Native Gate Sets

A native gate set is the short list of steps a device's control electronics can really do. The compiler rewrites textbook gates such as CX into these tuned steps. For example, it can build CX from a CZ with a Hadamard gate on each side. Two versions can give the same math but use different numbers of physical pulses, with different error. The two-qubit pulses cause most of the error.

How are textbook gates different from hardware instructions?

A device's native gate set is the short list of steps its control electronics can really do. Each native gate is backed by a calibrated pulse. That is a shaped microwave or laser signal, tuned until the physical step matches the gate's math. Everything else in your circuit gets translated.

Think of a phrasebook when you travel. You can say anything in your own language, but the listener only understands a few set phrases. So each idea must be rebuilt out of those phrases. The example breaks in one way: a clumsy translation still gets the point across, but every extra quantum pulse adds a little error.

This is the same idea as a normal chip's instruction set architecture (ISA). That is the list of operations it supports, what inputs they take, and, for a QPU, which qubit pairs they run on. Your code says CX. The hardware may have no CX pulse at all. Superconducting devices often offer CZ or ECR as their entangling step, the gate that links two qubits. (ECR, short for echoed cross-resonance, is another two-qubit microwave pulse.) Trapped-ion devices offer a two-qubit interaction of their own. The compiler builds CX out of whatever is there.

So a CX in a circuit listing is a claim about math, not about pulses. The two runs below make that difference something you can measure.