Silicon-based quantum random number generator with untrusted sources and uncharacterized measurements.

Random numbers are essential resources in science and engineering, with indispensable applications in simulation, cybersecurity, and finance. Quantum random number generators (QRNGs), based on the principles of quantum mechanics, ensure genuine randomness and unpredictability. Silicon photonics enables the large-scale deployment of integrated QRNGs due to its low cost, miniaturization, and compatibility with CMOS technology. However, current integrated QRNGs are typically based on perfect or partially perfect device models, deviating from real-world devices, which compromises the unpredictability of quantum random numbers. In this study, we implemented a silicon-based QRNG that makes no assumptions about the source and only uses trusted but uncharacterized measurement devices. In experimental demonstration, we show that our setup can generate secure random numbers with different choices of intensities of laser light, and achieve an optimized random number generation rate of up to 4.04 Mbps. Our work significantly advances the security, practicality, and commercial development of QRNGs by employing imperfect devices.