Skip to content

Magnetic-Field Learning Using a Single Electronic Spin in Diamond with One-Photon Readout at Room Temperature

MetadataDetails
Publication Date2019-04-29
JournalPhysical Review X
AuthorsR. Santagati, A. A. Gentile, S. Knauer, S. Schmitt, S. Paesani
InstitutionsBristol Robotics Laboratory, Collaborative Innovation Center of Quantum Matter
Citations49
AnalysisFull AI Review Included
  • Core achievement: Magnetic-field learning (MFL) used a single nitrogen-vacancy (NV) electronic spin in bulk diamond at room temperature, with an average of only one detected photon per algorithm step.
  • Magnetic-field sensitivity: The system achieved an absolute sensitivity of 60 nT·s1/2, including initialization, readout, and computational overheads, in approximately 10 ms.
  • Material platform: The experiments used two isotopically purified 12C diamond samples containing individually addressable NV centres. The sensing states were the NV electronic sublevels ms = 0 and ms = −1.
  • Method: Adaptive Ramsey precession times were selected using Bayesian particle filtering. The next precession time was approximately proportional to the inverse of the current field uncertainty, allowing the phase-accumulation time to increase as the estimate improved.
  • Multiparameter capability: The algorithm simultaneously estimated the magnetic field and the dephasing time T2∗, helping prevent measurements from exceeding the useful coherence window.
  • Dynamic sensing: MFL tracked stepwise changes in magnetic field, including an approximately 30-fold instantaneous change, and converged to the new field after approximately 10 epochs.
  • Practical significance: The work reduces the need for cryogenic single-shot readout and repeated photon collection, making room-temperature nanoscale magnetometry more compatible with biological sensing, nano-MRI, and compact instrumentation.
ParameterValueUnitContext
Operating temperatureRoom temperature—NV magnetometry without cryogenic operation
External magnetic field≈ 450GApplied approximately parallel to the NV centre axis
Electron gyromagnetic ratio≈ 2π × 28MHz/mTRelates the Zeeman frequency shift to magnetic-field strength
NV electronic statesms = 0 and ms = −1—Ramsey interferometry states
Optical excitation wavelength532nmSpin initialization and readout
Photoluminescence readout window350nsAvalanche photodiode detection interval
Readout contrast≈ 30%Approximate NV fluorescence contrast
Approximate room-temperature spin-state fluorescence difference≈ 35%Difference between optical readout probabilities
Laser-pulse overhead3µsTotal for initialization and readout
Relaxation/idle time1µsDelay between optical and microwave operations
TTL synchronization pulse20nsPhotodetector/control synchronization
Total microwave-pulse duration≈ 50nsTwo microwave π/2 pulses combined
Total experimental overhead per sequence≈ 4.07µsExcluding selected precession time
Photon detection rateApproximately 1 photon every 8 pulses—Limited by collection optics, internal reflection, and optical losses
Number of Ramsey sequences analyzed60,000sequencesLarge experimental data set used for MFL testing
Magnetic-field sensitivity, one-photon case60nT·s1/2Includes initialization, readout, and computation
Sensitivity, 20-photon case84nT·s1/2Approximately 12.8 Hz protocol repetition rate
Convergence, one-photon case≈ 500epochsFinal field uncertainty ≈ 0.45 µT; total time ≈ 18 ms
Convergence, 20-photon case≈ 150epochsFinal field uncertainty ≈ 0.3 µT; total time ≈ 78 ms
Typical one-photon running time≈ 10msPer epoch in tracking configuration
Computational overhead≈ 0.21ms/epochMFL computation in the tracking experiment
Average precession time0.41msTracking configuration
Experimental overhead16.28ms/epochExperimental routines dominate total runtime
Typical tracking convergence time≈ 10epochsAfter a stepwise magnetic-field change
Maximum demonstrated instantaneous field change≈ 30—Ratio between successive field levels in tracking test
Dynamic tracking rangeUp to 18µT/msTypical estimation error remained below 3%
Single-photon readout fidelity, state 0≈ 1.0—Effective fidelity estimate including losses
Single-photon readout fidelity, state 1≈ 0.54—Effective fidelity estimate including losses
Numerical tracking readout fidelity0.88—Readout-fidelity parameter used in simulations
Dephasing time, β1 data set16µsEstimated from Ramsey-fringe fitting
Dephasing time, ε2 data set64µsEstimated from Ramsey-fringe fitting
Dephasing-related saturation time≈ 0.1sApproximate scale where performance becomes dephasing-limited
Magnetic-field estimates, α1 data set52µTRepresentative MFL result
Magnetic-field estimates, α2 data set710µTRepresentative MFL result
Magnetic-field estimates, β1 data set8.3µTRepresentative MFL result
Field-tracking range, ε3 data set10 → 550µTStepwise magnetic-field changes
Isotopic composition12C—Isotopically purified diamond used to reduce unwanted spin noise
Field orientationNV symmetry axis—Initial experiment measured the field component parallel to the NV axis
  1. Prepare isotopically purified diamond samples containing isolated NV centres.

    • Use 12C-enriched material to reduce magnetic-noise contributions from naturally abundant nuclear spins.
    • Select NV centres suitable for confocal addressing and single-photon detection.
  2. Apply the external magnetic field.

    • Use a field of approximately 450 G.
    • Align the field approximately parallel to the NV centre symmetry axis.
    • Determine the magnetic field from the Zeeman shift of the NV spin transition.
  3. Initialize the NV electronic spin.

    • Apply a 532 nm laser pulse.
    • Prepare the sensing state using the NV optical transition.
  4. Execute the Ramsey sequence.

    • Apply a first microwave π/2 pulse to create an equal superposition of the ms = 0 and ms = −1 states.
    • Allow the spin to precess for an adaptive phase-accumulation time τ.
    • Apply a second microwave π/2 pulse to convert the accumulated phase into a spin-population difference.
  5. Read out the NV state.

    • Apply a second 532 nm laser pulse.
    • Collect red-shifted photoluminescence for 350 ns using an avalanche photodiode.
    • Normalize the photoluminescence count to obtain a likelihood estimate.
  6. Select the next precession time.

    • Approximate the next time using the particle-guess heuristic:
      • τ ≈ 1/σB, where σB is the current posterior uncertainty in the estimated magnetic field.
    • Increase τ as the Bayesian uncertainty decreases.
  7. Update the Bayesian posterior.

    • Represent the parameter distribution with a sequential Monte Carlo particle filter.
    • Compute the likelihood of the observed photon outcome for the current parameter hypothesis.
    • Reweight and resample the particles after each epoch.
  8. Estimate the magnetic field and dephasing time.

    • Begin with single-parameter estimation of the magnetic field.
    • Introduce simultaneous learning of B and T2∗ after approximately 100 epochs.
    • Select later precession times using the uncertainty of the joint parameter covariance.
  9. Process low-photon-count measurements.

    • For one-photon operation, incorporate readout infidelities and asymmetric state-assignment probabilities into the likelihood model.
    • Estimate the readout-fidelity parameter during a preliminary inference run.
  10. Track time-dependent magnetic fields.

    • Monitor whether recent observations invalidate the current posterior.
    • Reset the prior when the effective particle sample size remains below the resampling threshold.
    • Reinitialize the particle distribution when a large field change is detected.
  11. Compare against FFT analysis.

    • Apply FFT to the same Ramsey datasets for benchmarking.
    • Use cumulative FFT processing for tracking comparisons.
    • Note that FFT tends to retain the dominant spectral feature and can lag behind abrupt field changes.
6CCVD materialSuitability for the reported workRecommended qualification focus
Optical-grade single-crystal diamond (SCD)Best match for isolating and addressing individual NV centresLow fluorescence background, controlled NV creation, surface preparation, and long T2∗
Isotopically engineered SCDPreferred platform for reducing nuclear-spin noise and improving NV coherenceSpecify 12C enrichment, NV concentration, orientation distribution, and residual spin defects
Boron-doped single-crystal diamond (BDD-SCD)Not required for the demonstrated optical NV magnetometer, but potentially useful for electrically contacted sensor packagesBoron concentration, conductivity, optical absorption, and surface stability
Polycrystalline diamond (PCD)Useful for larger-area or multi-emitter sensor architectures rather than single-defect experimentsGrain size, uniformity, optical scattering, and spatial control of NV centres
Thin SCD membranes or platesSuitable for near-surface sensing and integration with microwave or photonic structuresThickness uniformity, surface roughness, NV depth, and membrane handling
Substrates and thicker diamond platesAppropriate for bulk-style experiments and stable mechanical mountingCrystal orientation, flatness, parallelism, and mounting-induced strain
  • Material: Optical-grade or isotopically purified SCD.
  • Crystal orientation: A selected orientation, such as [100] or [111], should be specified according to the required NV-axis geometry.
  • Thickness: 6CCVD can provide SCD from 0.1 µm to 500 µm and substrates up to 10 mm.
  • Surface finish: For optical and near-surface NV work, specify polished surfaces; 6CCVD can provide SCD polishing to Ra < 1 nm.
  • NV centre control: Define the target NV concentration, depth, spatial distribution, and acceptable level of other optically active defects.
  • Magnetic environment: Document the intended operating field range, field orientation, microwave architecture, and expected operating temperature.
  • Optical collection: Specify whether the design requires a single NV centre, multiple isolated centres, or a scalable ensemble.

The paper relies on several integration features that can be customized through 6CCVD’s internal processing capabilities:

  • Custom dimensions: SCD plates, wafers, membranes, and substrates can be produced according to the optical, microwave, and mechanical layout.
  • Laser cutting: 6CCVD offers laser cutting for shaping diamond components, including geometries for:
    • Confocal optical access.
    • Microwave striplines or resonators.
    • Electrical and thermal contacts.
    • Multi-sensor or array configurations.
  • Metallization: The required Ti/Pt/Au and other contact stacks can be evaluated and deposited using 6CCVD’s internal Au, Pt, Pd, Ti, W, and Cu metallization capability.
  • Surface preparation: The surface condition should be coordinated with the intended NV depth and sensing distance. Atomic-scale polishing, controlled damage layers, and post-polishing cleaning can affect both NV stability and photoluminescence.
  • Thin membranes: For near-field or scanning-probe magnetometry, a 0.1-500 µm SCD membrane can be considered, with thickness selected to balance photon collection, mechanical stability, and NV-to-sample distance.
  • PCD scale-up: Where single-defect addressability is unnecessary, PCD wafers up to 125 mm can support ensemble NV sensing and integrated sensor-array development.

Suggested stack for an NV magnetometer package

Section titled “Suggested stack for an NV magnetometer package”
ComponentCandidate material or processEngineering function
Diamond sensorOptical-grade SCD or 12C-enriched SCDNV host and optical sensing medium
Adhesion/contact layerTiPromotes adhesion between diamond and metal stack
Conductive diffusion-barrier layerPt or PdDefines conductive routing and contact geometry
Microwave conductorCu, Au, or PtDelivers microwave control pulses
Optical or thermal featuresPatterned Au/Cu or custom aperturesImproves light management, thermal control, or field access
Dielectric/passivation layerTo be specifiedProtects electrodes and reduces leakage or contamination
Mechanical supportDiamond substrate or selected carrierMaintains alignment and thermal stability

The metal stack and dielectric architecture should be selected after reviewing the microwave design, optical geometry, thermal budget, and intended readout electronics. Contact resistance, film adhesion, stress, optical absorption, and microwave loss should all be treated as qualification criteria.

6CCVD’s in-house PhD team can support material selection and application-specific development for similar room-temperature NV magnetometry projects, including:

  • Selecting between optical-grade SCD, isotopically enriched SCD, thin membranes, and larger-area PCD.
  • Defining the NV concentration and depth distribution.
  • Evaluating the effect of surface roughness, polishing damage, and contamination on fluorescence and spin coherence.
  • Selecting crystal orientation and NV-axis geometry.
  • Designing metallized structures and microwave routing around the diamond.
  • Evaluating Ti/Pt/Au, Ti/W, or other electrode stacks.
  • Comparing bulk, membrane, and ensemble sensor architectures.
  • Recommending the appropriate SCD, PCD, or BDD material for the intended electrical and optical environment.
  • Supporting prototype quantities before scaling to larger wafers or sensor arrays.
  • Advising on global delivery terms, with DDU as the default and DDP available on request.
Section titled “6CCVD specification questionnaire for a related project”
Specification itemInformation required
ApplicationSingle-NV magnetometry, ensemble magnetometry, biological sensing, nano-MRI, or other
Operating temperatureRoom temperature or reduced temperature
Field rangeExpected minimum and maximum magnetic field
Field orientationParallel to the NV axis, arbitrary orientation, or vector sensing
NV requirementSingle isolated centre, controlled few-centre array, or ensemble
Diamond typeSCD, isotopically engineered SCD, PCD, or BDD
Crystal orientation[100], [111], or other orientation
GeometryPlate, membrane, substrate, or patterned array
DimensionsLength, width, diameter, and tolerances
ThicknessTarget thickness and permitted variation
Surface finishOptical polish, Ra target, and surface damage requirements
MetallizationRequired metals, stack order, thickness, pattern, and alignment
Optical accessWorking distance, collection geometry, and wavelength
DeliveryDestination country and preferred Incoterm
Target quantityPrototype, pilot batch, or production volume

For custom specifications or material consultation, visit 6ccvd.com or contact our engineering team directly.

View Original Abstract

Machine learning enables a nitrogen-vacancy center to track a changing magnetic field with unprecedented sensitivity at room temperature, paving the way for new and practical nanoscale quantum sensing applications.

  1. 2012 - Machine Learning: A Probabilistic Perspective