Abstract
A quantum computer (QC) can generate true random numbers using the quantum superposition property of quantum bits (qubits). However, many types of noise impart bias into the created number, impairing its randomness. To compensate for the noise-induced bias and generate reliable random numbers, we propose gate parameter optimization of the TRNG circuit. We employ a hybrid quantum-classical loop to optimize the gate parameter. The parameter optimization routine can compensate for the bias and enhance the random number’s quality by utilizing even the lowest quality qubits. However, finding the ideal parameter involves lengthy repetitions between classical and quantum machines in the hybrid configuration. We conduct a series of error characterization and quantum tomography experiments to examine the effects of various noises such as gate error, decoherence, and readout error on QC-based true random number generators (TRNG)s. Leveraging insights from the study, we develop a regression-based machine learning approach that predicts the optimal gate parameter from qubit error specification without invoking the costly quantum-classical loop. Moreover, we propose another method by merging the hybrid loop and regression-based model to fine-tune the parameter. We validate our approaches by using experiments on real quantum computers from IBM and testing the generated bitstrings with the NIST statistical test suite. Experimental results suggest that the techniques can correct bias by up to 88.57%, even in worst-case qubits.
Original language | English (US) |
---|---|
Title of host publication | Design Automation of Quantum Computers |
Publisher | Springer International Publishing |
Pages | 69-86 |
Number of pages | 18 |
ISBN (Electronic) | 9783031156991 |
ISBN (Print) | 9783031156984 |
DOIs | |
State | Published - Jan 1 2022 |
All Science Journal Classification (ASJC) codes
- General Engineering