WorksPhysics and Astronomy
Quartet state of the neutral nitrogen-vacancy center in diamond and optimized spin readout under near-infrared two-photon excitation
Abstract
Photoionization of the excited nitrogen-vacancy center (NV-) in diamond leaves the neutral defect in a metastable quartet state of unknown energy. Using two-photon optically detected magnetic resonance with a tunable femtosecond laser, we resolve its onset in the spin contrast at 1.23 +/- 0.02 eV and, with the ground-state threshold from the same setup, place the quartet 0.54 +/- 0.04 eV above the neutral ground state. The threshold sets the optimum excitation for spin readout at 1060-1075 nm, in the near-infrared tissue window, with 26-37% higher shot-noise-limited sensitivity than at 1030-1040 nm.
Introduction
The negatively charged nitrogen–vacancy center (NV) in diamond, a substitutional nitrogen next to a vacancy Figure 1(a), is the leading solid-state quantum sensor [1, 2, 3, 4]. Two-photon excitation in the near infrared adds intrinsic optical sectioning and deeper penetration into scattering tissue; it has been used to image fluorescent nanodiamonds in cells [5, 6, 7] and, recently, to read out the spin by optically detected magnetic resonance (ODMR) [8]. Under intense femtosecond excitation, however, the defect is ionized [9], and the spin contrast depends on which ionization channels the laser opens.
Figure 1. (a) NV center: nitrogen (N) next to a vacancy (dot), axis along , and spin density of NV on the three carbon dangling bonds (red); adapted from Ref. [19]. (b) One-electron and levels in the gap and their occupation; A removes an electron from (), Q the minority-spin electron from , leaving three parallel spins (). (c) Total-energy diagram of NV and NV with the excitation (, dotted gray), ionization (A, I, Q) and recharge (R, R) channels; orange, spin-related; blue, spin-blind; dotted, two-photon steps; gray, channels through the singlets (ionization of and , recharge into ), which leave no separate signature (see text). The singlets are drawn at an illustrative singlet–triplet gap of 0.36 eV. (d) Top: measured photoluminescence (300 mW, circles) and the NV two-photon absorption cross section of Ref. [18] (triangles), each normalized at its maximum. Bottom: the channels of (c) in the order of the cycle, each drawn on the open side of its threshold (vertical tick) for the number of photons it needs; NV excitation, I and R are open over the whole range (arrows, with the threshold beyond it); the singlet channels, whose thresholds depend on the singlet–triplet gap, are omitted; the shaded part of the Q bar marks the onset broadened by the inhomogeneous width and the laser bandwidth. Dashed line, resonance (1042 nm); green band, measured optimum.
The central channel is photoionization of the spin-polarizing excited state . Ejecting an e electron from the configuration of leaves NV in an , whose lowest state is the quartet , not the ground state [10] Figure 1(b). The existence of the quartet state has been established by electron paramagnetic resonance (EPR) measurements [11, 12], but its energy relative to the ground state remains uncertain: calculations place it 0.48–0.68 eV above [13, 14], an estimate from room-temperature photoionization thresholds gives ≈ 0.4 eV [13], and a cavity experiment at two fixed infrared wavelengths bounds it by ≤ 0.58 eV [15]. One-photon ionization of has otherwise been used at a single wavelength, 1064 nm [16].
In this Letter we resolve the onset of the NV() channel spectrally. With one tunable source we measure its threshold and the ionization potential of the ground state, IP(), on the same sample at the same temperature, which gives the quartet energy without assumptions about temperature shifts, and locate the excitation wavelength that maximizes the sensitivity of two-photon ODMR of NV.
Channels and thresholds. Figure 1(c) shows the channels in a total-energy diagram. Two photons excite NV from to (zero-phonon line eV).
Two photons ionize the ground state into NV() above IP() (channel A); this route is spin-blind. A third photon ionizes the freshly excited , either into (Q) or into (I); both act on a spin-polarized state and remove spin contrast. Excited NV is recharged by one further photon (R), the quartet by two (R). The thresholds are linked by [10]
By Eq. (1), an onset near 1.2 eV requires a final state about 0.5 eV above . Group theory and Hund’s rule leave a single candidate: the quartet, the state identified by its EPR signal [11, 12]. Channel I is allowed by direct ionization into the admixture (8–28%) of the correlated ground state [13]; an Auger process [17] may contribute.
Figure 1(d) places the thresholds on the laser axis together with the measured excitation spectrum. The optimum is expected between the last spin-destroying onset (Q, R) and the fall of the two-photon absorption of NV [18].
Experiment. The sample is a single-crystal diamond grown by chemical vapor deposition, from Diamond Elements Pvt. Ltd. [20] ([N] = 10 ppm, 0.3 mm thick), with an NV density estimated at ≈ 0.1 ppm (NV/N ≈ 1%); it is measured in a single-objective epi-configuration. The beam of a tunable Coherent Chameleon Discovery femtosecond laser (680–1300 nm, 80 MHz, 100–150 fs, 13–16 nm spectral width; wavelength measured with a high-resolution spectrometer, APE waveScan) is attenuated by a half-wave plate and a polarizing beamsplitter cube, transmitted through an 805 nm longpass dichroic mirror and focused by an Olympus Plan N 40×/0.65 objective. The photoluminescence returns through the objective, is reflected by the dichroic mirror and integrated over 690–750 nm on an avalanche photodiode. Continuous microwaves (15 dBm), modulated at 50 Hz by the lock-in reference, are delivered by a wire loop; details are given in the Supplemental Material (SM) [21]. At each wavelength the photoluminescence (minus the laser-blocked dark level) and the spin signal (the decrease of on resonance) are read at the same setting, one immediately after the other; the contrast is . All data are taken at room temperature.
Figure 2 shows a wavelength scan at 300 mW across 1027–1130 nm. and peak together at 1060–1075 nm (a). The contrast is lowest there (b): 1.27–1.30% against 1.55–1.60% at 1030–1040 nm. The shot-noise figure of merit , however, peaks in the same window (c), at 91.7–92.2 against 72.6 at 1040 nm (+26%) and 66.8 at 1030 nm (+37%). The resonances are those of NV [(d); MHz, MHz].
Figure 2. Wavelength scan at 300 mW. (a) Photoluminescence and spin signal , relative to 1075 nm; top axis, photon energy. (b) Contrast . (c) Shot-noise figure of merit . Green band, 1060–1075 nm. (d) ODMR spectrum at 1075 nm (lock-in signal, in phase; the photoluminescence decreases on resonance).
Spin contrast at equal photoluminescence. The raw contrast mixes two effects: falls as rises, and both change with wavelength. We therefore compare each point with the contrast measured at 1075 nm at the same , by varying the power from 88 to 300 mW [Fig. 3(a)], and define the contrast loss at equal photoluminescence
A loss channel that is absent at 1075 nm appears as .
Rate model. To relate the contrast loss to the channels of Figure 1 we use a rate model of the NV center in both charge states, NV and NV [Fig. 1(c)]. It follows the populations of ten levels: the and sublevels of and , the singlets and , the and states of NV, and the and sublevels of . Each laser pulse is much shorter than any decay and is treated as instantaneous: it transfers population with probabilities set by the photon fluence and the cross sections of the two-photon excitations and of the ionization and recharge channels, including the ionization of excited by the same pulse; the lowest triplet above lies 2.385 eV higher [22], so absorbs our photons only into the continuum. The cross sections depend on the photon energy through their zero-phonon thresholds, broadened by the calculated phonon sidebands [10], an inhomogeneous width and the laser spectrum. Between pulses the populations follow linear rate equations for radiative decay, intersystem crossing, the decay of the singlets and of the quartet, spin–lattice relaxation, the microwave drive, electron transfer from to nearby nitrogen donors, and the return of to by electron tunnelling from the donors, with the distribution of tunnelling rates set by the random distance to the nearest donor at [N] = 10 ppm [23]. The steady state of the 80 MHz pulse train is the fixed point of the map over one period, found numerically. The photoluminescence follows from the radiative decays of and , weighted by their emission into the detection window; and are obtained with and without microwaves and averaged over the intensity profile of the focus. The rate equations, the rates and cross sections with their sources, and the numerical methods are given in Ref. [21].
The contrast loss is zero within 1% at 1060–1075 nm and rises on both sides [Fig. 3(b)]. We fit the model to 52 contrast-loss values measured at 150–300 mW [21]; each value carries an uncertainty of 1.5%, set by the choice of the reference measurements. The free parameters are the strengths of channels A, Q and , the two-photon absorption of and the share of the detected emission, with fixed at the value obtained together with the scan of Figure 4, and scanned. With channel Q, and with the cross section of channel I tied to that of Q, , the model reproduces these values with a root-mean-square deviation of 1.4%, within that uncertainty, for [Fig. 3(b)]. Without channel Q the deviation is 9.1%, up to 29% at the shortest wavelengths. The visible onset near 1055 nm is the tail of this zero-phonon threshold, broadened by the phonon sideband, an inhomogeneous width of 20 meV and the laser bandwidth. Inhomogeneous widths of 10–40 meV, ratios –0.13 ( fits worse) and a shift of IP by move the best threshold within 1.215–1.245 eV; together this gives . The fits need a photo-recharge of (), which we take at least as strong as channel A, as calculated for the corresponding one-photon transitions [24]; then stays below 0.4% populated, as assumed for green excitation [25].
Figure 4. (a) Scan at 40 mW: (squares) and (circles) relative to 980–1000 nm. Lines: model (black) and (orange) from the fit to both together with the contrast-loss values (solid, IP = 2.63 eV), and for a saturated ground-state channel (dashed, IP = 2.54 eV); top axis, two-photon energy. (b) Energy ladder of Eq. (1): IP() from (a), from Figure 3 and give the quartet energy.
The long-wavelength side carries independent information. Beyond 1080 nm the two-photon absorption falls, and matching requires a higher photon flux per excitation, so any one-photon ionization of grows relative to 1075 nm. Channel Q is closed there. Without channel I the model gives at 1080–1130 nm, against 3–11% measured; the root-mean-square deviation of these 11 values grows from 1.7% to 6.9%, and that of all 52 values from 1.4% to 3.4%. The data thus require a channel with –, an order of magnitude below direct ionization into at equal excess energy but not negligible: it ionizes 11–22% of the excited per pulse. Beyond 1030 nm, where only the broadened tail of the Q onset remains, this is several times the fraction ionized into the quartet, so that channel I is the main ionization route of there. The correlated well accounts for it. A cavity experiment likewise requires ionization at 1524 nm, below the quartet threshold, about an order of magnitude weaker than [15].
Channels of the singlets
The singlet shelf of adds three channels in the same spectral range: ionization of the short-lived [26] after the resonance at 1042 nm [27, 28] [29, 30]; two-photon ionization of , whose one-photon threshold has been measured with visible light [31–33]; and recharge of excited into the singlets instead of , whose threshold is known from transient absorption [34]. Their thresholds depend on the singlet–triplet gap and overlap with those of Q, and I, and several of these processes act in parallel with partly opposite effects on the spin contrast. None of them leaves a separate signature within the 1.5% uncertainty, as three tests show [21]. (i) Ionization of would make the contrast loss at the 1042 nm resonance grow with power. We measure a growth of at most 6% () from 150 to 300 mW, as expected without this route (+2.5%); this bounds the route to about a third or less of an estimate from the measured absorption, the selection rules and the calculated photoionization of [10]. (ii) Our photons (0.99–1.36 eV) lie below the one-photon threshold of , so is ionized only by nonresonant two-photon absorption; its cross section is below about 1 GM (Göpp ert-Mayer unit, 1 GM = ; see SM [21]), comparable to the two-photon absorption of NV [5, 18]. (iii) Recharge into the singlets does not improve the fit. These channels are therefore omitted from the fits shown here.
The photoluminescence step below 1060 nm [Fig. 1(d)], by a factor of 2.7 and independent of power from 100 to 500 mW, is not a charge-state effect in the model: a ten times stronger channel Q lowers S at 1030 nm by only 9% and removes most of the spin contrast. We assign it to the two-photon absorption spectrum of NV, which is resolved here with a narrower source than in Ref. [18].
Ground-state threshold and the quartet energy. To obtain we scanned 910–1000 nm at 40 mW [Fig. 4(a)]. At this power the charge cycling by two-photon excitation plus one-photon ionization, which scales with one more power of the intensity, is suppressed relative to the two-photon steps, so the edge is not masked by it. The contrast collapses at 940–960 nm to of its value at 980–1000 nm, while keeps about one third. The contrast-loss values alone do not fix the threshold, which they trade against the strength of channel A; the edge does. Fitted jointly with them, it gives eV, with unchanged, consistent with the photocurrent thresholds of 2.66(4) and 2.69(3) eV [24]. The edge could alternatively be read as the saturation of channel A, with its threshold well below the edge ( eV). The best fit of this reading [dashed lines in Figure 4(a)] keeps the contrast below 930 nm at 0.32 of its value at 980–1000 nm, whereas the measured contrast falls to 0.06–0.09; this reading is therefore excluded [21].
Equation (1) then places the quartet at
[Fig. 4(b)], the uncertainties of ( eV) and of ( eV) being independent. Both thresholds come from the same spot, the same setup and the same temperature, so the temperature shift of the band edge enters both onsets alike, and both are zero-phonon thresholds. The value is consistent with the room-temperature bound of Ref. [15]: that bound, eV, assumed eV, and with our IP the same argument gives eV. It is above the eV estimate from separate photoionization experiments and within
the calculated range (0.48 eV, HSE06, and 0.58 eV, PBE [13]; 0.68 eV, configuration interaction [14]).
Optimum excitation. The window of the highest figure of merit, 1060–1075 nm, lies just beyond the broadened onset of channel Q and before the two-photon absorption falls; at 1060–1070 nm the contrast loss at equal is to . The figure of merit at 1075 nm grows as over 88–300 mW [21]. At 200 mW, 1075 nm exceeds 1030 nm by 46%. For a shot-noise-limited measurement, the gain of 26–37% over 1030–1040 nm, the wavelengths of Yb sources used so far [8], shortens the acquisition by a factor of 1.6–1.9 at equal power or, by the measured power law, allows 30–40% less power at equal sensitivity, while the water absorption is nearly the same (0.16 and 0.18 at 1040 and 1070 nm) [35]. As a consequence, both the excitation and the 637–800 nm emission lie inside the near-infrared transparency window of tissue.
Conclusion. Two-photon ODMR with a tunable source resolves the threshold of photoionization from the NV excited state into the NV quartet,
1.23 ± 0.02 eV, and, with the ground-state threshold from the same setup, the quartet energy 0.54 ± 0.04 eV. The spin contrast also requires ionization of into the NV ground state, which sets the loss beyond 1080 nm. The quartet threshold and the excitation spectrum together fix the optimum excitation at 1060–1075 nm, a recipe that applies to two-photon sensing with nanodiamonds in biological media.
Data availability. The data and the code of the rate model (GNU GPL v3.0) are openly available in the HUN-REN research data repository at doi:10.5158/ARP/T7NV6O.
Acknowledgments
This work was supported by the EUREKA project quNV2.0 (2021-17465/NP/BILAT_HU_DE_1, 2023–2026) via the National Research, Development and Innovation Office (NKFIH) under Grant No. 2020-1.2.3-EUREKA-2022-00022. A.G. acknowledges funding from the European Commission through the QuS-PARC (Grant No. 101186889) and SPINUS (Grant No. 101135699) projects. We used the AI assistant Claude (Anthropic; Opus 4.8, 5 and 5.5, Fable 5 and 5.1) under the authors’ direction to write the code for the simulations, fits and figures, to structure and analyze the data, and to contribute to drafting and editing the text [21]. The authors specified the model and the analysis, checked all equations, inspected all fits and figures, and take full responsibility for the content.
References
Supplemental Material: Quartet state of the neutral nitrogen–vacancy center in diamond and optimized spin readout under near-infrared two-photon excitation
Nikoletta Jegenyes; Pál Maák; Vladimir Verkhovlyuk; László Himics; Tamás Váczi; Miklós Veres; Adám Gali
Sample and setup
Sample. The sample is a mm, mm thick (100) diamond plate grown by chemical vapor deposition, with a nominal nitrogen concentration of 10 ppm and an NV ensemble formed during growth ( ppm). All data were taken at room temperature.
Optics. The beam of a Coherent Chameleon Discovery femtosecond laser (680–1300 nm, 80 MHz, 100–150 fs) is attenuated by a broadband half-wave plate and a polarizing beamsplitter cube, transmitted through a Thorlabs DMLP805 longpass dichroic mirror and focused by an Olympus Plan N 40×/0.65 objective. Dispersion lengthens the pulse to fs at the sample. Powers are measured before the objective; the model takes an estimated 0.8 of this power at the focus (Table S2), and 0.9 for the scan below 1000 nm (S6). The photoluminescence is collected by the same objective, reflected by the dichroic mirror, filtered by an FELH550 longpass, an FES800 shortpass and a 690–750 nm bandpass, and detected by a Thorlabs APD410A/M avalanche photodiode. is the dc level of the photodiode minus the laser-blocked dark level, read at every point.
Wavelength scale and pulse spectrum. The wavelength and the laser spectrum are measured with a high-resolution spectrometer (APE waveScan); all wavelengths of the main text and of this SM are on this scale. The spectra have a full width at half maximum of 11–18 nm (mean 15 nm at 975–1130 nm, 12 nm at 910–1000 nm).
Microwaves and lock-in detection. An HP 8753C source (15 dBm) drives a wire loop at the sample. Its output is modulated with a 50 Hz sine wave by the internal reference of a Zurich Instruments MFLI lock-in amplifier, which demodulates the photodiode signal (time constant 500 ms; 50–200 averages per spectrum). The lock-in phase is fixed for each wavelength scan and set from a full spectrum at 1075 nm, so that the resonance lies in the in-phase component. On resonance the photoluminescence decreases, and the spectra are plotted with this sign (dips). The spin signal is the magnitude of this decrease, the in-phase amplitude of the resonance, from a full spectrum or from an on-resonance record minus the off-resonance record of the same setting; the contrast is thus positive. The modulation frequency is low because the spin signal follows the microwaves with an effective response time of 0.5–1 ms, measured on this sample under two-photon excitation at 30–50 mW with the frequency-domain method of Ref. [1], which also gives its spin–lattice relaxation time ms; at 50 Hz at least 95% of the signal is kept, at 1 kHz only a fraction of it. Although 50 Hz coincides with the mains frequency, pickup is below 0.1 V: spectra without a resonance are flat to V. and are read at the same setting of wavelength and power, one immediately after the other.
Representative ODMR spectra
Figure S1 shows representative spectra. The ensemble gives a strain-split doublet with equal amplitudes of its two lines (ratio at 995–1033 nm). At 1075 nm and 150–300 mW, –2869.08 MHz, –11.5 MHz and the full width at half maximum is 6.9–7.2 MHz; at 995–1033 nm and 200 mW, –2869.22 MHz, –11.8 MHz and 7.5–8.0 MHz; at 40 mW, –2870.18 MHz, –13.4 MHz and 9.1–10.3 MHz. At a given power the line positions and widths do not change with wavelength, so the wavelength dependence of is that of its amplitude. At 40 mW the resonance is clearly visible and can be fitted at every wavelength from 1000 nm down to 950 nm; at 940 nm it is only 3.6 above the noise, and at 910–930 nm it is not significant () [Fig. S1(b)].

Figure S1. Two-photon-excited ODMR spectra (lock-in signal in phase, off-resonance level removed; negative, a decrease of the photoluminescence). (a) 1075 nm at 300 and 150 mW and 995, 1015 and 1033 nm at 200 mW. (b) 40 mW across the ground-state threshold, 1000–920 nm.
Contrast loss at equal photoluminescence
Reference curve. At 1075 nm the contrast falls from 2.13% to 1.30% as rises from 3.9 to 50 mV (88–300 mW). The reference points are measured together with the wavelength scans and fitted as
with in mV and a common shape, and . Because the absolute contrast depends on the alignment, the level is fitted separately for the reference points measured with each wavelength scan ( and ; the levels differ by 2.5%, and by 0.1% where a single reference point sets the level). The root-mean-square residual is 1.3%. Each measured point is compared with the curve measured together with it, at its own ,
so that contains only the change of the contrast with wavelength at equal photoluminescence. The 300 mW scan reaches mV at 1065–1070 nm, 4% beyond the largest reference point.
Uncertainty. Four estimates agree: the residual of the reference fit (1.3%); the change of when one reference point is omitted (up to 1.1%); the agreement of repeated settings (0.5–1.2% in at 300 mW, 1–3% at 150 mW); and a repeated scan at 200 mW, whose values at seven common wavelengths (995–1033 nm) differ by +1.6% on average with a standard deviation of 1.2%. We assign to each value; where a setting was measured twice (1030, 1042 and 1046 nm at 150 and 300 mW), combines half the spread of the two readings (at least 1%) and 1% for the reference.
Data. The 52 values are listed in Table S1 and are included as a data file with the code (S9).
| (mW) | (nm) | (%) | (mW) | (nm) | (%) | (mW) | (nm) | (%) | (mW) | (nm) | (%) |
| 150 | 1030 | 200 | 1009 | 200 | 1033 | 300 | 1065 | ||||
| 150 | 1042 | 200 | 1009 | 200 | 1036 | 300 | 1070 | ||||
| 150 | 1046 | 200 | 1012 | 300 | 1027 | 300 | 1080 | ||||
| 200 | 975 | 200 | 1015 | 300 | 1030 | 300 | 1085 | ||||
| 200 | 980 | 200 | 1015 | 300 | 1030 | 300 | 1090 | ||||
| 200 | 985 | 200 | 1018 | 300 | 1035 | 300 | 1095 | ||||
| 200 | 990 | 200 | 1021 | 300 | 1040 | 300 | 1100 | ||||
| 200 | 995 | 200 | 1021 | 300 | 1042 | 300 | 1105 | ||||
| 200 | 995 | 200 | 1024 | 300 | 1045 | 300 | 1110 | ||||
| 200 | 1000 | 200 | 1027 | 300 | 1046 | 300 | 1115 | ||||
| 200 | 1003 | 200 | 1027 | 300 | 1050 | 300 | 1120 | ||||
| 200 | 1003 | 200 | 1030 | 300 | 1055 | 300 | 1125 | ||||
| 200 | 1006 | 200 | 1033 | 300 | 1060 | 300 | 1130 |
Table S1. The 52 contrast-loss values fitted in S4–S5 (power set before the objective).
Rate model
All parameters of the model, with their values and sources, are collected in Table S2; the three input curves are shown in Figure S3.

(a)

(b)

(c)
Figure S3. Input curves of the rate model, as read by the code ((S9)). (a) Two-photon absorption cross section of : circles, the points digitized from Fig. 6 of Ref. [3] (calibrated there to 0.45 GM at 1060 nm [17]); dashed, the continuation assumed where no data exist. (b) Calculated photoionization cross sections of and without vibrational broadening, against the excess energy above threshold, read from Fig. 7 of Ref. [2]; the model uses their shapes normalized at 0.10 eV (dotted line), with the strengths of channels A, Q and fitted ((S4.2)). (c) Construction of the ionization cross section, which is derived, not calculated ((S7)): the curve of (b) shifted to the threshold eV (dotted; eV), times the one-electron weight (dashed), times the photon-energy factor (solid, the cross section used). Gray band, photon energies of the scans; dashed line, the resonance (1042 nm).
Levels and assumptions
The model follows one NV center, in either charge state, together with its nitrogen environment. The populations are those of (; together), (; ), the singlets and of NV, the and states of NV, and the and sublevels of the NV quartet ; . The assumptions are:
Each pulse ( fs at the focus) is much shorter than every lifetime and is treated as instantaneous. Within a pulse, a state created by two-photon excitation can absorb a further photon from the remainder of the same pulse; on average it sees half of the pulse fluence.
The sublevels are treated together, hyperfine structure is not resolved, and the microwave drive is an incoherent rate between and (on resonance). All rates are room-temperature values.
Photoionization of into NV conserves the spin projection by the Clebsch–Gordan rule: reaches ; reaches with weight and with weight [2]. Recharge into NV produces with random (weights , ), or, for electron capture by , the singlet manifold with the statistical weight .
The environment consists of substitutional nitrogen donors at the density [N]; interactions between NV centers are neglected ([NV] [N]). An NV center returns from NV to NV by electron tunnelling from its nearest donor, at a rate set by the donor distance; the ensemble is a sum over donor-distance classes (S4.5).
The focal average is taken over the radial intensity profile of a Gaussian focus only.
The detected photoluminescence is the emission of NV and NV into the detection window with constant fractions and ; its absolute scale is not used.
Final state of channel Q. An onset near 1.2 eV requires a final state of NV about 0.5 eV above [Eq. (1) of the main text]. Photoionization of () removes one electron, and the one-electron selection rule restricts the final state to the configuration, whose multiplets are , , and . Of these, lies 2.156 eV above , and by Hund’s rule the quartet is the lowest in energy. The final state is therefore .
Photon fluence and cross sections
Pulse quantities. A pulse enters the model through the photon flux at the centre of the focus (photons per unit area and time). A one-photon step of cross section (cm) proceeds at the rate , and a two-photon step of cross section at the rate ; has units of cm s and is quoted in Göppert-Mayer units, . Integrated over a pulse, the two steps therefore depend on two different quantities: the photon fluence (photons per unit area per pulse) and . Both follow from three assumptions. (i) Each pulse carries the energy , i.e. photons, for an average power at the focus and a repetition rate . (ii) The focus is Gaussian, with the transverse profile of radius ; its area integral is , so the fluence at the centre is . For a focus formed by the same objective at all wavelengths, scales with the wavelength, . (iii) The temporal envelope is Gaussian with full width at half maximum , with and ; then . Together,
With first-order kinetics during the pulse, a two-photon step then occurs with the probability and a one-photon step of a state present before the pulse with (S4.3). A state created during the pulse by two-photon excitation, at a rate , sees only the photons that arrive after it is created; for a pulse symmetric in time this is on average half of the fluence, hence for the ionization of and the recharge of in S4.3. As an example, 300 mW at the focus at 1040 nm, with MHz, and fs, gives and , so that a two-photon cross section of 1 GM excites with a probability of about 0.18 per pulse. The two-photon absorption of NV, , is the measured spectrum of Ref. [3] [Figure S3]. That of NV is a vibronic band with zero-phonon line ,
with a normalized Gaussian of width 4 meV for and (Table S2) for . Every photoionization and recharge channel has a threshold and an excess-energy dependence : the calculated cross section without vibrational broadening from Fig. 7 of Ref. [2] (read from the vector graphics of the published figure; Figure S3), that of for the ionization of (channel Q) and that of for all other channels. It is normalized to , so that the fitted strength is the electronic cross section 0.10 eV above threshold. This applies to channel A (, threshold ), channel Q (, threshold ) and channel Rq (, threshold ). An independent calculation of the ionization of NV [4] (dielectric function from PBE Kohn–Sham orbitals in a 512-atom supercell, threshold shifted to the HSE06 SCF value) has the same shape as the curve used here within 6% at 0.3–1.4 eV above threshold, once the thresholds are aligned, and a magnitude 1.7 times larger. Closer to threshold the two cannot be compared, because there the intra-defect transition dominates the calculated dielectric function.
Vibronic onsets of channels A and Q. For these two channels the threshold is the zero-phonon threshold, and the vibrational broadening enters as the convolution of with the spectral function of electron–phonon coupling , as in Eq. (4) of Ref. [2]. For one photon,
where is the full width at half maximum of the laser spectrum and converts it to the standard deviation of the Gaussian , and for two photons (channel A) the same with and . The ingredients, shown in Figure S2, are:

Figure S2. Vibronic onsets of channels A and Q (S4.2). (a) Spectral functions of electron–phonon coupling of (A) and (Q) calculated in [2], at 0 K (dotted, with the multiphonon tail beyond 0.30 eV) and at 300 K (solid, as used); the zero-phonon lines are clipped. (b,c) Onsets of A and Q against the energy above the zero-phonon threshold: calculated electronic cross section without vibrational broadening (dotted), its convolution with the 300 K spectral function (dashed), and the cross section used in the fit, with the inhomogeneous width of 20 meV and the laser spectrum (solid). (d) Check of the convolution at 0 K: broadened cross sections of and published in Fig. 7 of Ref. [2] (thick) and our convolution of the published unbroadened ones with the spectral functions of (a) (dashed).
The spectral functions at 0 K, , of (channel A) and (channel Q) calculated in Ref. [2]: the modes from the generating function and the modes, which couple through the dynamical Jahn–Teller effect of and , from a multi-mode Jahn–Teller treatment. They are read from the published figures up to 0.30 eV and continued beyond the multiphonon tail of linear coupling; the clipped zero-phonon line carries the remaining weight, 0.349 (A) and 0.390 (Q). As a check, the convolution of the published cross sections without broadening with reproduces the published broadened cross sections (Fig. 7 of Ref. [2]) within 4%, and their onsets within 12 meV.
Room temperature: , with the thermal kernel of linear coupling
over the calculated spectral densities [2] with Bose occupations at 300 K. This is exact for the modes and an approximation for the Jahn–Teller modes. At 300 K the zero-phonon weight drops to 0.22 (A) and 0.27 (Q), and about 7% of the weight lies below the zero-phonon line.
Near threshold the published curves contain a numerical smoothing of 30 meV [2]. We place their zero at the calculated thresholds (2.67 and 1.15 eV, 11 and 13 meV above the points where the curves leave the axis of the published figure) and use the near-threshold law of Ref. [2] below 0.06 eV.
An inhomogeneous width , common to both thresholds, describes the spread of the local electrostatic and strain fields in the ensemble. The joint fit (S6) prefers meV among 0–40 meV; 10, 30 and 40 meV are tested in S5.
For channel , for which no spectral function has been calculated, the phonon broadening is a Gaussian of width meV: Eq. (7) with in place of and . Channel I () is open throughout the range (threshold eV); its cross section is tied to that of Q, . Two mechanisms can give this channel: direct ionization into the component of the correlated ground state, which carries 8–28% of it [5], and an Auger process in which the photoexcited electron is recaptured into while an electron is emitted [6]; the rate of the latter has been argued to be strongly overestimated because the emitted electron leaves the defect before the capture [2]. In the direct picture is the admixture, and the fitted –0.13 (S5) lies within the calculated range, so direct ionization alone accounts for the strength of the channel and no Auger process needs to be invoked. An Auger contribution of similar or smaller size would enter the same effective and cannot be excluded. The recharge of into (channel R, threshold eV [7]) is
The pulse map
A pulse maps the populations as . With the per-pulse probabilities
and the fraction of freshly excited that transfers its electron to a nearby donor within picoseconds, the pulse acts as follows (for and separately).
Ground state: a fraction is excited. Of the excited centers, remain in with the same ; end in ; end in . Of the centers not excited, a fraction is ionized into .
present at the pulse: a fraction ends in and in .
: a fraction is recharged into with random .
: a fraction is excited; of these, are recharged within the pulse, into (random ) with weight and into with weight ; present at the pulse is recharged with probability in the same way.
The channels through the singlets (S7) add the excitation followed by ionization of , two-photon ionization of , and recharge into and ; they are switched off in the fits of the main text.
Rate equations between pulses
Between pulses with
Here , is the microwave rate (zero with the microwaves off), the slow electron transfer from to nitrogen, and the tunnelling return of the center considered. Electron capture by the quartet is spin-blocked: only the fraction of the and pair states correlates with .
Donor-distance distribution
The rate in Eq. (9) is the return of to by tunnelling of an electron from a neutral substitutional nitrogen donor to the center. We take the model and parameters of Ref. [8], which were fitted there to the measured recovery of after photoionization in diamonds with 10 and 180 ppm nitrogen. The electron tunnels from the nearest donor, at distance ; the process releases about 1 eV, which is taken up by the lattice, mainly through the relaxation of the donor on ionization. The rate has the form of a multiphonon transition,
with an electronic and a vibrational factor.
is the electronic transfer (hopping) matrix element between the donor state and the NV level. Both defects are described in the zero-range-potential model: outside the defect core each bound wave function is that of a free particle of effective mass , , so that is the inverse decay length of the wave function. Here refers to the N donor and to the NV center, and
The effective mass is the average conduction-band mass of diamond. The values Å and Å are the fitted values of Ref. [8]. They are comparable to the effective-mass estimates of 0.51 and 0.64 Å for the binding energies of 1.7 eV (N) and 2.6 eV (NV). At large the slower decay, , dominates, and falls by about three orders of magnitude for every 8 Å.
is the vibrational factor: the phonon spectral function of the transition, i.e., the Franck–Condon-weighted density of final vibrational states, evaluated at the released energy. It is calculated in Ref. [8] from the configuration-coordinate potential-energy surfaces of and . Its value eV at 300 K is almost that at 0 K, 1.19 eV.
For a random distribution of donors at density , the probability that the nearest donor lies between and is
Here is the nitrogen concentration [N] = 10 ppm of the sample ( cm), for which peaks at Å. Because depends steeply on , the ensemble consists of centers whose return rates span many orders of magnitude. We represent it by 12 classes of equal probability , obtained by ordering the centers by ; each class is assigned the probability-weighted geometric mean of over its members. A class slower than s does not return within the measurement and is given ; such classes are merged. At 10 ppm nitrogen concentration, the return, with , and s (nearest donor closer than about 23, 29 and 34 Å, respectively). The remaining of the centers, whose nearest donor is farther than about 34 Å, return to only through the optical recharge channels R and R. The signals are averaged over the classes with their weights (S4.6).
Steady state, signals and contrast loss
The populations just before a pulse are periodic, . With the populations just after the pulse, the detected photons per pulse are
At a point of the focus with local average power (), is evaluated with the microwaves off and on, and the signals are averaged over the radial profile of the Gaussian focus, whose area element is , and over the donor classes of weight :
with (the edge of the region contributing to the two-photon signal). The model contrast loss is computed exactly as the measured one, , where is the model at 1075 nm over 20–400 mW, interpolated in against . Only , a ratio at equal photoluminescence, is compared with the data.
Microwave rate and damping factor. The microwave rate is calibrated to the measured microwave-power dependence of the spin signal, at 1040 and 1060 nm: with we use , for which the model of the final fit gives and 0.42 at the two wavelengths (300 mW). The model then gives at 1075 nm and 300 mW, against 1.30% measured. The ratio, the damping factor , collects the reductions of the measured contrast that the model does not describe; it multiplies and and cancels in . It contains the lock-in detection of the sine-modulated signal: the fundamental has a root-mean-square amplitude of 0.27–0.35 of the on/off difference, for a signal that follows the modulation linearly and less for the sublinear dependence , depending on how the source converts the modulation voltage into microwave power. The waveform does not affect the calibration of : for a power-law dependence the fundamental scales with the same exponent as the on/off difference. The remaining factor of 0.24–0.31 is the reduction of an ensemble measurement relative to the model of a single, ideally driven center: the inhomogeneous microwave field over the focus, the different microwave coupling of the four NV orientations, the unresolved hyperfine lines, and background light. is specific to a sample and setup and can be determined independently, from the absolute contrast at one wavelength and power; it does not enter any of the fits. The fitted threshold does not depend on within a factor of 3 around the calibrated value (Table S3).
| Fit | Best (eV) | Deviation (%) |
| (main) | 1.230 | 1.17 |
| 1.230 | 1.19 | |
| 1.230 | 1.94 | |
| meV | 1.215 | 1.41 |
| meV | 1.245 | 1.17 |
| meV | 1.245 | 1.37 |
| 1.230 | 1.16; 1.17 | |
| eV; 2.66 eV | 1.230 | 1.17; 1.18 |
| held at 10 GM | 1.230 | 1.22 |
| Gaussian onsets, meV | 1.230 | 1.25 |
| Gaussian onsets, meV | 1.215 | 2.49 |
| channel I off () | 1.230 | 3.29 |
| no channel Q | – | 9.11 |
Table S3. Fits of the 52 contrast-loss values ( eV, meV, unless stated, scanned in steps of 15 meV): best threshold and root-mean-square deviation.
| Symbol | Value | Meaning | Source |
| pulse repetition rate | this work | ||
| pulse duration (FWHM) at the focus | this work (100–150 fs at the source, lengthened by dispersion) | ||
| 15 nm | laser spectral width (FWHM) | this work (13–16 nm) | |
| focal radius at 1040 nm, scaled with the wavelength | this work (objective NA 0.65) | ||
| 0.8 | fraction of the set power reaching the focus | this work (estimate) | |
| radiative decay | Ref. [9] | ||
| intersystem crossing | Ref. [9] | ||
| intersystem crossing | Ref. [9] | ||
| decay | Ref. [10] | ||
| lifetime of | Ref. [11] | ||
| 0.8 | branching of into | assumed (literature range 0.7–1.0) | |
| spin-lattice relaxation of | this sample: [1] | ||
| microwave rate on resonance, microwaves on | this work: microwave-power dependence of (S4.6) | ||
| decay | Ref. [7] | ||
| 1 | radiative quantum yield of | assumed | |
| 3.9 | Huang–Rhys factor of the band | Ref. [5] | |
| 0.065 eV | phonon energy of the band | assumed | |
| 0.06 eV | width of each vibronic replica of the band | assumed | |
| 2.156 eV | zero-phonon line of | Ref. [12] | |
| 0.35 | fraction of the NV emission in the detection window | band-shape estimate (690–750 nm) | |
| 0.02166 | fraction of the NV emission in the detection window | FIT | |
| 1.945 eV | zero-phonon line of | Ref. [12] | |
| 5.47 eV | band gap of diamond | room-temperature value | |
| 2.63 eV | ionization threshold of (two photons) | this work (joint fit with the 40 mW scan, S6) | |
| 6.988 | two-photon absorption of NV relative to NV at 1040 nm | FIT | |
| 14.61 GM | channel A, : cross section 0.10 eV above threshold | FIT | |
| 0.03 eV | Gaussian phonon smearing of the and singlet onsets (and of A with Gaussian onsets) | assumed | |
| 1.23 eV | channel Q, : threshold | FIT (scanned) | |
| channel Q: cross section 0.10 eV above threshold | FIT | ||
| 0.03 eV | Gaussian phonon smearing of the Q onset, used only with Gaussian onsets (test, S5) | assumed |
Table S2. Parameters of the rate model, as listed in the parameter file of the code (S9). FIT: fitted jointly to the 52 contrast-loss values and the 40 mW scan (fit at eV, , eV, meV).
| Symbol | Value | Meaning | Source |
| 0.13 | channel I, : | admixture of , 8–28% [5] | |
| 0.717 eV | channel R, : threshold | Ref. [7] | |
| cm | channel R: cross section at 1040 nm | assumed | |
| 0.25 | fraction of electron captures that end in the singlets | spin statistics | |
| 14.61 GM | channel R, : cross section 0.10 eV above threshold | FIT, | |
| 1 | strength of R relative to channel A, , FIT (at the limit 1); from calculated one-photon cross sections [4] (S5) | fitted within 1–100 | |
| s | lifetime of the sublevels of | geometric mean of the measured 1.78–6.06 s, room temperature [13] | |
| 3 | decay of the sublevels of relative to | assumed | |
| 0.375 | spin-allowed weight of electron capture by | spin statistics (3/8) | |
| line shape | 1 | onsets of channels A and Q: 1, vibronic (S4.2); 0, Gaussian | this work |
| 300 K | temperature of the spectral functions | room temperature | |
| 0.02 eV | inhomogeneous width (Gaussian ) of the thresholds of A and Q | this work (joint fit; tested 10–40 meV, S5) | |
| 0.3487 | zero-phonon weight of the 0 K spectral function of channel A () | normalization of the spectral function of Ref. [2] (S4.2) | |
| 0.3895 | zero-phonon weight of the 0 K spectral function of channel Q () | normalization of the spectral function of Ref. [2] (S4.2) | |
| 0.011 eV | zero of the calculated onset above the point where the published curve leaves its axis | calculated threshold 2.67 eV [2] | |
| 0.013 eV | zero of the calculated onset above the point where the published curve leaves its axis | calculated threshold 1.15 eV [2] | |
| 0.06 eV | excess energy below which the electronic onset follows | Ref. [2] | |
| [N] | 10 ppm | substitutional nitrogen (donor) density | growth specification |
| 0.01 | fast (ps) electron transfer per excitation | Ref. [7], scaled to [N] | |
| s | slow electron transfer | Ref. [7], scaled to [N] | |
| effective mass in the donor-tunnelling model | Ref. [8] | ||
| 0.4 Å | decay constant 1 of the zero-range potential | Ref. [8] | |
| 0.89 Å | decay constant 2 of the zero-range potential | Ref. [8] | |
| 1.16 eV | vibronic overlap factor at 300 K | Ref. [8] | |
| 5 eV | threshold of ; 5 eV: channel closed | off in the fits; tested in S7 | |
| 0.64 | ionization of relative to that of at equal excess energy | selection rules and composition of [5]; share of : estimate (S7) | |
| scaling | 1 | value rescaled from its photon energy, IP + excess, to (1: on) | , Eq. (3) of Ref. [2] (S7) |
| cm | peak cross section of the zero-phonon line, room temperature | measured [14]; consistent with Ref. [15] | |
| 1042 nm | zero-phonon line | Ref. [16] | |
| 4 nm | width of the zero-phonon line, room temperature | assumed | |
| 0.36 eV | energy of above , for the singlet thresholds | illustrative (S7) | |
| 0.9 | Huang–Rhys factor of the sideband | Ref. [14] | |
| 300 K | temperature of the phonon occupation in the sideband | this work | |
| 5 meV | additional Gaussian broadening of the sideband | assumed | |
| 0 cm | recharge : cross section 0.10 eV above threshold | off in the fits; tested in S7 |
Table S2 (continued).
| Symbol | Value | Meaning | Source |
| 0 GM | recharge ; cross section 0.10 eV above threshold | off in the fits; tested in S7 | |
| 0.03 eV | phonon smearing of the recharge onsets | assumed | |
| 0.0846 | damping factor: measured over model contrast; multiplies and , cancels in (S4.6) | this work: 1.30% measured, 15.4% model at 1075 nm, 300 mW |
Table S2 (continued, PDF p. 16).
Numerical Solution and Fitting
Steady state. The propagator over one period, , is a 10 × 10 stochastic matrix; the matrix exponential is computed by scaling and squaring (Padé approximant), and is its eigenvector with eigenvalue 1, normalized to unit sum. The focal integral is evaluated by Gauss–Legendre quadrature in with eight nodes; with four nodes changes by less than 0.01%, and the result agrees within 0.05% with an even grid of 96 points.
Fit. The model is fitted to the 52 contrast-loss values of S3 by minimizing with (values measured twice carry their own uncertainties). For 52 values and five free parameters, a model that describes the data within the stated uncertainty gives . The fits of the 52 values alone give –33 (deviation 1.16–1.19%), so the 1.5% uncertainty is somewhat conservative; the joint fit of the main text gives for them (1.40%). We regard differences between fits as not significant and as a clear preference. The free parameters are the strengths of channels A, Q and (, , and in with ; see Fate of the quartet), the two-photon absorption and the emission fraction ; positive parameters are varied in , within bounds, by a trust-region reflective least-squares method with a finite-difference Jacobian. The 52 values do not determine , which they trade against the strength of channel A; it is fixed at 2.63 eV from the joint fit with the 40 mW scan (S6), and the fit shown in the main text and in Table S2 is that joint fit. The threshold is not a free parameter: the fit surface is flat along and correlated with , so is scanned on a grid (1.17–1.30 eV) and each grid point is fitted separately. The quality of a fit is quoted as the root-mean-square deviation .
Tests. Table S3 lists the best threshold and the deviation of the fits for the ratio , the inhomogeneous width , the microwave rate, , a fixed fast , the Gaussian onsets used before the vibronic line shape (S4.2), channel I switched off, and channel Q removed. Inhomogeneous widths of 10–40 meV are accepted, with the best threshold at 1.215–1.245 eV. Ratios and 0.13 fit equally well, 0.28 clearly worse. Neither the microwave rate (within a factor of 3 around its calibrated value, S4.6), nor (within eV), nor a fixed of 10 GM changes the threshold. Gaussian onsets of 30 meV give the same best threshold, 1.23 eV, and a sharp Gaussian onset (15 meV) is excluded. Over the accepted fits the best threshold lies in 1.215–1.245 eV, which gives eV. Without channel I the long-wavelength contrast loss is not reproduced (main text); without channel Q the short-wavelength rise is not.
Fate of the quartet. Once formed, can return to by photo-recharge (, threshold eV, two photons here), by electron capture from a nearby donor (spin-blocked to the weight ), or decay to , which is spin-forbidden and must release about 0.6 eV. Table S4 lists refits of the 52 values at eV in which these routes are changed. Without photo-recharge the fit is clearly worse (). Its strength is hardly constrained by the 52 values: between 0.2 GM (free) and 10 GM, changes by 3. The joint fit with the 40 mW scan prefers a weak recharge, 0.1 GM, by over 10 GM. Such a weak recharge is not plausible physically: the recharge of into is fully allowed for direct ionization [5], and the calculated one-photon cross sections for the emission of a hole from exceed that for the ionization of above threshold [4]. In the dielectric functions of that calculation, the emission of a hole from exceeds the ionization of by a factor of 3–7 at equal energy above the two thresholds, which lie within 0.1 eV of each other. We therefore take at least as strong as channel A, with , assuming that this ordering of the one-photon cross sections holds for the two-photon ones. The joint fit places at this limit, GM; the 52 values then give , the value expected for them. The steady-state population of is then small: averaged over the focus, at most 0.4% at 975 nm and 300 mW, 0.06% at 1030 nm and 0.02% in the optimum window. With the weak recharge preferred without the constraint it would reach about 10% at 975–1000 nm; the resonance of at 2D GHz ( MHz [18]) would show it directly. The threshold and the quartet energy are the same in both cases: with held at 10 GM the joint fit gives –1.23 eV and a quartet energy of 0.50–0.55 eV. The decay time to , varied over the measured 1.78–6.06 s [13], does not affect the fit; equal decay of the two sublevel pairs is worse (). A small population of agrees with the calculation that the quartet lives for microseconds only in the dark [5], and with the rate model of one-photon green (520 nm) excitation in Ref. [19], which found no measurable population of and omitted it; there a single 2.38 eV photon already exceeds the threshold of . Because returns to with random , the cycle Q followed by changes no charge state but randomizes the spin, at a rate that grows with intensity.
| Case | Deviation (%) | |
| at least as strong as channel A (; main) | 1.17 | 31.7 |
| free, not tied to A (0.2 GM) | 1.16 | 31.1 |
| held at 10 GM | 1.22 | 34.4 |
| held at 1 GM | 1.17 | 31.6 |
| switched off | 1.31 | 39.6 |
| decay time to 1.78 or 6.06 s (instead of 3.3 s) | 1.17 | 31.7; 31.9 |
| equal decay of both sublevel pairs () | 1.31 | 39.7 |
Table S4. Refits of the 52 contrast-loss values at eV with the routes out of NV changed.
Strength of channel I. The fitted – is small relative to channel Q: is 0.1 eV above its threshold, comparable to the calculated cross section of at that excess energy ( [2]), whereas applies 0.4–0.6 eV above its own threshold, where the calculated cross section is 2–3 times larger. At equal excess energy, channel I is therefore about an order of magnitude below direct ionization into , as expected for a channel that requires the minority component of or a two-body process. For scale, the strongest optical transition of the center, the absorption of NV, has at 532 nm [17], about 20 times and 130–240 times . In absolute terms it is not negligible: at 1030–1110 nm, where the onset of Q is only its broadened tail, channel I ionizes 11–22% of the freshly excited per pulse at 150–300 mW (centre of the focus), several times more than Q, and it is the main source of NV besides channel A and electron transfer to nitrogen (NV fraction 0.05–0.08 at 1030–1130 nm in the model).
This is consistent with the large NV fraction under one-photon green excitation, where the steady-state NV population is at most 75% [20] (about 70% in Ref. [21]). There, channel Q can itself feed the NV ground state: the photo-recharge of needs 2.27 eV or more, so a 532 nm photon (2.33 eV) lies only about 0.06 eV above its threshold, and depending on the intensity part of the quartet population relaxes to within microseconds instead of being recharged [5]; ionization of , whose threshold is near 2.3 eV [22, 23], adds a further route. In the present pulsed experiment two photons of the same pulse recharge the quartet far above threshold, so Q returns the center to NV, and a sizeable NV population requires channel I. The charge ratio under green light thus measures the sum of all routes into NV, not the strength of channel I.
Comparison with other experiments. A cavity experiment with green light and an infrared field at 966 or 1524 nm, analysed with a rate model that resolves both charge states, reaches the same picture [13]. Ionization of into is driven by a single 966 nm photon (1.28 eV), which bounds the threshold of channel Q by 1.28 eV, consistent with eV. At 1524 nm (0.81 eV), below that threshold, the charge-state dynamics require an effective one-photon ionization of , assigned there to an Auger process into ; at both wavelengths the rates into exceed those into by an order of magnitude [13], as for channels Q and I here. We compare these ratios only, not the absolute rates, which refer to the cavity field. The same model gives an intrinsic rate of 165–562 kHz (lifetime 1.78–6.06 s) and a green-light recharge of of 438–1048 kHz/mW [13]; at milliwatt green powers both routes out of the quartet are comparable, in line with the partial relaxation to discussed above. Picosecond pulses at 1064 nm (1.165 eV) quench of single centers in nitrogen-poor diamond into a dark state that returns to or within nanoseconds; there this was attributed to excitation rather than ionization, because the emission recovers and keeps the NV spectrum [24, 25]. No bound triplet lies within 0.99–1.36 eV above , the lowest higher triplet being 2.385 eV above it [26], so the nature of this state is open. A charge-neutral process of this kind would act on in our range as channel I does, bypassing the intersystem crossing; in the present fits its effect would be absorbed into the effective . The charge-resolved data of Ref. [13] show that a channel below the quartet threshold does produce NV.
Inside the fit: charge cycle and the role of the donors. Figure S4 decomposes the steady state of the fit into the channels that carry the charge cycle. For each channel the flux per pulse period is computed from the periodic populations (for the rate equations, from their exact time integral over the period). The fluxes are averaged over the focus and the donor classes as the signals are. Ionization and recharge balance to . Along the 300 mW scan [Fig. S4(a,b)], two cycles carry the charge:

Figure S4. Charge cycle of the fit of the main text. (a) Shares of the channels that ionize NV along the 300 mW scan: two-photon ionization of (A), ionization of into (I) and into (Q), and electron transfer from to nitrogen ( and ). (b) Shares of the channels that recharge NV: photo-recharge of (R) and of (R), and tunnelling return from the nearest donor (, from and from ). (c) NV fraction, time-averaged, along both scans. (d,e) As (a,b) along the 40 mW scan (laser width 12 nm), colors as in (a,b). (f) ODMR contrast of the 40 mW scan relative to 990–1000 nm, measured (S6) and for the fit of the main text (solid), without donor return at the same parameters (dotted), and without donor return refitted (dashed; Table S5). All quantities are averaged over the focus and the donor classes as the signals are.
Below about 995 nm, channel Q followed by , which changes no charge state on balance but randomizes the spin (S5, Fate of the quartet).
Above 1040 nm, channel I followed by the recharge R; channel I then provides 87–95% of the ionization.
Electron transfer to nitrogen ( and ) contributes 1–5% of the ionization, and channel A at most 13%, at the shortest wavelengths only. The donor return provides 10–13% of the recharge. All of it comes from the quartet of the centers with a donor within about 34 Å (S4.5): it carries all of the recharge of the fastest class and about a quarter of that of the next one (1075 nm). The NV fraction stays at 0.05–0.13 [Fig. S4(c)].
The 40 mW scan is different Figure S4,Figure S4). Below about 940 nm channel A provides 96–99% of the ionization and drives about 80% of the centers into NV. The optical recharge R is weak at this power, so the donor return carries most of the way back (78–97%). The contrast floor below the edge is therefore set by the competition between channel A and the donor return. The centers without a nearby donor are pumped into NV, and those with one keep part of their NV population.
Table S5 tests how much each of these ingredients matters. Each is changed with the other parameters of the fit kept (no refit), and then the joint fit of S6 (IP and the strengths free, eV; three starting points, the best kept, since IP and the strength of channel A trade against each other) is repeated.
| Change | NV | IP (eV) | Quartet (eV) | |||||
| none | 0.064 | 45.4 | 19.8 | 2.628 | 45.2 | 20.0 | 41.0 | 0.547 |
| no donor return, ([N] ppm) | 0.075 | 46.7 | 40.0 | 2.572 | 41.0 | 17.3 | 51.2 | 0.603 |
| 0.066 | 45.6 | 21.0 | 2.605 | 44.8 | 22.9 | 42.2 | 0.570 | |
| 0.061 | 45.3 | 19.3 | 2.640 | 45.2 | 17.9 | 38.9 | 0.535 | |
| [N] = 100 ppm | 0.018 | 51.8 | 36.8 | 2.663 | 50.0 | 12.0 | 69.7 | 0.512 |
| no cutoff of (all classes return) | 0.064 | 45.4 | 19.8 | 2.631 | 45.2 | 19.7 | 41.3 | 0.544 |
| no fast transfer, | 0.059 | 46.6 | 19.9 | 2.629 | 46.2 | 20.0 | 41.0 | 0.546 |
| no slow transfer, | 0.063 | 45.6 | 19.8 | 2.631 | 45.5 | 19.7 | 41.3 | 0.544 |
| no spin blockade, | 0.064 | 45.4 | 19.8 | 2.631 | 45.2 | 19.7 | 41.3 | 0.544 |
Table S5. Sensitivity of the fit to the donor and electron-transfer ingredients. Without refit: the fit of the main text with one ingredient changed; NV fraction at 1075 nm and 300 mW, and of the 52 contrast-loss values () and of the contrast of the 40 mW scan (10 points). Refit: the joint fit of S6 to and to and of the scan ( eV; IP and the strengths free, best of three starting points). Quartet: . The refits of the unchanged fit and of the two changes without effect (no cutoff, no spin blockade) agree within 0.003 eV in IP and 0.1 in the total , which indicates the precision of the refit columns.
The 52 contrast-loss values hardly depend on the donors. Removing the donor return, changing its rate tenfold in either direction, dropping the cutoff of S4.5, the electron transfer to nitrogen or the spin blockade of the quartet changes by at most 0.5%, and of these values by at most 1.3 without refit. The threshold , which these values determine, is therefore fixed independently of the donor model.
The 40 mW scan depends on the donors. Without any return (the case of [N] ppm, where no class exceeds s), the same parameters leave too little contrast below the edge Figure S4, dotted]. The refit restores it with a much weaker channel A and moves IP to 2.572 eV, at nearly the same total (109 against 106). At a tenth or ten times the calculated rate the refit gives IP = 2.605–2.640 eV.
A tenfold higher nitrogen density, 100 ppm, keeps NV near 2% and describes the data clearly worse even after the refit (total ).
For the sample studied ([N] = 10 ppm) the donor return exists, and the variants with return give quartet energies of 0.535–0.570 eV, within 0.54 ± 0.04 eV. The limit without return marks how far IP could move if the donor model were grossly wrong: the data alone do not exclude it (). Scanned over in this limit, the joint fit keeps the best threshold at eV and gives IP = 2.57 eV, so that the quartet energy would be 0.60 eV, just above the quoted range.
Ground-state ionization threshold
Data. The scan covers 910–1000 nm in 10 nm steps at 40 mW, with full ODMR spectra and read at every point (Table S6). At this power is 0.18–0.60 mV above a dark level of 32.5 mV, whose drift over the scan (0.02–0.04 mV) is included in the uncertainty of . The 1075 nm reference at the start and the end of the scan gives and 0.85% and and 2869.84 MHz.
| (nm) | 910 | 920 | 930 | 940 | 950 | 960 | 970 | 980 | 990 | 1000 |
| (eV) | 2.725 | 2.695 | 2.666 | 2.638 | 2.610 | 2.583 | 2.556 | 2.530 | 2.505 | 2.480 |
| 0.09(6) | 0.06(6) | 0.08(5) | 0.30(9) | 0.37(6) | 0.58(4) | 0.68(5) | 0.84(3) | 1.05(4) | 0.95(6) | |
| 0.48 | 0.38 | 0.33 | 0.31 | 0.43 | 0.48 | 0.78 | 1.03 | 0.97 | 0.79 |
Table S6. 40 mW scan: contrast relative to its mean at 990–1000 nm and photoluminescence relative to its mean at 980–990 nm.
The edge. The contrast falls from its plateau at 980–1000 nm to a floor of 0.06–0.09 below 930 nm; it crosses half of the plateau at 956 nm ( eV). The photoluminescence does not follow: it drops to about one third at 930–940 nm. The remaining light carries no spin information, as expected when the ground state is ionized into NV. faster than it is repumped and polarized. Below 930 nm rises again, to 0.48 at 910 nm, which the model does not reproduce (0.19 at 910 nm; these points give about half of the of in Table S7). The contrast, which fixes IP, is not affected: the fit to alone gives nearly the same IP (2.64 against 2.63 eV). Over the edge the laser spectrum (12 nm) and the inhomogeneous width (20 meV) broaden by about eV, and the phonon sideband shifts the rise of channel A above its zero-phonon threshold (S4.2); the edge itself, 940–960 nm, spans –2.64 eV.
| Fit | IP (eV) | Channel A (GM) | Deviation of (%) | , | , |
| 52 values only, IP held | 2.60 | 0.7 | 1.17 | 20 | 461 |
| 52 values only, IP held | 2.66 | 1.8 | 1.18 | 16 | 731 |
| joint, of the scan | 2.642 | 1.4 | 1.17 | 12 | 814 |
| joint, and of the scan | 2.631 | 15 | 1.39 | 20 | 41 |
| joint, , and at nm only | 2.652 | 46 | 1.39 | 17 | 49 (21) |
| joint, and , IP held (main text) | 2.63 | 15 | 1.40 | 20 | 41 |
| joint, and , IP held (saturated reading) | 2.54 | – | 1.58 | 68 | 53 |
Table S7. Fits of the 52 contrast-loss values together with the 40 mW scan (10 values each of and ), eV, , meV: threshold IP, strength of channel A, deviation of the 52 values, and of the scan (of at all ten wavelengths; in parentheses, at the eight fitted ones).
Joint fit. The 52 contrast-loss values do not fix IP: refitted with IP held at 2.60, 2.63 and 2.66 eV, they give the same deviation (1.17–1.18% at eV), while the strength of channel A rises from 0.7 to 1.5 and 1.8 GM; such fits miss the photoluminescence of the 40 mW scan. This scan removes the degeneracy, because its contrast floor and photoluminescence depend on where the threshold lies relative to the edge. We therefore fit both data sets together, with IP free in addition to the five parameters of S5, the conditions of this scan (laser width 12 nm, 0.9 of the set power at the focus), and scanned (Table S7). With the contrast and the photoluminescence of the scan, the best threshold is – eV and IP = 2.63–2.66 eV; with the contrast alone, the fit leaves the photoluminescence far off (). We take eV, which covers these fits, the inhomogeneous widths of S5 and the variants of the donor model (Table S5) and lies within the edge. It agrees with the photocurrent thresholds of NV ionization in irradiated type-Ib diamond, 2.66(4) and 2.69(3) eV [4]. Fitting only at nm, where the model reproduces it (see The edge), gives IP = 2.65–2.66 eV and a quartet energy of 0.51–0.52 eV at the same eV, inside the quoted ranges, and a contrast floor of 0.08–0.09 (Table S7). The fit of the main text has IP held at 2.63 eV and includes both and of this scan at all wavelengths; it leaves a floor of 0.12, slightly above the measured 0.06–0.09.
Saturated reading. The alternative reading places the threshold well below the edge, which then marks the saturation of channel A. With IP held at 2.54 eV the best joint fit leaves a contrast floor of 0.32 below 930 nm, against 0.06–0.09 measured, and describes both data sets worse (Table S7); this reading is excluded.
Channels through the singlets
Figure S5 collects the photoionization and recharge channels of the two charge states and places their thresholds on the laser axis for an illustrative singlet energy eV ( eV). The channels through the singlets have thresholds within the range of the data for any between 0.30 and 0.42 eV. The – gap of 0.334–0.375 eV, inferred from calculated and measured intersystem-crossing rates [27], gives with the zero-phonon energy of 1.945 eV –0.42 eV, at the upper end of this range. The singlet channels were tested with the parameters of the fit of the main text, without refitting (Table S8).

Figure S5. (a) Total-energy diagram with the photoionization () and recharge () channels, numbered; lifetimes of the initial states. (b) Threshold of each channel as a laser wavelength (one or two photons) with its open side; filled symbols, onset seen in the data; open symbols, not seen or not resolved. Gray bars, spans of the scans; vertical lines, the zero-phonon line and the long-wavelength cut-offs of the two-photon absorption of and .
| Channel | Strength | Model | Measured |
| ionization: change of from 150 to 300 mW (%) | |||
| full | 8.6–9.8 | (2) | |
| 5.9–6.6 | |||
| 3.8–4.4 | |||
| 2.9–3.1 | |||
| off | 2.5 | ||
| full, or 0.24 eV (closed) | 2.5 | ||
| full, eV | 7.3 | ||
| full, eV | 8.8 | ||
| two-photon ionization: at 150 / 300 mW (%) | |||
| 1 GM | 4.8–5.5 / 10.8–13.8 | 6.4 2.0 / 6.9 2.0 | |
| 3 GM | 7.2–8.7 / 16.2–20.9 | ||
| off | 3.2 / 6.0 | ||
| Recharge into or : deviation of the 52 values (%) | |||
| free | 1.15–1.18 | – | |
| off | 1.17 | – |
Table S8. Singlet channels, with the parameters of the fit of the main text ( eV), for , 0.36 and 0.42 eV.
Ionization of at the resonance. A center shelved in absorbs one photon at the 1042 nm resonance: a Lorentzian zero-phonon line of 4 nm width with its phonon sideband [14] and the measured room-temperature peak cross section cm [14], consistent with the calculated radiative lifetime of [15]. Within the same pulse can absorb a further photon; under continuous illumination its 100 ps lifetime makes this unlikely [5]. By the one-electron selection rule, ionization of reaches only : removing an electron from the part of (about 90%) gives the part of (71% [5]), whereas () cannot be reached from a singlet and lies far above the threshold. We therefore take 0.09 0.71 = 0.064 the calculated cross section of [2] at the same excess energy, above the threshold IP – – 1.190 eV (1.02–1.14 eV, 0.05–0.17 eV below the photon energy). The curve is appropriate for also because the two states have nearly the same equilibrium geometry: relative to that of , relaxes by only 16 meV, against 256 meV for [27]. At equal excess energy the final conduction-band states, their density and the one-electron matrix elements are those of the calculation, as assumed in Ref. [2] for the ionization of ; the cross section, , then differs only by the photon energy, and we rescale the value from its own photon energy, IP plus the excess energy, to , a factor 0.43 at the resonance (constant ; a constant momentum matrix element would invert this factor). This cross section is thus derived, not calculated; its construction is shown step by step in Figure S3. The signature is a contrast loss at 1042 nm that grows with power. Measured, at 150 mW and 4.6 1.4% at 300 mW, a growth of at most 6% [2]; the model without this route gives +2.5%, with it +8.6 to +9.8% (3.1–3.7 above the measured growth). The route is close to saturation, so this value hardly depends on the strength: scaled by 0.3 it gives +5.9 to +6.6%, at the bound, by 0.1 +3.8 to +4.4%, by 0.03 +2.9 to +3.1%. The data thus bound the route to about a third of this estimate or less. The route would close only if lay 0.24 eV or less above , which places the threshold at or above the photon energy; the measured ionization energies, –0.38 eV (Sec. S6 and Refs. [22, 23]), exclude this, and over 0.27–0.42 eV the route stays open (Table S8).
Two-photon ionization of . Off resonance, is ionized by two photons above IP , 2.27 eV for eV ( at 1092 nm); the tests with –0.42 eV cover thresholds of 2.21–2.33 eV (1064–1122 nm). Because the shelf population grows with power, the channel adds a contrast loss that grows faster than linearly with power at short wavelengths. The excess measured is 6.4 2.0% at 150 mW and 6.9 2.0% at 300 mW; the model gives 3.2 and 6.0% without the channel, 4.8–5.5 and 10.8–13.8% with 1 GM, and 7.2–8.7 and 16.2–20.9% with 3 GM. At 300 mW, 1 GM already exceeds the measurement by 2.0–3.4; the two-photon cross section is therefore below about 1 GM.
Comparison with the one-photon measurements of the threshold. The ionization energy of has been measured with one visible photon: 1.91–2.25 eV from the charge-resolved photoluminescence after the singlet is populated by spin control and illuminated with filtered visible pulses [28]; 2.25–2.33 eV for 1.6–300 K from magnetic-field-modulated photoluminescence [22]; and, with the same method, 2.29–2.33 eV for 50–150 K [23]. In the field-modulation experiments continuous visible light excites the center and, at the same photon energy, ionizes in a single step at any time during its lifetime. A magnetic field mixes the spin sublevels of and thereby switches the population on and off, and the change of the emission between the two settings isolates the ionization of from all other channels [22, 23]. Three differences remove this signature from the present experiment. (i) Our photons (0.99–1.36 eV) lie below the one-photon threshold, so is ionized only by two photons absorbed within one pulse, a nonresonant process whose cross section the data bound below about 1 GM. (ii) For the bracket 2.25–2.33 eV, which holds up to room temperature [22], the two-photon channel opens at 1064–1102 nm, around the window where the contrast loss vanishes within the 1.5% uncertainty; just above threshold the cross section rises from zero and is broadened by the laser spectrum and by phonons, and it becomes appreciable only above threshold where it acts together with channel Q and the resonance. (iii) We detect the spin contrast in the band, to which the ionization of contributes only together with channels Q, I and R and the ionization of ; no setting of the present experiment changes the population alone. The tests above cover two-photon thresholds of 2.20–2.32 eV, which include most of that bracket; the lower values of Ref. [28] would open the channel over the whole range and are not tested separately.
Recharge into the singlets. Recharge of excited can end in (one photon, threshold ) or (two photons) instead of . Fits of the 52 values with either channel made explicit and its strength free give deviations of 1.15–1.18%, against 1.17% for the same model with both strengths at zero; in most cases the fitted strength goes to zero. The data neither require nor exclude these channels.
Photoluminescence step and optimum
Photoluminescence step. Below 1060 nm falls by a factor of 2.7 at every power (Table S9). The step lies between 1045 and 1060 nm and is flat below 1040 nm; it is reproducible and independent of power from 100 to 500 mW. In the rate model a stronger channel does not deplete by this factor, because the quartet is recharged within the same pulse: a ten times larger cross section lowers at 1030 nm by only 9% and removes 70% of the spin contrast, whereas the measured contrast loss at 1030 nm is about 10%. We therefore attribute the step to the two-photon absorption spectrum of .
| Power (mW) | (nm) | |
| 100 | 1070 | 0.37–0.40 |
| 150 | 1070 | 0.35–0.37 |
| 200 | 1070, 1075 | 0.38–0.39 |
| 300 | 1070 | 0.36–0.37 |
| 500 | 1057 | 0.34 |
Table S9. Photoluminescence at 1030 nm relative to that at the reference wavelength .
Optimum against power. Figure S6 shows the shot-noise figure of merit against wavelength at 200–500 mW, normalized at 1030 nm. The 300 mW scan resolves the maximum, at 1070 5 nm. The 200 and 500 mW scans end at 1063 and 1057 nm; up to there the figure of merit rises from 1030 nm by 77% and 24%, and at 500 mW it follows the 300 mW scan to within 8%. At 200 mW, 1075 nm exceeds 1030 nm by 46%. At 1075 nm the figure of merit grows as over 88–300 mW and as over 150–300 mW [Figure S6].

Figure S6. (a) relative to 1030 nm against wavelength; at 200 mW two scans are shown. Green band, 1060–1075 nm; dotted lines, the zero-phonon line and the visible onset of channel Q. (b) at 1075 nm against power with power-law fits; square, 500 mW at 1057 nm, the end of its scan.
Code and data
The rate model, the fitting procedure, the parameter set of Table S2, the 52 contrast-loss values and the 40 mW scan of S6 are available as a Python package in the HUN-REN research data repository at doi:10.5158/ARP/T7NV6O (code under the GNU General Public License v3.0, data under CC BY 4.0). All fixed parameters and the literature curves (the NV two-photon absorption of Ref. [3], and the photoionization cross sections, spectral functions and spectral densities of Ref. [2]) are read from CSV files, so that the calculation can be repeated for other samples (for example another nitrogen density) or other excitation conditions. For each wavelength and power the model returns
, , and . The package includes the scripts that recompute every model number and model curve of the main text and of this SM, and the model curves of Figure 3(b) and Figure 4(a) of the main text. The package also contains the raw data: the lock-in files of all ODMR records and spectra behind the 52 contrast-loss values, the 1075 nm reference series and the 40 mW scan, with an index that lists the wavelength, power, , and of every measurement; from this index the reference curves, the 52 values and the scan of S6 are recomputed.
Figure 3. (a) Contrast against photoluminescence: 1075 nm at 88–300 mW (open circles) with the reference fit, and the 300 mW scan colored by wavelength. (b) Contrast loss at equal for the 300 mW scan (error bars, 1.5%), with the rate model fitted to all 52 values, without (dashed) and with (solid, eV) the channel; .
Use of AI tools. The scripts for the numerical solution of the rate model, the fits and the tests, and those for structuring and plotting the data, were written by the AI assistant Claude (Anthropic; models Opus 4.8, 5 and 5.5, Fable 5 and 5.1, used through Claude Code) following the model specified by the authors: levels, channels, assumptions and parameter choices. The authors checked all equations and inspected all fits and figures.
References
- [1]V. Verkh...
- [2]L. Razinkovas, M. Maciaszek, F. Reinhard, M. W. Doherty, and A. Alkauskas, Phys. Rev. B 104, 235301 (2021).
- [3]P. Ji, R. Balili, J. Beaumariage, S. Mukherjee, D. Snoke, and M. V. G. Dutt, Phys. Rev. B 97, 134112 (2018).
- [4]E. Bourgeois, E. Londero, K. Buczak, J. Hruby, M. Gulka, Y. Balasubramaniam, G. Wachter, J. Stursa, K. Dobes, F. Aumayr, M. Trupke, A. Gali, and M. Nesladek, Phys. Rev. B 95, 041402(R) (2017).
- [5]G. Thiering and Á. Gali, J. Appl. Phys. 136, 084401 (2024).
- [6]P. Siyushev, H. Pinto, M. Vörös, A. Gali, F. Jelezko, and J. Wrachtrup, Phys. Rev. Lett. 110, 167402 (2013).
- [7]M. T. Luu, A. T. Younesi, and R. Ulbricht, Mater. Quantum Technol. 4, 035201 (2024).
- [8]A. Alkauskas, C. G. Van de Walle, L. Razinkovas, and R. Ulbricht, Phys. Rev. Lett. 137, 136401 (2026).
- [9]L. Robledo, H. Bernien, T. van der Sar, and R. Hanson, New J. Phys. 13, 025013 (2013).
- [10]R. Ulbricht and Z.-H. Loh, Phys. Rev. B 98, 094309 (2018).
- [11]V. M. Acosta, A. Jarmola, E. Bauch, and D. Budker, Phys. Rev. B 82, 201202(R) (2010).
- [12]M. W. Doherty, N. B. Manson, P. Delaney, F. Jelezko, J. Wrachtrup, and L. C. L. Hollenberg, Phys. Rep. 528, 1 (2013).
- [13]P. K. Shandilya, V. K. Kavatamane, S. Flågan, D. P. Lake, D. Sukachev, and P. E. Barclay, arXiv:2411.10638v2 (2026).arxiv.org/abs/2411.10638
- [14]P. Kehayias, M. W. Doherty, D. English, R. Fischer, A. Jarmola, K. Jensen, N. Leefer, P. Hemmer, N. B. Manson, and D. Budker, Phys. Rev. B 88, 165202 (2013).
- [15]M. Bockstedte, F. Schütz, T. Garratt, V. Ivády, and A. Gali, npj Quantum Mater. 3, 31 (2018).
- [16]L. J. Rogers, S. Armstrong, M. J. Sellars, and N. B. Manson, New J. Phys. 10, 103024 (2008).
- [17]T.-L. Wee, Y.-K. Tzeng, C.-C. Han, H.-C. Chang, W. Fann, J.-H. Hsu, K.-M. Chen, and Y.-C. Yu, J. Phys. Chem. A 111, 9379 (2007).
- [18]S. Felton, A. M. Edmonds, M. E. Newton, P. M. Martineau, D. Fisher, and D. J. Twitchen, Phys. Rev. B 77, 081201(R) (2008).
- [19]D. Wirtitsch, G. Wachter, S. Reisenbauer, M. Gulka, V. Ivády, F. Jelezko, Á. Gali, M. Nesladek, and M. Trupke, Phys. Rev. Research 5, 013014 (2023).
- [20]N. Aslam, G. Waldherr, P. Neumann, F. Jelezko, and J. Wrachtrup, New J. Phys. 15, 013064 (2013).
- [21]A. A. Wood, A. Lozovoi, R. M. Goldblatt, C. A. Meriles, and A. M. Martin, Phys. Rev. B 109, 134106 (2024).
- [22]S. M. Blakley, T. T. Mai, S. J. Moxim, J. T. Ryan, A. J. Biacchi, A. R. Hight Walker, and R. D. McMichael, Phys. Rev. B 110, 134109 (2024).
- [23]K. V. Ung, C. A. Ronconali, R. L. Walsworth, and S. M. Blakley, Phys. Status Solidi A 223, e202500538 (2026).
- [24]Y. Cao, Y. Liu, J. Lin, M. He, X. Yuan, B. Wu, and E. Wu, Appl. Phys. Lett. 126, 144004 (2025).
- [25]Y. Cao, J. Lin, M. He, X. Yuan, B. Wu, and E. Wu, Diamond Relat. Mater. 160, 112987 (2025).
- [26]M. T. Luu, C. Linderälv, Z. Benedek, Á. Ganyecz, G. Barcza, V. Ivády, and R. Ulbricht, Appl. Phys. Lett. 126, 234001 (2025).
- [27]Y. Jin, J. Park, M. M. McMillan, D. D. Ohm, C. Barnes, B. Pingault, C. Egerstrom, B. Huang, M. Govoni, F. J. Heremans, D. D. Awschalom, and G. Galli, Phys. Rev. Lett. 135, 036401 (2025).
- [28]S. A. Wolf, I. Meirzada, G. Haim, and N. Bar-Gill, Phys. Rev. Applied 19, 034076 (2023).



