Qyantram’s verification scales with used qubits, not the entire device.

Sadique Khatib and colleagues at K. K. Wagh Institute of Engineering Education and Research have developed Qyantram, a quantum computing platform where verification scales with used qubits, not the entire device. The platform focuses on the active components of a circuit, leaving untouched qubit lines significantly reducing computational cost. This approach allows for full per-compilation verification on 127- and 156-qubit layouts, achieving speed-ups of around four orders of magnitude. Qyantram distinguishes itself by bundling every result with supporting evidence and subjecting it to independent double-checking.

Qyantram: Transparent Quantum Compilation Pipeline Stages

The Qyantram platform separates its compilation pipeline into distinct stages, each modifying only one circuit property at a time, ensuring traceability of added gates and a level of transparency uncommon in quantum computing systems. This stage-by-stage approach, covering decomposition, layout and routing and basis translation, maintains an invariant where alterations to gate arity, qubit placement, or gate set occur in isolation, simplifying debugging and verification processes.

Each modification is therefore easily linked to a specific stage within the pipeline, allowing engineers to pinpoint the rationale behind each transformation. Qyantram’s verification process focuses exclusively on actively used qubits, a strategy that dramatically reduces computational load. Untouched qubit lines are effectively eliminating the need to verify their state and enabling a mathematically exact scaling of verification cost as $2^k$, where k represents the number of used qubits.

The restricted check consistently completes below a millisecond, a speed enabled by this focused verification strategy. This commitment to verification extends to the routing stage, where Qyantram publishes a list of SWAP candidates that were not selected, along with their associated scores.

SABRE Router Records Decision-Making Process

The SABRE routing heuristic forms the foundation of Qyantram’s approach to circuit optimization, but the platform diverges from standard implementations by meticulously recording each decision made during the routing process. Unlike many systems that treat routing as a black box, Qyantram maintains a detailed log of SWAP candidates considered and their associated scores, even for those not ultimately selected.

This allows for a level of post-hoc analysis uncommon in quantum computing, offering insight into why specific routing choices were made and potentially revealing areas for improvement. While the platform’s router exhibited a mean factor of 1.27 greater SWAP count than SABRE, the detailed logging provides a benefit for debugging and refinement.

2^k Verification Scales with Used Qubits

This transparency extends to verification, where the system focuses solely on actively used qubits, leaving any untouched qubit lines a technique that drastically reduces computational load. The cost of this verification then scales as $2^k$ for k used qubits, a mathematically exact reduction from the $2^N$ complexity of verifying the entire device. This focused verification approach enables full per-compilation verification on circuits using 127 and 156 qubits, with checks completing in under a millisecond.

Every compilation limit is explicitly defined, including a maximum of 400 source instructions, 60 circuit snapshots per stage and a 900-gate limit for snapshot size, preventing coding errors from triggering unintended hardware jobs. Qyantram’s fourth subsystem focuses on training variational models, extending the platform’s capabilities beyond compilation and verification into the realm of quantum machine learning.

Pauli Error Identification with Exact Validation

Error identification within the Qyantram platform avoids approximation by exhaustively checking every possible Pauli error pattern, achieving agreement with closed-form predictions to 2.8× ${10}^{–17}$. Unlike many systems that rely on sampling to estimate logical error rates, Qyantram enumerates each possibility, a staggering 262,144 distinct patterns for the nine-qubit code, calculating the exact probability, syndrome, correction and outcome class using vectorised algebra. This approach delivers an exact rate, not an estimation, and provides a level of validation uncommon in current quantum computing systems.

Table XI details this comparison across a range of noise values, demonstrating consistent accuracy. The machine-learning engine within Qyantram further enhances efficiency, training roughly 25 to 50 times faster than comparable systems and landing within 2.0× ${10}^{–15}$ of quantum_info.

The platform supports three codes in the stabiliser formalism, the three-qubit bit-flip and phase-flip repetition codes, and Shor’s nine-qubit code, providing the check operators, syndrome-to-correction lookup tables and logical X and Z actions for each. The exhaustive enumeration process, rather than relying on statistical sampling, ensures that any discrepancies are not due to estimation errors, but rather to floating-point precision limits. “We do not estimate logical error rates by hoping repeated samples will cover the space,” the documentation states, emphasizing the platform’s dedication to verifiable quantum computation.

Qyantram’s Machine Learning Engine Benchmarks

Qyantram’s compilation engine, a core component of the platform, operates as a pure function devoid of internal state or direct access to hardware submission modules, preventing accidental execution of unverified circuits. This design choice, detailed in the platform’s documentation, ensures that each compilation remains isolated until intentionally deployed, a level of control uncommon in many quantum systems.

The engine itself comprises 4,187 lines of code, part of a larger service layer divided into six independent packages: compilation, machine learning, error identification, assessment, learning and authentication, with minimal inter-package dependencies. A key innovation lies in the platform’s verification process, which scales with the number of used qubits rather than the total device size.

Qiskit Comparison: Training Speed and Accuracy

Beyond routing, the speed advantage of Qyantram’s compilation engine becomes pronounced when processing batches of circuits, a common operation in training quantum machine learning models. The accuracy of Qyantram’s error identification module also demonstrates a high degree of fidelity, aligning with closed-form algebraic solutions to within 2.8 x 10^-17. The machine learning engine agrees with quantum_info to within 2.0 x 10^-15 while maintaining the 25- to 50-fold speed increase on the batched training path. The platform does not improve routing quality, and uses approximately 1.27 times as many SWAPs as SABRE.

Restricted Check Achieves Millisecond Verification

Verification of quantum computations, a traditionally resource-intensive process, now occurs in under a millisecond using the Qyantram platform when focused on actively used qubits, a feat previously unattainable on 127- and 156-qubit layouts. This speed is achieved by a restricted check that bypasses verification of idle qubit lines, an approach mathematically proven to be exact rather than an approximation. The router adds about 1.27 times as many SWAPs as SABRE because it skips optimisation; that extra cost is the price of keeping every routing decision visible from start to finish.

SWAP Gate Costs Reveal Routing Transparency

Qyantram’s routing process publishes a detailed record of candidate SWAP gates that were considered but not used, moving beyond the typical “guessing game” inherent in quantum circuit routing. Consequently, every added gate can be traced to a specific cause, allowing for targeted remediation when a circuit’s size unexpectedly increases. The platform’s approach to routing impacts overall circuit efficiency, as demonstrated by comparisons with other routing methods. While Qyantram adds approximately 27% more SWAPs, the detailed routing record facilitates easier auditing and understanding of the process.

Qyantram’s Four-Subsystem Platform Architecture

prioritizes a traceable record of circuit modifications, a departure from many quantum computing platforms that treat compilation as a black box. This transparency moves beyond a simple routing decision and provides insight into the platform’s optimization process, allowing users to understand why certain routing choices were made. This contrasts with systems where routing appears as an opaque process, obscuring the trade-offs between circuit depth and gate count. The detailed record also facilitates targeted remediation when a circuit’s performance degrades, enabling developers to address the root cause of inefficiencies.

More like this

Quantum AlgorithmsResearchers show how to automatically uncompute ancilla qubits with measurements

Quantum AlgorithmsA single library now links quantum algorithms to IBM and IQM hardware

Quantum Error CorrectionFaultScope software pinpoints errors in quantum error correction circuits

Quantum PhysicsMoth Quantum’s Braided Halloween game splits you into two quantum halves