High-Speed Device-Independent Quantum Random Number Generation without a Detection Loophole

Quantum mechanics provides the means of generating genuine randomness that is impossible with deterministic classical processes. Remarkably, the unpredictability of randomness can be certified in a manner that is independent of implementation devices. Here, we present an experimental study of device-independent quantum random number generation based on a detection-loophole-free Bell test with entangled photons. In the randomness analysis, without the independent identical distribution assumption, we consider the worst case scenario that the adversary launches the most powerful attacks against the quantum adversary. After considering statistical fluctuations and applying an $80\mathrm{Gb}\times 45.6\mathrm{Mb}$ Toeplitz matrix hashing, we achieve a final random bit rate of $114\mathrm{bits}/\mathrm{s}$, with a failure probability less than ${10}^{-5}$. This marks a critical step towards realistic applications in cryptography and fundamental physics tests.

Figure 1

Schematics of the experiment. Light pulses of 10 ns, 100 kHz from a 1560 nm seed laser (LD) are amplified by an erbium-doped fiber amplifier (EDFA), and up-converted to pulses at 780 nm via second-harmonic generation (SHG) in an in-line periodically poled lithium niobate (PPLN) waveguide. The residual 1560 nm light is removed by a wavelength-division multiplexer (WDM) and spectral filters. A half-wave plate (HWP) and a liquid crystal (LC) is used to adjust the pump polarization. The 780 nm light pulses are focused into a periodically poled potassium titanyl phosphate (PPKTP) crystal in a Sagnac loop to generate polarization entangled photon pairs. After removing the 780 nm pump pulses by dichroic mirrors (DMs) and a 1 mm thick silicon plate, the entangled photons at 1560 nm are subject to polarization state measurements and then sent to superconducting nanowire single-photon detectors (SNSPDs) via a 130 m optical fiber.

Figure 2

Randomness generation versus number of experimental trials. We set the expected CHSH game value to be ${\omega }_{\exp }=3.52\times {10}^{-4}$, ${ϵ}_s={ϵ}_{\mathrm{EA}}=1/\sqrt{n}$ and ${\delta }_{\mathrm{est}}=\sqrt{10/n}$ for the finite data rate.

Figure 3

Randomness generation versus expected CHSH game value. We set ${ϵ}_s={ϵ}_{\mathrm{EA}}=1/\sqrt{n}$ and ${\delta }_{\mathrm{est}}=\sqrt{10/n}$ for finite data rate curves.