The “Colossus” Qubit
In Marvel’s X-Men franchise, “Colossus” is a character who is described as being nearly indestructible. There are some caveats to his invulnerability, thanks to the attributes of other characters, but there is nothing you or I could do in the real-world to physically damage such a character, if he were to actually exist.
That’s how I felt while playing with Alice & Bob’s Boson 4.
What’s a Cat Qubit?
Qubits experience two classes of errors: bit-flip errors and phase-flip errors. A cat qubit is a superconducting qubit that is designed to exponentially suppress bit-flip errors while allowing linear growth in phase-flip errors as a trade-off. The idea is that quantum error correction codes can essentially ignore bit-flip errors – aside from redundancy – and focus only on phase-flip errors. As a consequence of this, far less error correction overhead should be required as compared to the significant overhead associated with correcting both types of errors. According to one paper, the overhead could be reduced up to 200X.
Playtime
Superconducting qubit lifetimes are measured in microseconds, so I started there. I initialized the qubit in the |1> state, set a delay in microseconds, and then measured it. I went from 10s of microseconds to 100s of microseconds to milliseconds to 10s of milliseconds to 100s of milliseconds to seconds. Like the Colossus image above, Boson 4 seems to shrug off bit-flip errors and return perfect results. In the microsecond range, if you run 20,000 shots, you might catch a few. But if you run 100 shots with 1-second delays and the maximum 16 photons, for example, you can expect a perfect result.
Technical note: with cat qubits, the |0> and |1> states are both excited states, so the z measurement decays to 0.5, not to 0.
Record-Breaking
Alice & Bob recently announced that the bit-flip lifetime of Boson 4 is actually 430 seconds: longer than 7 minutes and longer than any other superconducting qubit. The maximum runtime is set to 15 minutes, so you can’t run more than 2 shots of this at a time. But if you have the patience to try this, and to submit it over and over again, it should take a really, really long time before you encounter even a single bit-flip error.
Usage note: the 15-minute runtime limit serves two purposes; it prevents a single user from blocking access to everyone else, and it also protects users from incurring charges during their 1-hour of free usage.
Limitations
Despite the impressive achievement, Boson 4 has several limitations. First of all, it only has one qubit and you can only apply Z gates, so you can’t yet run algorithms. All you can really do, at the moment, is test the bit-flip lifetime, the phase-flip lifetime, and Z gate errors. Second, the bit-flip lifetime is so long that it takes incredible patience to test it. If you run 1,000 shots with a 1-second delay, that’s 1,000 seconds, which is almost 17 minutes, which is more than the 15 minutes you’re allotted. No matter how you set the delay and the shot count, you’re essentially waiting 15 minutes just to get another perfect result. Third, the linear growth in phase-flips is evident, so x measurements will be far from perfect until qubit quality can be improved and error correction implemented.
Availability
You don’t have to take my word for it. For that matter, you don’t have to take their word for it, either. Alice & Bob has made Boson 4 available through Google Cloud Marketplace. The first hour is free, so you can verify all this for yourself. Usage beyond one hour is reasonably priced.
The Road Ahead
Boson 4 demonstrates a 120,000X improvement over Boson 1. Their short-term goal is 100 logical qubits with a 10-8 error rate, which is more than halfway met by this bit-flip record; a bit-flip lifetime of 13 minutes is required. If you’re wondering about the name, the next generation chip will be called Hydrogen, and the next generation chip after that will be called Helium. The goal of Hydrogen will be to detect errors, and then the goal of Helium will be to correct phase-flip errors.
Conclusion
Even without correcting phase-flip errors, it would be interesting to explore what might be done simply by adding a few qubits and some gates. Then there’s the question of single-qubit (besides Z) and multi-qubit gate errors, which could be significant, but the lifetime of Boson 4 is nevertheless encouraging. There are superconducting quantum computers on the cloud that cannot perform computation for more than a few microseconds. At a minimum, Boson 4 has the lifetime to run certain deep circuits that are well beyond the lifetimes of any other superconducting quantum computers.
