Prof. John M. Martinis of the Martinis Group for Josephson Junction Quantum Computing at University of California, Santa Barbara, calls it "+Hdirt ... the Hamiltonian of dirt." That's his term for the nitty-gritty physics that comes up in his group's engineering of practical devices capable of maintaining coherent states of duration and quality sufficient for the needs of working quantum computers. Prof. Martinis spoke from his office in Santa Barbara.
- Transitioning to Multicore Development
- Crime Prediction and Prevention: A Safer Public through Advanced Analytics
JW: I was watching your awesome presentation Decoherence from Two-Level States. I've spoken with many theoreticians and it's good to see a lot of heavy engineering!
JM: My focus on superconduciting qubits has been a little bit more in the engineering and materials aspect than many people in the field. I'm a physicist, but I enjoy engineering. We've thought about these issues a lot and have worked hard to improve our qubit.
JW: You toss off a few interesting asides in your presentation, for instance, the iswap gate, describing it as a universal gate, a component of more complex gates like the cnot .
JM: If you have two qubits, the swap gate will turn |01> -> |01> , |00> -> |00> , |11> -> |11> , |01> -> |10> . The iswap gate puts an i phase factor, so |01> -> i|10> , etc.
JW: Is that the universal gate, like a NAND is in transistor boolean logic?
JM: That's an interesting question. If you talk to the mathematicians and the computer engineers, even a lot of physicists and theoreticians, they will say the cnot is like the AND gate. My point as an experimentalist developing real phyiscal systems is that the iswap is a bit more generic, more of a pratical universal gate.
JW: What is an iswap gate physically? Is it a Josephson junction? Is it two or more?
JM: Think about the larger context of any kind of qubit. If you put two qubits on resonance with each other so they have same oscillation frequency, the same energy level, then almost any interaction will cause a photon of energy to exchange between one qubit and the other. Generically what happens is when you get this photon of exchange, you have an iswap gate of which you can make everything.
JW: You do this in the lab, and it requires a certain number of physical devices ...
JM: You have to have your qubit. In our case, to make the iswap gate we have two Jospehson junction qubits and we couple them together, a small capacitance between the qubits, and that will cause the photon to exchange between the two qubits and you get the iswap from which you can make cnot in other entangling gates.
JW: And in a useful device this information would be maintained in that realm, or would it be exchanged back and forth with conventional computing and then the state reinitialized for the next computation?
JM: This iswap gate, along with single qubit gates, you can make into a cnot, and then from the cnot with other single qubit gates, you can build any arbitrary quantum logic. With these gates you can then couple them to classical logic to do any kind of computation. How you connect it depends on the algorithm.
JW: Whether there are very many useful quantum algorithms is subject at the present time to debate.
JM: We're learning how to build small circuits. We're focussing on the underpinnings of quantum logic, the hardware issues, how to build better materials, how to make our qubits have longer memory times.
JW: With these circuits one could build a general-purpose quantum CPU.
JM: Various people have done certain CPU kinds of tasks. We're trying to do it. Other people are trying many different ways. Just within the superconducting qubit community, there are different ways to do it. We use each other's work and share the concepts that are developed in different implementation. We're all working on our own particular hardware and trying to get that to work right. It's complex to figure out that way, but it also makes it interesting.
JW: What do you think of other major approaches to quantum computing such as ion-trap and photon qc? You obviously chose superconducting qc for certain reasons ...
JM: That's because one has a certain kind of expertise. In my case it was superconductors, seeing that we could do quantum computing and make some progress.
Ion traps are an extremely powerful and interesting direction. Most would agree that they are the leaders in the field right now. I have a lot of respect and admiration for what people are doing there.
But I think that superconductors have a unique advantage compared to other approaches, which is why I'm particularly interested in that approach.
There's a problem in quantum computing in that you have to have a unique coherence of the state, a good memory of the qubit state over a long time.
JW: A long time in this instance is "about 500 nsec".
JM: That's much longer than it take to do an operation. You typically want a thousand or 10,000 operations in a coherence climate. You need long qubit memory, but you also want to manipulate the qubit state. As you try to manipulate the state by applying control fields, you need those control fields to strongly interact with the qubit or you are not going to be able to manipulate it. If you want strong manipulation, whatever you do allows the qubit to lose its memory by dissipating energy to other modes. Decoherence and control work against each other.
The example that I give is that if all we cared about was qubit coherence, then we would make our qubits out of neutrinos, because neutrinos don't interact with anything!
JW: Like keeping your computer safe by never turning it on.
JM: Exactly. You need both long coherence time, but you have to be able to control it. It particularly becomes a problem when you try to connect two qubits together. Then you want them to interact strongly to make a fast gate, faster than the coherence time, but that strong interaction means it can connect to other modes and lose its memory.
JW: Is there any payback analogous to what there is in DRAM, where an operation refreshes the decay time of the memory?
JM: Generally not until you start talking about quantum error correction. You don't need infinite coherence, just enough so you can do what you said, and that comes through a quantum error correction protocol.
Decoherence is always an issue. There are ways to operate the qubits to minimize decoherence. By complicated algorithms we can correct these errors, but they have to be small to start with. It's like correcting classical computing errors, but more complex because these are quantum errors. But if your communication channels are too noisy, then you'll never be able to correct.
JW: The basic priniciples of engineering apply to quantum computing like anything else.
JM: That's right. We're trying to make quantum gates that have low enough error rates to allow us to establish and perform correction algorithms.
JW: And all this precedes beginning to link these together to make a CPU.
JM: People are running simple algorithms to test how good the coherence and functionality is.
The point is that you have a balance between how much coherence you have and how much control you have over the qubit. The advantage of superconductors is that it's easy to get the qubits to interact with one another.
JW: Compared to ion traps?
JM: The ion traps have very long coherence times, the qubits can interact, it's a little bit slow, but fast enough to do real operations. As soon as you try to get them close to interact with one another, various decoherence mechanisms pop up and things start to go wrong, and you have to do various things to make this work.
JW: So you think with superconduction this is a more tractable engineering problem?
JM: In the end, we are building microwave integrated circuits which are fabricated on wafer in the same way conventional integration is done. We lay down qubits and wires between the qubits and get the interactions we design. We can scale up.
Right now people are working with coupled qubits, trying to manipulate and measure single quantum systems, single atoms. We can do that pretty well. In the end you want to connect tens or thousands or millions of qubits together. You're probably going to use some integrated circuit fabrication technology. The superconducting qubits are completely compatible with the technology of the day.
They're building ion-trap chips to do the very same thing. I certainly don't want to say they can't build ion-trap integrated circuits. They're working on it, the technology looks very promising. But the problems and challenges of ion traps are different than those of the superconducting qubit.
The difference is interesting. You just have to do the experiments and see what nature and technology give you. It's unclear to me which approach will be the winner.
JW: When will you have a CPU?
JM: There's a group at Yale who wrote a recent paper who built a two-qubit superconducting processor and ran the Grover algorithm on it. In some sense people have already done something. You can do only a little bit with two qubits.
When will we have it? In five or ten years we'll hopefully scale up to ten or twenty qubits and then go on from there.
There a nice quote from Bill Phillips that the chances for building a quantum computer are 50/50: fifty percent chance in fifty years!
It's very hard. It's going to take a long time to get the engineering right. Maybe if we can solve some of the materials issues, maybe in twenty years we'll be building things that are kind of interesting, not thousands and thousands of qubits, but enough algorithms and error correction to show we're on the right path.
We're making progress. As long as we're making progress, people are optimistic. That's what you see on all fronts. Someone may come up with a fantastic idea, a game changer, and it might be a lot easier. It happens in physics all the time.