Quantum Computing RFP Template: Tests, SLAs & Failure Modes
Introduction
Procuring a quantum computer is unlike any classical hardware purchase: the device’s value hinges on fragile error rates, calibration stability, and vendor transparency that can evaporate post-contract. Technical buyers need a rigorous quantum computing RFP template that forces vendors to commit to measurable acceptance tests, enforceable SLAs, and disclosure of hidden failure modes before capital leaves the building.
This guide delivers a production-grade RFP skeleton, concrete acceptance criteria, SLA language, and a checklist of failure modes that have already cost early adopters millions. Whether you are evaluating superconducting, trapped-ion, or photonic systems, the template below protects your organization from optimistic marketing slides and delivers evidence-led procurement.
A recent procurement at a Tier-1 financial institution illustrates the risk: the selected vendor quoted 99.9 % two-qubit fidelity; post-delivery the median fidelity collapsed to 96.4 % under sustained 24-hour workloads, rendering error-corrected algorithms unusable. The contract lacked continuous benchmarking clauses and the organization absorbed a seven-figure write-down. Such hidden failure modes are now preventable.
Executive Summary
TL;DR: A quantum computing RFP template must mandate statistically valid acceptance tests (e.g., randomized benchmarking with 10^5 shots, p99 fidelity floors), 30-day stability SLAs with liquidated damages, and mandatory disclosure of crosstalk, leakage, and calibration drift profiles.
- Require vendor submission of full tomographic data and error budgets before shortlisting.
- Embed independent randomized benchmarking and quantum volume protocols in acceptance criteria.
- Define SLAs around uptime, fidelity stability, and error-correction decoder latency with financial penalties.
- Force disclosure of all known hidden failure modes: spectator errors, leakage, TLS defects, and cosmic-ray upsets.
- Use our Quantum Computer Procurement: Technical Due Diligence Checklist to cross-validate vendor responses.
- Include continuous benchmarking rights and escrow of calibration scripts in contract terms.
Direct Answers for Common Queries
What are the core acceptance test criteria for a quantum computer? Statistically powered randomized benchmarking (RB) with ≥10^5 shots per gate, median two-qubit fidelity ≥99.5 % (p99 ≥99.0 %), quantum volume ≥2^10, and <24-hour calibration stability verified by daily GST runs.
What SLA requirements should appear in a quantum computing contract? 99.5 % system availability, fidelity drift <0.3 % over 30 days, decoder latency p99 < 120 µs, and mandatory root-cause reporting within 4 hours of any fidelity excursion beyond agreed bounds.
How do you expose hidden failure modes during quantum hardware procurement? Demand full error-budget spreadsheets, leakage population measurements, TLS defect density maps, cosmic-ray upset rates, and spectator-qubit error correlations; cross-check against independent quantum benchmarking methodology.
How a Quantum Computing RFP Template Works Under the Hood
An effective RFP is not a checklist but a technical contract scaffold. It decomposes the quantum system into four measurable layers: physical qubits, control electronics, compiler/runtime, and error-correction stack. Each layer carries explicit metrics, test protocols, and remedies.
Physical-qubit metrics focus on T1, T2*, single- and two-qubit gate fidelities obtained via interleaved RB and gate-set tomography (GST). Control-layer SLAs target pulse-timing jitter (< 2 ns p99) and AWG phase noise. The compiler/runtime layer must expose compilation depth, routing overhead, and native-gate decomposition fidelity. Finally, the error-correction layer requires decoder latency, logical-error-rate suppression curves, and leakage detection efficiency.
Hidden failure modes appear at the intersections: TLS defects couple to qubit frequencies causing sudden T1 drops; cosmic rays induce correlated multi-qubit errors; spectator qubits accumulate phase errors that standard RB misses. The RFP forces vendors to quantify these effects with concrete diagnostics rather than hand-waving “typical” values.
Our companion article Quantum Error Correction Readiness: Judging Logical-Qubit Claims provides deeper guidance on interpreting the logical-error-rate data vendors will submit.
Implementation: Production RFP Patterns
Below is a modular RFP template excerpt. Copy, adapt, and embed into your organization’s procurement workflow.
1. Vendor Information & Evidence Requirements
Respondents must supply:
- Full error budget for every qubit (T1, T2, readout, gate, crosstalk, leakage).
- Raw RB and GST datasets for at least three calibration epochs spaced 30 days apart.
- Device topology graph and calibrated frequency map at delivery.
2. Acceptance Test Criteria
# Example acceptance test script skeleton (Python/PyQuil style)
import numpy as np
from cirq import Circuit, measure, X, CZ
from qiskit.ignis.verification import randomized_benchmarking as rb
def run_rb_acceptance(backend, num_qubits=2, lengths=[10,50,100], shots=100000):
rb_experiment = rb.RandomizedBenchmarking(
num_qubits, lengths, shots_per_length=shots)
result = rb_experiment.run(backend)
# p99 fidelity floor must exceed 0.99 for two-qubit gates
assert np.percentile(result.fidelities, 1) >= 0.99, "p99 fidelity below floor"
return result.fidelities
# Daily stability test (must pass 30 consecutive days)
def daily_stability_test(backend):
fidelities = []
for day in range(30):
f = run_rb_acceptance(backend)
fidelities.append(np.median(f))
drift = np.max(fidelities) - np.min(fidelities)
assert drift < 0.003, f"Fidelity drift {drift:.1%} exceeds 0.3 % SLA"
Insert the above or equivalent into the RFP as mandatory deliverable code. Require vendors to execute it on their hardware and submit signed reports from an independent calibration lab.
3. SLA Requirements & Liquidated Damages
Contract language example:
“Vendor guarantees median two-qubit fidelity ≥ 99.5 % (p99 ≥ 99.0 %) measured daily via interleaved RB with 10^5 shots. If fidelity falls below floor for > 72 consecutive hours, Buyer may invoke a service credit equal to 5 % of monthly subscription fee per percentage point shortfall, capped at 50 % of total contract value.”
Additional SLA clauses:
- System availability ≥ 99.5 % measured monthly (excludes scheduled calibration windows < 4 h).
- Decoder p99 latency ≤ 120 µs for surface-code distance-5; measured on vendor-provided FPGA or cloud endpoint.
- Maximum calibration time 8 h per week; any overrun triggers 2 % credit.
Reference our Quantum Error Mitigation Decision Tree for NISQ when negotiating which error rates justify credits.
4. Hidden Failure Mode Disclosure & Mitigation
The RFP must contain a dedicated “Failure Mode Inventory” section. Vendors submit a table with columns: FailureMode, ProbabilityPerHour, DetectionMethod, Mitigation, ResidualLogicalError.
Typical entries to demand:
- TLS-induced T1 fluctuation (probability, frequency map).
- Cosmic-ray burst rate and spatial correlation length.
- Leakage to non-computational states (measured via randomized benchmarking with leakage detection).
- Spectator-qubit dephasing (measured via simultaneous RB on all qubits).
Comparisons & Decision Framework
Use the following rubric when scoring proposals. Each criterion is weighted by organizational risk tolerance.
| Criterion | Superconducting | Trapped-Ion | Photonic | Minimum Score |
|---|---|---|---|---|
| Median 2Q Fidelity (p99) | 99.6 % | 99.8 % | N/A (measurement based) | 99.0 % |
| Calibration Stability (30 d drift) | 0.4 % | 0.1 % | 0.05 % | <0.3 % |
| Logical Qubit Suppression Factor | 10× at d=5 | 50× at d=3 | Early stage | ≥8× |
| Cosmic-Ray Upset Rate | 1.2 /h/m² | 0.3 /h/m² | Negligible | Documented & mitigated |
Score each vendor against the rubric, then apply the Quantum Computer Procurement: Technical Due Diligence Checklist before final selection. Cross-reference vendor manufacturing claims against our independent survey in Quantum Computer Manufacture: Who Builds Them & What Scales.
Failure Modes & Edge Cases
Hidden failure modes repeatedly surface after procurement:
- TLS Defects: Sudden T1 drops from 80 µs to 12 µs. Diagnostic: repeated T1 spectroscopy over 48 h. Mitigation: frequency tuning away from defect; require vendor to supply tunable-frequency map.
- Cosmic-Ray Correlated Errors: Multi-qubit phase flips at ~1 event per hour in large chips. Diagnostic: monitor parity of ancilla qubits. SLA clause: vendor must demonstrate detection efficiency > 90 %.
- Leakage Accumulation: Population leaks into |2⟩ state, invisible to standard RB. Require leakage-aware benchmarking (e.g., randomized benchmarking with leakage detection) with residual leakage < 10^{-4} per gate.
- Calibration Drift Under Load: Fidelity collapses after 10 h of continuous algorithmic workload. Acceptance test must include 24-hour stress run with representative circuit depth.
- Spectator-Qubit Crosstalk: Idling qubits accumulate phase errors proportional to number of active gates. Demand simultaneous RB data on all qubits.
For each mode, the RFP must require a quantitative mitigation plan and residual error after mitigation.
Performance & Scaling
Key performance indicators to track post-acceptance:
- p95 two-qubit fidelity across daily runs (target > 99.4 %).
- Logical error per round for distance-5 surface code (target < 10^{-3} within 12 months).
- Decoder latency p99 (target < 120 µs; critical for real-time feedback).
- Effective quantum volume under sustained load (target ≥ 2^{12} within 18 months).
Monitoring recommendations: integrate vendor telemetry with Prometheus exporters for fidelity, T1/T2, and leakage population. Set alerts at 2σ deviation from baseline. Require quarterly independent audits using the protocols described in our Quantum Benchmarking Methodology article.
Scaling clause: contract must include volume discounts and performance multipliers when qubit count doubles, tied to maintained fidelity floors.
Production Best Practices
1. Escrow all calibration, tuning, and decoder source code with a neutral third party; trigger release on material SLA breach.
2. Mandate on-site or remote “shadow” access for your own calibration engineers during the first 90 days.
3. Include a continuous benchmarking clause: vendor must expose a public API endpoint that runs daily RB and GST; results feed into your internal dashboard.
4. Security: require quantum-safe encryption of classical control channels and audit rights for supply-chain provenance of cryogenic components.
5. Rollout runbook: begin with NISQ proof-of-concept workloads, graduate to logical-qubit experiments only after 30-day stability sign-off.
6. Exit criteria: predefined logical-error-rate thresholds that allow contract termination without penalty if vendor cannot demonstrate progress within 18 months.
Further Reading & References
- Blume-Kohout et al., “A Randomized Benchmarking Suite for Quantum Computers,” arXiv:1712.06550 (2018).
- Quantum Economic Development Consortium (QED-C) Benchmarking Suite, 2024 release.
- Google Quantum AI, “Suppressing quantum errors by scaling a surface code logical qubit,” Nature 614, 676–681 (2023).
- Our technical due-diligence companion: Quantum Computer Procurement: Technical Due Diligence Checklist.
- Verification-focused follow-up: Quantum Computer Procurement: Verify Vendor Claims Before You Buy.
- IonQ & Rigetti public error-budget disclosures (2024–2025).
Adopt this RFP template verbatim or adapt it to your risk model. The upfront engineering effort pays for itself many times over by preventing seven-figure write-downs and accelerating time-to-value on quantum workloads.