Quantum Random Number Generators

• Quantum Random Number Generators (QRNGs) are devices that harness intrinsic quantum uncertainty—through methods like vacuum-state fluctuations or single-photon detections—to produce truly random numbers.
• They utilize diverse physical mechanisms, including homodyne detection, SPAD arrays, and radioactive decay, achieving throughput from a few bits per second to multi-Gbps rates.
• QRNGs play a vital role in secure cryptographic key generation, Monte Carlo simulations, and quantum key distribution, ensuring randomness with rigorous statistical validation.

Quantum Random Number Generators (QRNGs) are physical devices that exploit fundamentally indeterministic quantum processes to produce statistically provable random numbers. Unlike classical random number generators, whose unpredictability is ultimately constrained by deterministic physics and initial conditions, QRNGs leverage quantum measurements—e.g., vacuum-state quadrature fluctuations, single-photon path choices, or nuclear decay events—where the outcome probabilities are set exclusively by the Born rule and immune to any classical “hidden variable” prediction. QRNGs are critical for cryptographically secure key generation, Monte Carlo simulation, and fundamental tests of quantum physics.

1. Fundamental Physical Mechanisms and Principles

QRNGs derive their entropy sources from irreducible quantum uncertainty. The most widely realized QRNG types operate on the following principles (Ma et al., 2015, Herrero-Collantes et al., 2016):

• Vacuum-state quadrature fluctuations: Balanced homodyne detection measures one quadrature of the electromagnetic vacuum, described by the operator $\hat{x} = (\hat{a} + \hat{a}^\dagger)/\sqrt{2}$, with the quantum shot-noise variance $\langle(\Delta x)^2\rangle=1/2$ in the vacuum state. The measured outcome is a continuous Gaussian random variable (Bai et al., 2021, Bruynsteen et al., 2022, Qiao et al., 16 Sep 2025, Haider et al., 3 Jun 2025).
• Single-photon path and time-of-arrival: A single photon impinging on a beam splitter or time-gated detector gives an intrinsically random click position or arrival time, each governed by quantum amplitudes and statistics, e.g., Poisson processes for emission times (Ma et al., 2015, Stipčević et al., 2014, 0807.4111).
• Radioactive decay: The spontaneous $\beta$-decay timing of nuclei is quantum-mechanical tunneling, yielding Poisson-distributed time intervals $$P(t)\mathrm{d}t = \lambda e^{-\lambda t}\mathrm{d}t$$; counts per window or interval parity serve as random outputs (Herrero-Collantes et al., 2016, Zafar et al., 2024).
• Solid-state emitters, SPAD shot noise, and LED emission: Single-photon sources (e.g., defects in hexagonal boron nitride), avalanche diode arrays, and even smartphone cameras can resolve discrete photoelectronic events—Poissonian in origin—modulated by quantum uncertainty in emission or absorption (White et al., 2020, Regazzoni et al., 2021, Sanguinetti et al., 2014, Moeini et al., 2023).

Trust models explicitly distinguish practical (trusted-component), semi-self-testing (source- or measurement-independent), and self-testing (fully device-independent, certified by Bell-inequality violations) QRNGs, with respective tradeoffs in throughput and adversarial robustness (Ma et al., 2015, Herrero-Collantes et al., 2016, Xu et al., 2017).

2. Physical Realizations and System Architectures

QRNG architectures vary primarily by the quantum process harnessed, device integration level, and extraction strategy.

• Integrated photonic vacuum-state QRNGs: Silicon-on-insulator (SOI) or silica photonic integrated circuits (PIC/PLC) realize homodyne detectors, combining optical splitters (e.g., 2×2 multimode-interference couplers, Y-splitters), matched photodiodes (InGaAs, Si), and on-chip or co-packaged transimpedance amplifiers. Examples include 18.8 Gbps (15.6×18.0 mm²) real-time QRNGs (Bai et al., 2021), and miniaturized chips (16.6×7.8 mm) with on-board MCU/ADC for 5.2 Mbps sustained operation across –40°C to 85°C (Qiao et al., 16 Sep 2025).
• SPAD array and CMOS/FPGA QRNGs: Large ($128\times128$) arrays of single-photon avalanche diodes illuminated by an LED, with on-chip XOR-matrix post-processing, achieve 400 Mbit/s information-theoretically secure extraction, at <500 mW power (Regazzoni et al., 2021). Flip-flop schemes with periodic sampling of toggling state, fed by Poissonian SPAD events, operate at up to 20 Mbit/s per cell, requiring minimal logic (Stipčević et al., 2021).
• Free-space/mechanical setups: On-demand optical pulse systems convert an external trigger, laser emission, and fast single-photon detection into a single bit per request, with guaranteed in-future-of-request action and <10 ns latency (Stipčević et al., 2014).
• Radioactive decay: Beta sources (Co-60, Sr-90), Geiger-Müller tube detectors, and programmable counting intervals produce raw bits (entropy >0.997 per bit), with NIST SP 800-22 pass rates up to 87.5% depending on source and geometry (Zafar et al., 2024).

Integrated systems often feature on-chip or in-package digitization and randomness extraction modules, e.g., 10-bit 2.5 GSa/s ADC plus FPGA for 18.8 Gbps throughput (Bai et al., 2021), or on-board Toeplitz-hash extraction in MCUs for compact modules (Qiao et al., 16 Sep 2025).

3. Randomness Quantification and Extraction Methods

Quantum entropy must be quantified conservatively, separating quantum from classical/statistical noise with provable lower bounds. Key tools include (Ma et al., 2015, Ma et al., 2012, Guo et al., 2021):

• Min-entropy: $H_\infty(X) = -\log_2 \max_x P(x)$ for the output distribution (raw or conditioned). Worst-case guarantees are mandatory for crypto-seed use; e.g., vacuum-state homodyne QRNG (with SNR-optimized LO) yields $H_\infty \approx 7.7$ bits/sample on a 10-bit ADC (Bai et al., 2021), $H_\infty\approx 5.8$ on 8 bit ADC (Qiao et al., 16 Sep 2025), and $H_\infty / \text{bit}\approx0.73$ for raw SPAD data (Regazzoni et al., 2021).
• Toeplitz hashing: A universal hash function—multiplying a Toeplitz binary matrix by blocks of raw bits—extracts (by the Leftover Hash Lemma) $m$ almost uniform bits from $n$ input bits when $m\leq H_\infty n - 2\log_2 (1/\varepsilon)$ (statistical distance). Implementations parallelize extraction in FPGAs to achieve multi-Gbps throughput (Bai et al., 2021, Guo et al., 2021).
• Alternative extraction (unbiasing, FIR, LFSR): In some architectures, classical finite-impulse-response (FIR) filters, XOR logic, or linear feedback shift registers suffice for flattening distributions (e.g., arcsine or Gaussian) when security requirements are not composable (Marangon et al., 2018, Moeini et al., 2023, Haider et al., 3 Jun 2025).

Extractors must be information-theoretically secure, with outputs passing stringent statistical test suites (NIST SP 800-22, Diehard, TestU01), and stability monitored in real time for operational integrity (Bai et al., 2021, Marangon et al., 2018, Regazzoni et al., 2021, Qiao et al., 16 Sep 2025).

4. Performance Metrics and Statistical Validation

State-of-the-art QRNGs exhibit wide performance variation depending on source, integration, and security model (Ma et al., 2015, Herrero-Collantes et al., 2016, Bai et al., 2021, Bruynsteen et al., 2022):

Approach	Typical Bit Rate	Integration	Notable Features
Vacuum-state homodyne (PIC)	5.2–100 Gbps	Full PIC/SiP, FPGA, MCU	Gaussian statistics, scalable min-entropy certification
SPAD array/LED (CMOS)	0.1–0.4 Gbps	Full CMOS	Hardware/post-processing co-designed; energy-efficient
Laser phase noise/interference	8–16 Gbps	Bulk or miniaturized optoelectronics	Phase decorrelation, FIR-based unbiasing
Single-photon path/arrival time	1–10 Mbps	Discrete optics; simple logic	No extractor required in bias-free assignment
Beta decay (radioactive)	≤50–500 bps	Bench-top; GM-tube	Highest entropy per bit, but inherently low throughput

Statistical testing for autocorrelation, uniformity, and randomness is universal: all competitive QRNGs demonstrate pass rates above confidence thresholds for all NIST tests; some (e.g., (Marangon et al., 2018)) are validated over multi-petabit, multi-month continuous runs without interruption, demonstrating reliability comparable to conventional TRNGs or CSPRNGs.

5. Security Models and Trust Paradigms

QRNGs are classified by the degree of hardware trust and allowed adversarial knowledge (Ma et al., 2015, Herrero-Collantes et al., 2016, Xu et al., 2017):

• Practical/trusted-device QRNGs: All components are characterized; entropy is certified via detailed physical and noise modeling. Throughputs of 10 Mbps–80 Gbps are typical.
• Self-testing QRNGs (device-independent): Security is certified by Bell-inequality violation, independent of device details. Randomness per trial is directly bounded by observed nonlocality; practical rates remain ≪1 kbps due to low event rates and experimental constraints.
• Semi-self-testing QRNGs (source- or measurement-independent): One subsystem is trusted and monitored (e.g., random switching of measurement basis). Continuous-variable source-independent protocols with vacuum-state detection have reached 15 Gbps offline and >6 Gbps real-time post-processing (Xu et al., 2017).

Information-theoretic security demands extraction be robust to adversaries with quantum or classical side information. Formalism based on the min-entropy conditioned on side information ($H_\infty(X|E)$) is standard. For scenarios with detector arrays or SPAD matrices, scalable methods for conditional min-entropy calculation (e.g., using restricted Stirling numbers) provide tight, composable bounds even when device parameters are partially adversarially influenced (Marangon et al., 2016, Regazzoni et al., 2021).

6. Practical Implications and Application Domains

QRNGs are essential in domains requiring unpredictable, certifiable randomness:

• Cryptographic key generation and PQC: Deployment in post-quantum cryptographic algorithms (e.g., ML-KEM, ML-DSA, SLH-DSA) replaces PRNG seeds with QRNG outputs, greatly improving resistance to side-channel and nonce-reuse attacks (Chen, 24 Jul 2025).
• Monte Carlo simulation and randomized algorithms: High-throughput QRNGs replace PRNGs in sensitivity-critical simulations, providing unbiased uncertainty quantification (Marangon et al., 2018).
• Quantum key distribution (QKD) and fundamental tests: Fast, in-future, and basis-selection QRNGs enable loophole-free Bell tests and continuous QKD (Stipčević et al., 2014, Ma et al., 2015).
• Embedded, mobile, and IoT security: QRNGs in CMOS or camera-artifact form factors (e.g., smartphone camera QRNGs at 1–3 Gbps) expand practical deployment in low-SWaP (size, weight, and power) systems (Sanguinetti et al., 2014, Regazzoni et al., 2021).

Integration trends emphasize modular scaling, single-package solutions, and compatibility with CMOS/fiber/photonic standards (Bai et al., 2021, Qiao et al., 16 Sep 2025). Ongoing work focuses on parallel, real-time, hardware-efficient randomness extraction (e.g., multi-channel FPGA Toeplitz hashing at >8 Gbps (Guo et al., 2021)) and rigorous composable security certification.

7. Open Directions and Comparative Analysis

• Throughput ceilings: The fastest demonstrated vacuum-state QRNG achieves 100 Gbps using PIC, high-speed TIA, 20 GS/s ADC, and FIR-based digital equalization; the limiting factors are detector bandwidth, ADC speed, and extractor throughput (Bruynsteen et al., 2022).
• Security-composability bottlenecks: Toeplitz extractors dominate high-speed applications; Trevisan's extractor offers optimal seed efficiency but slower performance, with ongoing software/hardware acceleration research (Ma et al., 2012, Ma et al., 2015).
• Device-independent rates: Fully DI-QRNGs remain orders of magnitude slower due to Bell-test constraints; hybrid semi-SI/MDI models offer practical tradeoffs (Ma et al., 2015, Xu et al., 2017).
• Side-channel considerations: Optical/thermal fluctuations, electronic crosstalk, detector bias, and adversarially controlled classical noise are modeled and monitored in real time, with built-in self-test modules issuing health status and alarms (Marangon et al., 2018, Bai et al., 2021, Qiao et al., 16 Sep 2025).
• Physical diversity: QRNGs continue to proliferate into solid-state, SPAD/MOS, LED/photodiode, nuclear, and photonic integration domains, each presenting unique tradeoffs in throughput, reliability, and integration complexity (Regazzoni et al., 2021, Moeini et al., 2023, Zafar et al., 2024).

In summary, QRNGs provide a diversity of architectures, both continuous- and discrete-variable, each with experimentally validated quantum entropy sources and robust information-theoretic extraction, enabling secure, high-rate, and operationally reliable randomness for next-generation communication, computation, and cryptography (Bai et al., 2021, Ma et al., 2015, Bruynsteen et al., 2022, Xu et al., 2017).