Add ESP readout support to Qiskit

[zachschoenfeld33]

Summary

Adds flag use_measure_esp to assemble, which sets the final measurement in a circuit to use Excited State Promoted (ESP) readout. See https://arxiv.org/pdf/2008.08571.pdf for reference.

A backend configuration flag measure_esp_enabled is included. The ESP feature can only be used if measure_esp_enabled=True.

[zachschoenfeld33]

@taalexander @dtmcclure

[mtreinish]

I'm not clear why this needed to be merged into terra at all. It's an ibm backend only option right? Why isn't this just exposed as an option from the ibmq provider backend? It should be passed through the kwargs for the run command (and execute() by extension) if it's exposed through the options interface correctly.

Also we shouldn't be updating the schemas here anymore. They're deprecated and frozen slated for removal soon (see #6558 )

[taalexander]

I'm not clear why this needed to be merged into terra at all. It's an ibm backend only option right? Why isn't this just exposed as an option from the ibmq provider backend? It should be passed through the kwargs for the run command (and execute() by extension) if it's exposed through the options interface correctly.

Agreed, this is a backend-only option. Unfortunately, the provider is calling out to assemble to build the Qobj payload still, so until this function is migrated to the provider this seems like the required place for this logic. We avoided any other IBM specific changes in this PR. See Qiskit/qiskit-ibmq-provider#952 for backend-specific changes.

Also we shouldn't be updating the schemas here anymore. They're deprecated and frozen slated for removal soon (see #6558 )

Ok, these changes are just a mirror of Qiskit/ibm-quantum-schemas#10. We will not make further updates to these schemas.

[mtreinish]

I'm not clear why this needed to be merged into terra at all. It's an ibm backend only option right? Why isn't this just exposed as an option from the ibmq provider backend? It should be passed through the kwargs for the run command (and execute() by extension) if it's exposed through the options interface correctly.

Agreed, this is a backend-only option. Unfortunately, the provider is calling out to assemble to build the Qobj payload still, so until this function is migrated to the provider this seems like the required place for this logic. We avoided any other IBM specific changes in this PR. See Qiskit/qiskit-ibmq-provider#952 for backend-specific changes.

Sure, I agree that we should copy the assembler and qiskit.qobj over to the ibmq provider and longer term deprecate and remove it from terra (but we should be conservative on that given it's former central usage). But, even without that being done the way assemble() is written you don't need to add explicit kwargs for every single option in the run config. If you pass in arbitrary kwargs to the function it gets added directly the run config. For example I ran:

from qiskit.compiler import assemble
from qiskit import QuantumCircuit
import qiskit

print(qiskit.__version__)
qc = QuantumCircuit(2)
qc.h(0)
qc.cx(0, 1)
qc.measure_all()
print(assemble(qc, use_measure_esp=True))

with the current release of qiskit-terra and it returns:

0.17.4
QASM Qobj: a5005853-6158-4e00-9abe-9f061adeb129:
Config: {'init_qubits': True,
 'meas_level': <MeasLevel.CLASSIFIED: 2>,
 'memory': False,
 'memory_slots': 2,
 'n_qubits': 2,
 'parameter_binds': [],
 'shots': 1024,
 'use_measure_esp': True}
Header: {}
Experiments:

QASM Experiment:
Header:
{'clbit_labels': [['meas', 0], ['meas', 1]],
 'creg_sizes': [['meas', 2]],
 'global_phase': 0.0,
 'memory_slots': 2,
 'metadata': {},
 'n_qubits': 2,
 'name': 'circuit-0',
 'qreg_sizes': [['q', 2]],
 'qubit_labels': [['q', 0], ['q', 1]]}
Config:
{'memory_slots': 2, 'n_qubits': 2}

	Instruction: h
		qubits: [0]

	Instruction: cx
		qubits: [0, 1]

	Instruction: barrier
		qubits: [0, 1]

	Instruction: measure
		qubits: [0]
		memory: [0]

	Instruction: measure
		qubits: [1]
		memory: [1]

So all this is doing in practice is raising an exception at assemble() call time if use_measure_esp is enabled and the backend doesn't support it. But that arguably should be checked before the provider is calling to terra's assemble() function.

[zachschoenfeld33]

One thing I will say is it was far easier to debug and test from assemble than the provider. assemble has local unit tests, while the provider seems to require some server connections...

[zachschoenfeld33]

This can be revoked in favor of Qiskit/qiskit-ibmq-provider#952.