Quantum Random Number Generation Using Nanodiamonds and Nanopillar-Isolated Single NV Centers
Paper details
Section titled “Paper details”| Field | Value |
|---|---|
| Journal | Nanomaterials |
| Published | 2026-03-27 |
| Authors | Oskars Rudzitis, Reinis Lazda, V. Krumins, Heinrihs Meilerts, Mona Jani |
| DOI | 10.3390/nano16070404 |
Both room-temperature diamond QRNG prototypes produced conditional min-entropy above 0.98 bits after Von Neumann debiasing. The nanodiamond ensemble reached 0.9786 bits, while the nanopillar source reached 0.9817 bits. The nanopillar source retained a larger fraction of raw data, but it had much lower photon flux. Neither dataset was long enough for full NIST SP 800-22 validation.
Random bit extraction and entropy measurement
Section titled “Random bit extraction and entropy measurement”The nanodiamond source used a time-of-arrival extractor. The authors divided arrival times into bins with width Δt = n/λ, where λ was the event rate and n was the expected count per bin. They set n = ln(2), which gives P = 0.5. Since λ changed between datasets, the bin width was recalculated. Its average value was 45 µs, with photon fluxes on the order of tens of kilo-counts per second.
The nanopillar source used a 50:50 beamsplitter. A click at detector A produced 0, and a click at detector B produced 1. The authors measured conditional min-entropy from detector click probabilities and available side information. The scale runs from 1 bit for a uniform binary source to 0 bits for a deterministic source. They also ran NIST SP 800-22 tests on each generated sequence.
Confocal setup and detection
Section titled “Confocal setup and detection”The authors built a table-top confocal microscope with a single-mode fiber instead of a pinhole. They chose the fiber for lower drift sensitivity, better spatial filtering, mode cleaning, and background rejection.
A 532 nm laser supplied 1.0 W output power. Power after the plate polarizer was 10 mW. Free-space mirrors carried the beam to a Nikon Plan Fluor 100x/1.30 Oil objective. The objective was backfilled with a beam expander and used immersion oil with a glass slide. Its ideal XY resolution was about 300 nm for red light at 650 nm. The translation stage moved in 0.25 µm X and Y steps.
A dichroic mirror, longpass filter, and single-mode fiber separated red fluorescence from the excitation light. A 50:50 beamsplitter sent fluorescence to two single-photon detectors. Each detector had a 100 µm active-area diameter, 70% efficiency at 670 nm, 45 ns dead time, and 0.2% afterpulsing probability. Their TTL signals were synchronized on an oscilloscope with a 200 µs per division time scale.
Nanodiamond ensemble performance
Section titled “Nanodiamond ensemble performance”The ensemble source used 100 nm scale fluorescent nanodiamonds with an NV concentration of about 3 ppm. A typical nanocrystal contained on the order of 100 NV centers. The authors diluted and sonicated the particles in isopropanol, then deposited a drop on a glass slide. They used this ensemble to calibrate the microscope and optimize fluorescence collection and data collection. The setup resolved NV fluorescence at about 90% contrast against background.
The brightest nanodiamond spot produced about 140 kcps. Before debiasing, detector A and B click probabilities were PA = 0.2130 and PB = 0.7870. The 31,998 registered counts gave 0.3455 bits of conditional min-entropy.
Von Neumann debiasing produced PA = 0.5001 and PB = 0.4999. It retained 4,025 counts and reached 0.9786 bits, a 7.9-fold data reduction. The raw imbalance came from multiple emitters, the larger excitation volume, background fluorescence, and unequal coupling into the two detectors. The authors selected Von Neumann debiasing because it gives uniform output without assumptions about the bias structure. Its cost was low retention for this strongly biased source.
Nanopillar-isolated single-emitter performance
Section titled “Nanopillar-isolated single-emitter performance”The second source used a diamond surface containing fabricated nanopillars. The manufacturer stated that 30% of pillars held one NV center at the tip. Other pillars held several centers or none. The protective hexagonal pattern had 40 µm sides and 2 µm spacing between pillars. The paper states that this geometry gives about 20 times higher collection efficiency than typical bulk diamond.
Under this microscope, the nanopillar sample gave about 28 times fewer counts than the nanodiamond ensemble. Its signal remained visible as a few or single NV centers per bright spot. The selected bright spot produced about 5 kcps.
Before debiasing, detector A and B click probabilities were PA = 0.4319 and PB = 0.5681. The 19,276 registered counts gave 0.8158 bits of conditional min-entropy. Debiasing produced PA = 0.4941 and PB = 0.5059. It retained 5,015 counts and reached 0.9817 bits, a 3.8-fold data reduction.
The authors treated this source as the single-emitter regime. They expected antibunching, but the experiment could not confirm it through a measurement of g(2)(0).
NIST validation and the data-length limit
Section titled “NIST validation and the data-length limit”Both datasets were shorter than the one million bits required for the full NIST SP 800-22 suite. Four tests were therefore ineligible for both sources. These were Binary Matrix Rank, Overlapping Template Matching, Maurer’s Universal, and Linear Complexity.
Random Excursion failed for both datasets. Random Excursion Variant passed for nanodiamonds but failed for the nanopillar source. The authors attributed these failures to statistical fluctuations caused by the short sequences. All eligible tests passed. They also stated that longer datasets are required before claiming full suite validation.
Source tradeoffs and unresolved measurements
Section titled “Source tradeoffs and unresolved measurements”The comparison was phenomenological rather than controlled. Nanodiamonds had higher brightness because each particle contained many emitters and a larger excitation volume. The nanopillar photon rate was limited by the 10 to 20 ns excited-state lifetime of a single NV center. Differences in raw entropy could also reflect multi-photon emission, background, signal-to-noise ratio, detector response, and setup throughput. The authors say a controlled comparison would require normalization of photon statistics and emitter number.
Von Neumann debiasing removes first-order bias, but it does not remove higher-order correlations or memory effects. The authors also did not test whether time binning introduced temporal correlations through detector dead time, afterpulsing, or slow system fluctuations. A separate coincidence measurement of g(2)(0) lacked sufficient statistics. Longer acquisition and higher throughput were required for both measurements.
They identify Peres extraction and Toeplitz matrix hashing as more efficient post-processing choices for a production system. Peres extraction processes discarded bit pairs, while Toeplitz hashing approaches the min-entropy bound with a valid entropy lower bound.
An earlier NV-center single-photon QRNG study reported about 34 random bits per second. This paper does not report a comparable end-to-end output rate. By comparison, integrated systems based on vacuum fluctuations or laser phase noise reach Mbps to Gbps. The NV approach provides a direct link between photon emission and the output, at lower throughput.
Device engineering implications
Section titled “Device engineering implications”Diamond source selection depends on whether raw count rate or intrinsic entropy efficiency matters most. The nanodiamond ensemble delivered the higher flux. It also required more post-processing because its raw detector bias was much larger. The nanopillar source gave the better raw entropy and retained a larger share of events after debiasing.
For a nanopillar QRNG, device engineering must balance emitter isolation, emission rate, collection efficiency, and background rejection. The 50:50 beamsplitter also requires balanced coupling and stable alignment. Detector dead time and afterpulsing can affect temporal correlations in either extraction method.
A stronger validation would require longer bit sequences and a measured g(2)(0) to support the single-emitter claim. The authors suggest that nanopillar optimization should maximize emission while reducing background. Their results identify this as a design target rather than a completed optimization.
Materials and support
Section titled “Materials and support”For 6CCVD, the diamond materials catalog includes MPCVD-grown single-crystal and polycrystalline diamond plates and wafers up to 125 mm. The paper’s room-temperature quantum random number generator needs an NV-active diamond surface shaped into nanopillars for emitter isolation. That specific geometry requires the custom fabrication described on the engineering support page.
View Original Abstract
Quantum random number generation (QRNG) provides fundamentally unpredictable randomness derived from intrinsic quantum processes. In this work we demonstrate two solid-state, room-temperature QRNG implementations based on nitrogen-vacancy (NV) centers in diamond, i.e., ensemble fluorescence from nanodiamonds and single-photon emission from single NV centers located at the tips of fabricated diamond nanopillars for enhanced light collection efficiency, spatial isolation and minimized crosstalk. We compare entropy rates (above 0.98 bits), statistical performance, and robustness of both approaches in our experimental setup, the results contribute to establishing diamond-based QRNG as a scalable solution for quantum-secure randomness generation.
High Concentration Quantum Grade Single Crystal Diamond
View Specs →
Single Crystal Cvd Diamond Plate
View Specs →
Boron Doped Single Crystal Diamond Plate
View Specs →References we cover
Section titled “References we cover”- High ODMR contrast and alignment of NV centers in microstructures grown on heteroepitaxial diamonds
- Sensitivity optimization for NV-diamond magnetometry
- Single Photon Randomness based on a Defect Center in Diamond
Original source
Section titled “Original source”References
Section titled “References”- 2018 - Molecular Dynamics Simulation for All [Crossref]
- 1970 - Quantum-Mechanical Random-Number Generator [Crossref]
- 1997 - An integrated analog/digital random noise source [Crossref]
- 2017 - Quantum random number generators [Crossref]
- 2023 - Optical quantum random number generators: A comparative study [Crossref]
- 2024 - A fast and robust quantum random number generator with a self-contained integrated photonic randomness core [Crossref]
- 2025 - Noise-rejecting photonic integrated circuit for robust quantum random number generation [Crossref]
- 2013 - The nitrogen-vacancy colour centre in diamond [Crossref]