yeah we're taking away the things but like
A lot of people have worked very, very, very, very hard in an area that very few people can even work in. Yeah. To accomplish that. So that's like – because the tuning piece, again, it's one sliver. But it's a very important – Yeah. I mean it's part of the scalability argument. Right. Right. Right. Of the three layers of the criterion. Anyway, I'm just trying to give you props. But like that – it's dude this is so guys i'm really actually just like on a personal note i'm very happy that we did this because now my podcast partner actually knows what i do on a date this is as a day job right and like now i know when we go out and we're people i'm like you know what he does and i'll be able to give him the spiel yeah exactly
So that was the control part. Okay. And although towards the end, we kind of got into scalability. Yeah. Which is like, if we want to build a bigger thing, you better be able to tune that thing autonomously. Right? You don't want a human being there sitting there like a monkey doing this. 100%. And so finally, we get into scalability and economics. And this is the kill shot for silicon. If you want to build 100,000-cubit trapped... Ion machine, you're going to have to invent a hundred thousand laser optical miracle. But if you want a million superconducting qubits, you're going to have to build a warehouse sized priostat.
Boo. But if you want a million spin qubits... You just put it on a standard 300 millimeter silicon wafer. You run it through the exact same photolithography machines at TSMC, ASML, Intel, that made the processor in your iPhone. And there you'll have it. Right. Scaling is the key. Yeah. And scaling is the future. Yeah. HRL is not the only one that is pursuing this stuff. So I want to give a shout out to some of the other players in the space.
Intel had a really amazing talk after Thaddeus' talk at the APS March meeting that showed uniformity in a lot of their fab. I mean, they're really good at fab, right? And they could make their chips along with like 16 metal layers of back end, which is kind of crazy. And they had a roadmap to scale up to larger qubits. So there's already scaling to larger numbers of qubits coming in from Intel. And then, of course, there is Dirac, which is where I am at right now.
This is an Australian startup that is trying to make millions of spin qubits on a single silicon chip. It's a great place to work, to be honest. And Dirac also publishes in Nature on the regular. So on the next slide, we've got just a few of the ones that came out in the last two years. um spin qubit control with um a millikelvin cmos chip so they're getting into the cryo cmos type of stuff um they've got really high fidelity they're getting into that and they've got high fidelity spin qubit operation and initialization above one kelvin so they're trying to do things without going even into the millikelvin stage it's an open problem right like do you even need um
You could do LD qubits, Lost D Vincenzo qubits, but the energy split is big enough that even at one Kelvin, the temperature is lower than that energy split. So spin qubits are really making a resurgence, I think. And... I think you don't have to invent a supply chain is the point. Yeah. The supply chain already exists. We're piggybacking off of 60 years of Moore's law and trillions of dollars of CMOS and silicon manufacturing infrastructure.
there's a reason why your laptop is a miracle of engineering, and yet it's still actually kind of cheap. Yeah. If you think about it, like people who build computers, they have to do the same drastic, like the people who are at Apple and like Dell, they have the same level of like crazy specific knowledge, right? But they're making laptops on the Chief because they can scale, right? 100%. And the point is that when manufacturing that first wafer that's going to get you that first silicon quantum computer at scale that is fault tolerant and all that other stuff, that first wafer is going to be hella expensive.
Yeah, yeah. Okay? And it's still not there yet. Yep. But once you get that one wafer... But once you get that one wafer, the second wafer... It's going to be a joke. The path of acceleration is going to be so much faster if you can do it on a silicon wafer. Yeah. Because we already, it's already all there, which is, I can't overstate that as a part of this. Yeah. Um, It's part of the reason why the acceleration of AI has been so quick because NVIDIA has already been building GPUs, which were the substrate that's really good for a lot of these transformer algorithms, the matrix multiplication.
And so imagine that we did not have GPUs at the level that NVIDIA was today. Good point. Right? We would not have seen the acceleration of AI that quickly. And it wouldn't have been that big a deal. But they already – it's global. Mm-hmm. It's already been – they have the distribution partners. Anyway, you get the idea. So that's how I make the connection here. As someone coming from more of the technology and like operational context in the real world, it's like as soon as you cross that Rubicon, again, there's going to be the battle at the chip makers with their other business lines for time.
Yeah. But, you know, a quantum chip... It's going to be hard to argue why that shouldn't be at the front of the line. Exactly. And I mean, there's a... So IBM has its own chip-making facilities, which is, I think, one of the reasons why it probably acquired HRL. It also probably... I'm not going to speak for IBM, but maybe I am a little, in saying that they may see the writing on the wall when it comes to superconducting qubits. I mean, there was... You know, at APS, actually, I should mention that Northrop Grumman...
at a talk where they showed they used the sort of same trick that these guys are using with Exchange Only, where they had superconducting qubits, but you could rig them up to create qubits where there were single triplets made out of transmons. But now I only have to communicate with them using DC. Now, that's all fine, but it's still not in silicon, right? And it's still massive, still big. So you still have that scaling problem, even though maybe you don't have the wiring problem anymore. Yeah. there's still an advantage in using silicon um and because silicon is just the bread and butter of our economy right it is the best um so finally we do have our ffp audit that we are going to um put on spins the last overlay of the day yeah so uh qubit quality 10 out of 10 yep easily electron spins are dope yep
whether it's lost even Chenzo or exchange only cubit control again, 10 out of 10, you're using maybe a single microwave that's going through for lost even Chenzo. And then your baseband pulses or just only baseband pulses with exchange only again, all DC amazing. If you can, if you can handle the isotopic enrichment and you can handle the Valley States like HRL has, you're good to go with 10 out of 10, which is now totally solvable. So yeah, That's great to know. And scalability and economics, 10,000 out of 10.
It's obvious. The total is 10,020 out of 30. Just to bring it up again, neutral atoms was at negative 3,000 in our proprietary criteria. Just saying to the German guy who is not listening, but if somebody knows the Germans who work on neutral atoms, please send this to them because I really hope they see all the comments that are coming in. But silicon is key. And so... There's a reason why other modalities in classical computing like vacuum tubes and electric relays are in the museums.
And it's because the transistor took over. And so I firmly believe that even though silicon might not be the first... quantum computer to achieve fault tolerance. And there might be a world where ions get there first or superconducting computers get there first or even neutral atoms get there first. I don't think it is going to be the future quantum computer. Once it's achieved, I think silicon is going to be the one that actually gets scaled because it's cheap and it's easy to make. We've seen this happen. Perfect example. MySpace was first, but we still all use all.
facebook related phase because being first doesn't mean being what becomes the standard yeah exactly and now you can finally understand the ai generated song that was in the beginning of this podcast um it i don't have the backstory for it but if somebody has the backstory for it from the group chat it was just posted in the group chat and i was like this is dope i'm gonna use this um a few points of note on that song um it says exchange only cubits Because I think the AI doesn't know how to pronounce it. And also Scalable Quantum.
Scalable. Scalable. So you'll hear that. It's a bonix. It's fine. You said it, not me. That's true. That is true. And before I end my segment, I'd just like to thank some of my co-authors. Now there's 250 of them. I'm not going to read out every single one. But there are a few that I will name that I had personal relations with at HRL and continue to have personal relations with. Today, first, the three corresponding authors, Thaddeus Ladd, Jacob Blumhoff, and Matt Reed. Thaddeus Ladd very generously sold me one of his extra copies of the Nature Physical.
He bought like 10 of them. I think he's very excited that this is finally seeing the light of day, obviously. I hope you made it this far. He's also one of the people responsible for actually getting me to APS because the deadline was on a Friday. Yeah. I didn't know because I don't keep track of these things. I thought somebody would tell me that he's told me. He emailed everyone like, hey, is everyone like signed up and like taken care of? And I'm like, what? And then it's like, dude, there's an hour left. And I'm like scrambling. And he helped me figure out like what session to be in and things like that.
So thanks for that. Also, like the nature of... Getting physical copies of nature is like a mom and pop shop. Yeah. Yeah. Like you got to, you got to like email someone. Then they email you back being like, Hey, what did you want? And then you're like, I want a copy. And then they email you back a form being like, okay, fill out this form manually. And then they send it. And a lot of my coauthors who've got these, they got water damage. Oh, wow. Like I didn't even know mail got water damage these days, but somehow like he got 10 of these copies. Eight of them had water damage.
This one doesn't. This one's in good condition. This one is in great. And we're going to encase it in a cryostat so that it remains in good quality because this is a big deal. Yeah. Yeah. So a few other people. Patrick Harrington, congratulations on being a new father. um cliff plesha thank you for uh sending me some of the videos from ap s so i could relive the the big talks tina garcia um thomas harris and kevin he these guys uh these guys kept my fridge cold at hrl for me so i didn't have to go and refill the liquid nitrogen every week that was pretty great um
Cameron Jennings, Paul Jerger, Stephen Carr, John Carpenter, who's the artist behind this, Faustin Carter, Matthew Borselli, who's been working on SpinCubits at HRL for a very, very long time. So congratulations on this paper. Kevin Chen, Adam Daly. Adam Daly taught me everything that I know about spin. dilution refrigerators and taught me how to like operate one keep it cold that was pretty great jp dotson galen gladhill who is a co-worker of mine now at durac so i see him every week and that's pretty fun aaron mitch jones um really great in like bringing me into the fold with the um hrl quantum project raj kati joseph kirkoff justin christensen who was a classmate of mine at ucla um
um phd and then we both ended up at hrl andrew pan matthew raker matthew raker had a great um like cubits class where he actually taught me from fundamentals from for its principles all right the exchange cubit worked things like that he was he was very patient with me that was this pod in future classes yeah uh rashan sajad christian snibel congrats on your second kid bosun Skylar Turner and Alan Sinanian, who are both now colleagues of mine at Dirac.
Aaron Weinstein. Aaron Weinstein was the first author of the 2023 paper, the Universal Logic one. So props on that. And then Adam Holmes. And finally, the automation team that I was a part of. um alwina lu she's a big fan of the podcast i hope you've listened all the way to where uh you can now hear your name i hope the sound is fixed i know you've been complaining about that let us know yeah let us know if it's fixed for you um ian jenkins joe kern um john maeda uh last no not last but uh definitely least i have to say sam mumford
We have a long history, this guy. He was a classmate of mine at Princeton. Okay. So he lived down the hall from me in 1941. Oh, really? Yes, former Wilson College, then first college, then it got demolished, and now it's called Hobson College. I don't know if 1941 Hall is going to make it back, but he lived. down the street from me that's i'm not down the street literally down the hallway um uh so in freshman year so he was he was one of one of my first friends at princeton we took um physics classes together and he was in physics with me um he no you don't know him he never hung out at ivy he would i went around other places than ivy don't you don't you throw dirt on my name
he'd be hanging out with the uh with the engineers at charter um but he you know jokes aside very instrumental in bringing me up into the fold he was uh you know we're old friends and i hope you're listening to this one episode i know you don't like listening to it um but whatever you know it is what it is um so sam sam mumford and last um definitely not least um maybe most is theresa brecht Long time fan of the pod. Number one fan.
A patron of the podcast arts. But more importantly, from a personal level, she was my boss at HRL. And she was the best boss that I've ever had. And she also took a chance on me. with getting me into the spin qubits program because i started at hrl as um just a machine learning guy right and i was in the intelligence systems lab and i was telling you that like i was working on like these tiny bespoke machine learning projects wasn't really you know fitting my um ambition levels and like interest levels and then the quantum silicon team was um
in need of like machine learning people to do the automation stuff that we had just talked about. And I joined that group. And then finally, when, when time came, I asked her, he sort of like formally put me into the HRL quantum team, which she did. She gave me a chance. Um, and I ran with it. And so because of her, really, I'm on this, I'm on this paper. So thank you, Teresa for that. And that will end my three and a half hour tirade about, um, a paper that I am on the cover of nature for. So, you know,
It's tough, but I don't really regret spending three and a half hours because this is kind of a dream come true. It's monumental. It's on the paper. It's on the cover of Nature, dude. Yeah. even like the the you know i didn't write a single word of the paper but i was involved in the science and that's something that i think we can all be proud of all 250 of us that's a really big deal congratulations again thank you i i will just note for those who are tracking the information security on the project i've had no idea what krishna did for work for the longest time because it was classified
And that was so annoying, but I knew when the time would come, I would get to learn what it is you did because I knew you were so passionate about it and you were excited. And to be able to sit through both parts of this and really understand at a real level kind of what it is you've been working on in the world you've been living in for so long and continue to do so. Yeah. It has just been such a blessing. And just congratulations to our resident PhD. We have already kept you all long enough, so I will not pontificate any longer.
If you can share a like, share a... If you are still here... And you're trying to listen to the song. Let us know. Let us know. Seriously. Just say, I'm still here. That's going to be mind-blowing. Anything. This was meant to be two parts. We didn't want to make it three parts, so we just said we're going to do it. A like, a share, a comment, a five-star can really do a lot for us to get this podcast to more people. We are the best science show on the planet. People want us to make it shorter, but we can't because the science is just so good.
There's a lot of people who don't want to make it shorter. There is. And clearly, we're not. As always, I am Lester Nari, joined by my co-host in our resident PhD and recent cover, story, co-author... Krishna Chowdhury, our resident PhD. If you are here this episode, you now really understand how quantum computing works. What is the layout of the frontier? I feel very educated. I am now ready to allow my brain to decompress because that was some deep stuff.
We will see you all for a more chilled and relaxing and giving our resident PhD some time to decompress episode next week. long. It'll be a little bit more fun. We're going to figure out what we're going to do, but this was really fantastic. We'll see you all next week. HRL research group, late night grind. Lots of quantum dots, keep the layout fine. Scaling it up while the losses stay low. Prior quantum processing unit in the snow.
54 exchange, couple quantum dots in a row. Tiny little pulses make the logic all flow. Distance 5, repetition cold. Lock it in tight. Distance 2, error detecting. Catch it mid-flight. Superconducting with... We'll be right back.
New result dropping from the fresh QPU. Order of magnitude, different kind of view. Lots of new developments humming in tune. Automated tune up, calibrate the room. Large valley splitting how they pull that off. Tadis Lad leading while the numbers talk. HL research group steady on the path. Stacking every code in the low temp app. Super conducting ribbon camp. Kelvin Dream. Low noise tunnel where the waveforms glide. New kind of signal with the chill in the ride.
Do pigeons navigate with their livers? A surprising study proposes that iron-rich immune cells help homing pigeons sense Earth's magnetic field. But other researchers argue that the mechanism does not hold up. Lester...
How can a weather satellite see Ramadan, Christmas and a blackout? Dr. Miguel Román joins Krishna Choudhary to explain NASA's Black Marble: nightly maps of the lights on Earth, from a single band on the VIIRS...
Can sound destroy a tumor without an incision? Dr. Zhen Xu joins Krishna Choudhary to explain histotripsy: focused ultrasound that creates tiny bubbles to mechanically break apart tissue, without relying on heat. From a...
What happens when AI moves beyond the chatbot and into the laboratory? From Claude-designed proteins that bind scorpion toxins to robots that assemble optics experiments, we examine what AI can do for science and what...
Can light restore aspects of sight? Can recalling a cold experience change an animal's metabolism? And what can ghost particles and mirror-image molecules tell us about the universe? Lester Nare and Krishna Choudhary...