Skip to content

Guide: IBM Quantum Hardware

This guide walks through demo/03_qiskit_ibm.py — running a 5-step VQE on a real IBM Quantum backend with full HilbertBench recording, including the hardware calibration snapshot.

Run it directly (IBM token required):

python demo/03_qiskit_ibm.py

Prerequisites

pip install hilbertbench qiskit-ibm-runtime

You need an IBM Quantum account. Set your token in the demo file:

IBM_TOKEN = "YOUR_IBM_QUANTUM_TOKEN_HERE"

What changes on hardware

Two additional steps are required compared to the simulator demo:

  1. Transpile to ISA — the circuit must be compiled to the backend's native gate set and topology
  2. Layout alignment — the observable must be aligned to the transpiled circuit's qubit layout
from qiskit_ibm_runtime import QiskitRuntimeService, EstimatorV2
from qiskit.transpiler.preset_passmanagers import generate_preset_pass_manager

service  = QiskitRuntimeService(channel="ibm_quantum_platform", token=IBM_TOKEN)
backend  = service.least_busy(operational=True, simulator=False)

pm       = generate_preset_pass_manager(optimization_level=1, backend=backend)
isa_qc   = pm.run(circuit)
isa_obs  = abstract_obs.apply_layout(isa_qc.layout)   # ← required on hardware

The proxy usage is identical to the simulator:

with HilbertTape("runs/ibm_vqe", tags={"backend": backend.name}) as tape:
    estimator = HilbertEstimatorProxy(tape, real_estimator=EstimatorV2(mode=backend))

Calibration snapshot

On hardware, HilbertBench automatically records the device calibration snapshot (T1, T2, readout error, gate error) at the time of the run. After the trace is sealed:

from hilbertbench.analysis import noise_profile

result = noise_profile(trace)
print(f"Noise verdict: {result['status']}")
print(f"T1 (mean):     {result['t1_us']['mean']:.1f} µs")

Note on iteration count

The demo uses N_ITER = 5 to minimize hardware queue time. For a production experiment, increase to 50–100 iterations.

Extending this example

  • Compare noise profiles across different backends using the same circuit
  • Record the calibration at multiple times during a long run to observe drift
  • Pair with circuit_structure to understand which gates dominate the hardware fidelity estimate