Back to Research Notes & Blog
RESEARCH ARTICLE • TIER A FINDINGS
COMPILER LIVENESS RESEARCH • EXP-03-MCM-REUSE

Reusing Physical Qubits with Compiler Liveness Analysis on IBM Heron

How compiler-driven mid-circuit measurement reduced physical register width by a median 78.1%.
BY: Q-Psi Research Team
•
HARDWARE: ibm_marrakesh (IBM Heron r2)
•
JOB ID: da1u0om3kjvs738777cg

1. The Register Width Problem in NISQ Hardware

In quantum algorithm design, logical qubits are traditionally assumed to persist for the entire duration of a circuit. However, on physical quantum hardware, maintaining idle qubits incurs continuous decoherence ($T_1/T_2$ decay) and spectator cross-talk errors.

By performing static liveness analysis on the compiler's intermediate representation, the Q-Psi compiler identifies when a logical qubit's information has been fully extracted or measured. The physical slot is then reset ($|0\rangle$) and reassigned to a new logical wire.

2. Experimental Execution on IBM Marrakesh

We evaluated 6 streaming benchmark circuits on IBM Heron r2 (24,576 shots total). Logical register requirements ranged from 8 to 16 qubits:

16 Logical Qubits (STREAM_16_A):Compressed to 2 Physical Slots (87.5% reduction)
16 Logical Qubits (STREAM_16_B):Compressed to 3 Physical Slots (81.3% reduction)
12 Logical Qubits (STREAM_12_A):Compressed to 2 Physical Slots (83.3% reduction)
8 Logical Qubits (STREAM_8_A):Compressed to 2 Physical Slots (75.0% reduction)
Median Physical Width Reduction:78.1% Reduction

3. Non-Inferiority Verification

Reducing physical register width is only useful if observable output fidelity is preserved. In 4 out of 6 cases, the reused circuit achieved strictly lower observable RMSE than the uncompressed circuit, because eliminating idle qubits reduced cumulative $T_1$ decoherence. All 6 cases satisfied the pre-declared non-inferiority margin ($\epsilon = 0.05$).

4. Structural Claim Boundary

EXPLICIT SCIENTIFIC SCOPE

Qubit reuse is mathematically bounded by the treewidth and sequential liveness graph of the workload. Workloads with all-to-all simultaneous entanglement cannot be compressed without swapping overhead. We do not claim 16-qubit quantum computing generally runs on 2 qubits.

RESEARCH NOTE LINK: Mid-Circuit Measurement Research Note
RAW RESULTS SHA256: ec0f2cea637f433f9d134d05a7168b3114b4526c2f5483288efbdea01d83e8e4