How to create custom fake backend?

Question

Right now I'm trying to figure out how to create custom fake backends using qiskit with data we are retrieving from our real backends in the laboratory.
As I understand, the best way to create one is to use GenericBackendV2.

Documentations says, that we can specify:

• Number of qubits
• List of basis gate names
• Coupling map for the backend
• Instruction calibration settings

But what about other characteristics? For instance, I want to set T1 / T2 times or each qubit's frequency and maybe even some measurement errors.

Documentations says:

The remainder of the backend properties are generated by randomly sampling from default ranges extracted from historical IBM backend data.

But can I somehow set it manually?

Answer 1

For setting T1/T2 times, measurement error, and other errors (like reset errors etc.) you can make a custom noise model by using the NoiseModel() function from the qiskit_aer.noise module.

Here is a high-level overview of how you can achieve this:

1. To use this function, you first define what errors you want to model and on what gates you want to apply this errors.

2. Then you add these errors to a noise model, and here you get to pick if you want to apply the noise models to all the qubits or just some specific qubits.

3. After doing this, you can use your noise model in the AerSimulator() to simulate your circuit on the simulator or you can specify your custom backend to transpile and run your circuit.

Here is the link to the documentation for building noise models. Hope this helps!

Answer 2

One option is to create a new class that inherits from GenericBackendV2 and accepts "noise settings". The below code shows how to do that

class MyFakeBackend(GenericBackendV2):
    def __init__(
        self,
        num_qubits: int,
        basis_gates: list[str] | None = None,
        *,
        coupling_map: list[list[int]] | CouplingMap | None = None,
        control_flow: bool = False,
        calibrate_instructions: bool | InstructionScheduleMap | None = None,
        noise_settings: dict,
        dtm: float | None = None,
        seed: int | None = None,
    ):

        self.noise_settings = noise_settings
        super().__init__(num_qubits, basis_gates,
            coupling_map=coupling_map,
            control_flow=control_flow,
            calibrate_instructions=calibrate_instructions,
            dtm=dtm,
            seed=seed)

    def _get_noise_defaults(self, name: str, num_qubits: int) -> tuple:
        if name in self.noise_settings:
            return self.noise_settings[name]

        if num_qubits == 1:
            return _NOISE_DEFAULTS_FALLBACK["1-q"]
        return _NOISE_DEFAULTS_FALLBACK["multi-q"]

Now, you can instantiate your fake backend as follow:

cmap = CouplingMap.from_full(27)
noise_settings = {
    "cz": (7.992e-08, 8.99988e-07, 1e-5, 5e-3),
    "id": (2.997e-08, 5.994e-08, 9e-5, 1e-4),
    "rz": (0.0, 0.0),
    "sx": (2.997e-08, 5.994e-08, 9e-5, 1e-4),
    "x": (2.997e-08, 5.994e-08, 9e-5, 1e-4),
    "measure": (6.99966e-07, 1.500054e-06, 1e-5, 5e-3),
    "delay": (None, None),
    "reset": (None, None),
}

fake = MyFakeBackend(num_qubits=27,
    basis_gates=['cz', 'id', 'rz', 'sx', 'x'],
    coupling_map=cmap,
    noise_settings=noise_settings)

Note that, noise_settings follows the same structure used internally in GenericBackendV2 for defining _NOISE_DEFAULTS, where there are two possible formats:

• (min_duration, max_duration, min_error, max_error), for ranges.
• (duration, error), for fixed values.