It is convenient to label different sources of noise in a quantum device (as there are usually many) and we categorize them to better understand them. We call noise that comes from preparing and measuring a quantum state SPAM error.
To prepare a quantum system, we need highly controlled conditions to initialize our physical qubits and often we require applications of unitaries/gates to prepare a specific quantum state before implementing our main algorithmic circuit – all this preparation can introduce noise/errors in our overall computation. Making a quantum measurement at the end of a computation, often includes applying specific unitaries on certain qubits (physical or logical) and involves various methods for gathering classical outcomes. Since this, again, includes many steps, measurement introduces a lot of errors too. We group these two sources of noise together because (i) they can contribute significantly to overall noise, and (ii) we often want to isolate them in order to characterize gate or circuit performance more accurately.
Frequently asked questions
- What are some examples of SPAM error?
Failing to initialize all qubits in the intended state, e.g. due to thermal excitations, Cross-talk, where operations on one qubit affect the state of another during preparation or measurement. In photonic systems distinguishability of photons and loss of photons both contribute significantly to SPAM error.
- How do we separate SPAM in experiments?
In randomized benchmarking SPAM errors are separated from the errors on gates, because the results of the protocol entirely depend on the sequences and sequence length while SPAM is independent of this. Gate set tomography can model SPAM errors entirely by learning all possible information about the state. Experimental tests to calibrate the device also tell us about the SPAM noise and it’s strength.
- How can we reduce SPAM error?
That is a good question and one that researchers are actively working on. In general, we can keep making improvements to hardware by, e.g., increasing coherence time and reducing cross-talk noise and we can also use error mitigation and error correction to ensure SPAM does not impact our computations. We can also optimize readout methods to make them more efficient and accurate.