Quantum random number generation based on a perovskite light emitting diode

Perovskites have enabled efficient, low-cost photonic devices and are promising for LEDs due to narrow emission and tunable bandgaps, with simple solution processing and potential sustainability advantages. Quantum information technologies require practical, integrable components; however, complete quantum information tasks using perovskite devices had not been demonstrated. This work asks whether a metal-halide PeLED can serve as the light source for a modern, secure quantum random number generator with competitive bitrate and certified privacy. The study targets a measurement-device-independent (MDI) QRNG that maintains security even if detectors are untrusted, aiming for cheap, fast, integrable, and potentially more sustainable implementations.

QRNGs exploit intrinsic quantum measurement randomness. Security levels range from device-independent (DI) approaches offering strongest certification but low rates, to device-dependent (DD) systems with high rates but requiring full characterization. Semi-device-independent (SDI) approaches, including MDI QRNGs, balance security and performance by making reasonable assumptions (e.g., trusted state preparation, untrusted measurement). Detector side-channel attacks motivate MDI schemes that certify private randomness despite untrusted detectors. Homodyne-based QRNGs can achieve very high rates in both DD and SDI settings. Prior perovskite quantum optics efforts existed but not full quantum information tasks. The authors situate their PeLED-based MDI-QRNG among competitive systems, showing raw bitrates comparable to commercial devices.

Device and source: The PeLED uses an ITO-coated glass substrate. The active layer is ~50 nm formamidinium lead iodide perovskite with a pimelic acid additive for stability. Electron and hole transport layers are PEIE-modified ZnO and TFB (~40 nm), followed by MoOx (~7 nm) and Au (~80 nm) contacts. Under forward bias the device emits at 804 nm with ~41.6 nm FWHM and ~18% peak external quantum efficiency. Experimental setup (MDI-QRNG): The preparation stage (trusted) houses the PeLED to produce weak coherent states. A linear polarizer filters unpolarized emission; a liquid crystal waveplate (LCWP) with <50 ms switching implements state selection. A 50 mm focal length ball lens collimates the light. A calibrated optical attenuator sets the mean photon number per detection window to keep multiphoton probability <0.28%. State preparation and protocol: For each round, the user chooses between test states |H⟩ or |V⟩ (for measurement device checking) or a superposition state |R⟩ = 1/√2(|H⟩ + i|V⟩) (for randomness generation). The measurement device (untrusted) consists of an uncharacterized PBS ideally projecting onto |H⟩ and |V⟩ with outputs sent to silicon SPADs D1 and D2 (PerkinElmer), ~25% overall detection efficiency at 800 nm. The measurement stage outputs a bit “0” or “1” to FPGA electronics. Events with no click or double clicks are discarded. The PBS has unequal transmission/reflection, leading to differing success probabilities for |H⟩ and |V⟩. Measurement scheduling: Before each data block, an external software RNG (demo setting) chooses measurement mode with 99% probability for randomness generation and 1% for testing (in practical deployments a trusted low-rate RNG or feedback from generated bits would be used). Security certification (MDI): From estimated p(α|ωx) for x in {H,V,R}, an optimization (per prior work) upper-bounds the eavesdropper’s guessing probability P(x*), even allowing entanglement and full control of detectors. Certified private randomness per round is H_min(x) = −log2 P(x*). Data handling and extraction: The FPGA buffers raw bits in blocks of 2^16 and streams via UDP to a host server, storing data by state for analysis and extraction. Toeplitz hashing extracts near-uniform bits from rounds prepared as |R⟩: raw sequences are split into N sequences of length n=400; an n×m Toeplitz matrix multiplies each to yield m hashed bits. The Toeplitz matrix is defined once per run with a seed of n+m−1 bits from the start of the raw sequence and reused; outputs Tη1, Tη2, … are concatenated. A security parameter of 2^−100 is used per the leftover hash lemma. Statistical testing: 5 Gbits of extracted bits are subjected to the NIST SP 800-22 statistical suite (16 tests). Sequences are divided into b=5000 blocks (typical block size 10^6 bits), with significance level α=0.01. The confidence interval for pass proportion is r ± 3√(r(1−r)/b), giving 0.9858 ≤ r ≤ 0.9942. A test is passed if the observed pass proportion lies within this interval. Operational conditions: The PeLED was driven at a constant current density of 0.24 mA cm−2 during long-term rate/lifetime observations.

- Raw random bit generation rate reached a maximum sustained 10.35 Mbit s−1; the overall average over the full run was 9.01 ± 1.30 Mbit s−1. - The success probability for test states remained stable over time despite PeLED emission decay; average P_suc ≈ 0.97 ± 0.01. - Certified private randomness per experimental round in the MDI scenario was H_min ≈ 0.71 ± 0.01 bits, bounding eavesdropper knowledge even with detector control. - The extracted 5 Gbit sequence passed all NIST SP 800-22 tests: mean p-values per test exceeded the 0.01 threshold and pass proportions met the confidence criterion (0.9858 ≤ r ≤ 0.9942). - Operational lifetime: stable performance for ~8 days with gradual rate decay thereafter; after 22 days the PeLED retained more than half of its initial brightness under 0.24 mA cm−2 drive. - Demonstrates a compact PeLED-based source can serve as a high-quality light source for quantum information tasks with performance comparable to commercial QRNGs.

The study demonstrates that a metal-halide PeLED can act as a practical quantum light source for an MDI QRNG, addressing the need for cost-effective, integrable, and potentially more sustainable components in quantum technologies. By implementing an MDI framework with trusted state preparation and untrusted measurement, the system certifies private randomness against detector side-channel attacks. The high sustained raw rates (up to 10.35 Mbit s−1) and successful NIST testing on 5 Gbits indicate competitive performance. Stability of test-state success probabilities over time shows that degradation in source brightness did not compromise the security certification; the certified min-entropy per round remained high (~0.71 bits). The results place PeLED-based QRNGs among leading implementations and suggest that perovskite emitters can support secure quantum information tasks while offering manufacturing and environmental advantages associated with perovskites.

PeLEDs were successfully employed to implement a modern, highly secure MDI quantum random number generator. The device achieved up to 10.35 Mbit s−1 raw rate with an average of 9.01 ± 1.30 Mbit s−1, passed the NIST SP 800-22 tests on 5 Gbits, and certified ~71% private randomness per round via the MDI approach. The PeLED operated for at least 22 days while retaining more than half of its initial brightness. These results demonstrate feasibility of perovskite-based quantum technology systems and indicate a path toward more sustainable and integrable QRNGs with performance comparable to commercial devices. Future directions include leveraging recent advances to further improve PeLED operational stability and lifetime, enabling broader practical deployment.

- Source degradation: The PeLED emission decayed over time, leading to a gradual reduction in bitrate after ~8 days; lifetime observed to 22 days with >50% initial brightness remaining. - Detection efficiency and balancing: The measurement stage used SPADs with ~25% overall detection efficiency at 800 nm and a PBS with unequal transmission/reflection, which can limit rates and introduce bias (mitigated in extraction and testing). - Event discarding: No-click and double-click events were discarded, reducing effective throughput. - Demonstration control: Block-type selection used an external software RNG for demonstration; practical secure deployments require a trusted low-rate randomness source or feedback from generated bits. - Weak coherent source requirement: The mean photon number had to be attenuated to keep multiphoton probability <0.28%, which constrains brightness and rate.