One of the comments on the flux qubit post asked an important question: where does the decoherence come from? I dealt with this a bit in the thread itself, but this post will be a less technical treatment.
In general, decoherence is a result of the fact that the qubit under study isn’t in isolation, but interacts with some larger environment. Through this interaction, information that starts out concentrated in the qubit dissipates out into the environment, and likewise information in the environment mixes into the qubit. Of course, the state of the environment isn’t known beforehand so the information that mixes in just looks random, and averages out over a large number of experiments.
In the case of our qubit, what matters is the electromagnetic environment—the electric and magnetic fields that act on the qubit. Any fluctuations in these fields can produce decoherence, and just about everything produces some level of field noise.
All resistors, for example, produce Johnson noise, fluctuations in the voltage across the terminals of the resistor. Any voltage across a resistor produces a current, and the noise currents generate fluctuating magnetic fields nearby. These currents are proportional to the square root of the temperature and inversely proportional to the square root of resistance, so we can mitigate these by keeping the system at low temperature (hence, the 0.04 Kelvin operating point) and using larger resistors when possible. Even better is to avoid resistors entirely and use superconductors, as they are immune to Johnson noise, but this is not always practical.
Electromagnetic waves are another source of noise: especially in a physics lab, the air is full of radio waves being generated by various electronics, and our qubit operates distressingly close to frequencies used by cell phones and WiFi stations. Fortunately we can shield these out pretty effectively. Our circuit is enclosed in a lead-plated cavity, and lead is another superconductor, so that at low temperatures this becomes a perfectly conducting shield, essentially a Faraday cage. The device is further enclosed in a can made of mu-metal, an alloy used for magnetic shielding. Finally the entire apparatus is inside a small room walled in sheets of copper, providing yet another Faraday cage. (Rooms of this type were built by the NSA in US embassies during the Cold War to thwart Soviet spying.) All this gives us pretty good protection from external sources of EM waves, and it does seem to be necessary: just by opening the door of the copper room we can see the qubit’s performance degrade as radiation leaks in.
Ultimately to measure the qubit state we need to be able to interact with it, and this implies a tradeoff between qubit isolation and measurement sensitivity. The best case would be a measurement device which is “off”—producing no noise—for as long as we want to keep the qubit in its coherent state, and only turns on when we’re ready to measure. Our device (a Superconducting Quantum Interference Device, or SQUID) is designed this way: the components closest to the qubit are entirely superconducting except during the measurement process, and is arranged symmetrically around the qubit so that noise currents coming down the wires connected to the SQUID pass by on both the left and right, producing opposite magnetic fields that cancel each other out.
However, in practice it’s impossible to make the device perfectly symmetric; there’s always going to be some small deviation, and even a 1% difference can allow a lot of noise in. We have additional measures in place to reduce this noise: extensive filtering on the wires that connect to the SQUID, and electrical isolation from the main power lines in the building (which are incredibly noisy, as they are shared by all the other physics experiments at Berkeley). However, some of the commercial electronics we use to perform the measurement generate quite a lot of noise (Agilent does not seem to be targeting the solid-state quantum computing market), and there’s a limit to how much we can filter them: they do need to be tightly connected to the device in order to make the measurement. So we believe that these electronics are a major source of our decoherence, and one of my current projects is building a homemade low-noise pulse generator to replace the one we’re currently using.
We can generally calculate how much noise is present from our decoherence times, using what’s known as the “spin-boson model”, a theory for decoherence developed by Tony Leggett (who won the Nobel Prize in 2003 for his work in superfluids). We used this theory to explain an interesting effect—when we increase the rate at which we make measurements, the qubit’s decoherence time gets shorter. This is a result of the fact that each measurement produces a small amount of heat on the chip, which comes in two forms: electrons that break out of the superconducting state (quasiparticles) and vibrations in the metal (phonons). Both quasiparticles and phonons interact with the qubit, and we were able both to calculate and to demonstrate this effect. In essence the measurement rate was a knob we could turn to increase or decrease the decoherence. We’ve presented these results at meetings and will be preparing a publication on this later in the year.