Sample preparation<\/p>\n
The hBN thin flakes were tape-exfoliated from a monocrystalline hBN crystal and transferred onto Si\/SiO2 substrates. Then we irradiated the hBN flakes with 2.5\u2009keV 13CO2 (99.0% 13C, Sigma-Aldrich) ions with a dose density of 1012\u2009cm\u22122 using a home-built ion implanter. The sample is then annealed at 1,000\u2009\u00b0C at 10\u22125\u2009Torr for 2\u2009h to activate the carbon-related defects. For ODMR measurements, we transferred the hBN flakes to a coplanar waveguide using the standard dry transfer method with propylene carbonate stamps. The waveguide is made of 200-nm-thick silver with a 4-nm-thick Al2O3 layer on top.<\/p>\n
Sensitivity of a single spin defect<\/p>\n
The ODMR contrast varies between defects and can reach as high as 200% (Supplementary Fig. 24<\/a>). Among the more than 100 spin defects investigated, approximately 25% exhibit a contrast higher than 10%. A single hBN spin defect in our sample has a typical sensitivity of \\(5\\,\\mu {\\rm{T}}\/\\sqrt{{\\rm{Hz}}}\\) for DC magnetic-field sensing, calculated using \\((8{\\rm{\\pi }}\/3\\sqrt{3})(1\/{\\gamma }_{{\\rm{e}}})(\\Delta \\nu \/C\\sqrt{I})\\) (ref.\u200929<\/a>), in which \u0394\u03bd is linewidth (20\u2009MHz), C is the contrast (30%) and I is the photon count rate (170\u2009kcts\u2009s\u22121). Furthermore, although most group II and III defects exhibit stable behaviours under a weak laser excitation (\u226415\u2009\u03bcW), the stability of group I defects varies greatly.<\/p>\n Estimation of nuclear spin polarization<\/p>\n We estimate the polarization of the 13C nuclear spin by evaluating the imbalance between III-2 and III-4 in the ODMR spectrum. The ODMR is taken after the SWAP gate to transfer the electron polarization to the 13C nuclear spin. By using the fitted relative populations of the hyperfine basis states, the polarization can be calculated by the equation<\/p>\n $$P=\\frac{{\\sum }_{{m}_{I}}{m}_{I}{\\rho }_{{m}_{I}}}{I{\\sum }_{{m}_{I}}{\\rho }_{{m}_{I}}}=\\frac{{\\rho }_{1\/2}-{\\rho }_{-1\/2}}{{\\rho }_{1\/2}+{\\rho }_{-1\/2}}.$$<\/p>\n \n (2)\n <\/p>\n Spin readout efficiency<\/p>\n The efficiency of a single-shot spin readout is an important factor to estimate how efficiently we can determine the electronic spin state of a spin defect, which is highly dependent on the defect properties. The readout efficiency is defined by the signal-to-noise ratio from a single readout pulse and can be expressed as38<\/a><\/p>\n $${\\eta }_{{\\rm{s}}}=1\/{\\sigma }_{{\\rm{s}}}={\\left(1+2\\frac{{\\alpha }_{0}+{\\alpha }_{1}}{{({\\alpha }_{0}-{\\alpha }_{1})}^{2}}\\right)}^{-1\/2},$$<\/p>\n \n (3)\n <\/p>\n in which \u03b10 and \u03b11 are the mean numbers of detected photons for a single measurement of the brighter state and darker state, respectively. We estimate the efficiency based on the pulsed ODMR measurements. The pulsed ODMR contrast is 17.5% when we set the readout duration at 5\u2009\u03bcs under a PMW\u2009=\u200960\u2009mW microwave drive. The contrast reaches 28% when PMW\u2009=\u20092\u2009W. For each readout laser pulse, we obtain approximately 0.9\u2009photons from the darker state under the 15-\u03bcW laser pumping. These yield the efficiency of 0.08 and 0.12 for PMW\u2009=\u200960\u2009mW and PMW\u2009=\u20092\u2009W, respectively. See details in Supplementary Information Section\u00a0IV<\/a>.<\/p>\n Gate fidelity<\/p>\n For Rabi oscillations limited by a pure dephasing process, we can write the \u03c0-gate fidelity as F\u03c0\u2009=\u20090.5(1\u2009+\u2009exp(\u22121\/Q\u03c0)), in which Q\u03c0\u2009=\u2009TRabi\/T\u03c0 is the quality factor of a \u03c0 gate39<\/a>. We extract the coherence time and \u03c0-gate time of nuclear spin Rabi by fitting the results in Fig. 3<\/a> to the function C(\u03c4)\u2009=\u2009asin(\u03c0\u03c4\/T\u03c0\u2009+\u2009b)exp(\u2212\u03c4\/TRabi)\u2009+\u2009d, in which C(\u03c4) is the signal contrast of Rabi. As a result, we obtain a F\u03c0,n\u2009=\u200999.75% \u03c0-gate fidelity, with T\u03c0,n\u2009=\u20090.60\u2009\u03bcs and TRabi,n\u2009=\u2009117\u2009\u03bcs. Similarly, we also estimate the electronic spin \u03c0-gate fidelity to be 96.2%, using the same defect and transition in the nuclear spin control experiments.<\/p>\n DFT calculations<\/p>\n We use Quantum Espresso40<\/a>, an open-source plane-wave software, to perform the DFT calculations. Both the Perdew\u2013Burke\u2013Ernzerhof functional and the Heyd\u2013Scuseria\u2013Ernzerhof hybrid functional (the factor of 0.32 for Fock exchange)41<\/a>,42<\/a> are used for the exchange-correlation interaction. We use the optimized norm-conserving Vanderbilt (ONCV) pseudopotential43<\/a>,44<\/a> for the calculations of excitation energy and the GIPAW pseudopotential45<\/a> for the calculation of hyperfine interaction parameters and ZFS. We set the kinetic energy cut-off to be 55\u2009Ry, which is adequate for converging the relevant properties. Geometry optimizations are carried out with a force threshold of 0.001\u2009Ry\u2009Bohr\u22121. We select the 6\u2009\u00d7\u20096\u2009\u00d7\u20091 or higher supercell size of hBN for the calculations of hyperfine parameters and excitation energies. For these calculations, we sample a k-point mesh of 3\u2009\u00d7\u20093\u2009\u00d7\u20091 for the calculation of excitation energies46<\/a> and the \u0393 point for the hyperfine parameters and ZFS28<\/a>,46<\/a>. We calculate the zero-phonon line by the constraint occupation DFT method47<\/a>, the hyperfine parameters using the QE-GIPAW code48<\/a>, the ZFS by using the ZFS code46<\/a> and we cross-compare results between the ZFS code and the PyZFS code49<\/a>. The key results are summarized in Supplementary Tables 2<\/a> and 3<\/a>.<\/p>\n Simulation of ODMR spectrum<\/p>\n The continuous-wave ODMR spectra shown in Fig. 5<\/a> are simulated using the MATLAB toolbox EasySpin50<\/a> based on data from the ab initio calculations. EasySpin also takes the nuclear Zeeman and quadrupole interaction into account. Therefore, the continuous-wave ODMR linewidth can be determined according to the hyperfine couplings with the most abundant nuclear-spin-active isotopes: 13C, 11B and 14N. In our simulation, we consider a 13C nuclear spin, ten nearest 11B nuclear spins and two proximate 14N nuclear spins. The other nuclei, located further away, couple more weakly to the electron, scaling with \u221d\u20091\/r3 (in which r is the distance from the central carbon site) and, thereby, have a negligible effect on the ODMR linewidth.<\/p>\n Spin-pair model<\/p>\n The S\u2009=\u20091 transitions are consistently observed alongside the S\u2009=\u20091\/2 transitions within the same emitters. To explain the coexistence of both spin transitions in ODMR, we use a weakly coupled spin-pair model32<\/a>,33<\/a> and perform numerical simulations to investigate the underlying mechanism.<\/p>\n Extended Data Fig. 1a<\/a> provides a simplified representation of the spin-pair model, consisting of two independent defects (defects A and B), separated by \u22651\u2009nm, forming a defect complex. This complex hosts two unpaired electrons that establish different internal charge states depending on their spatial occupancy. When both electrons are localized on the same defect (defect A), they form a closed-shell spin singlet GS with a metastable spin triplet state (S\u2009=\u20091), which can be accessed through laser excitation and intersystem crossing transitions (left panel in Extended Data Fig. 1b<\/a>). This state corresponds to a strongly coupled spin-pair charge state and explains the S\u2009=\u20091 transitions.<\/p>\n Alternatively, laser excitation can induce charge hopping that transfers one electron from defect A to defect B, forming a weakly coupled defect pair. In this configuration, each defect hosts a single electron (S\u2009=\u20091\/2), as illustrated in the right panel of Extended Data Fig. 1b<\/a>. This spin-dependent charge hopping yields a corresponding spin-dependent PL signal32<\/a>.<\/p>\n The actual GS, which is also the optically active state, is determined by the lowest-energy charge configuration and depends on the specific defect species and the local Fermi energy level. We considered two possible energy-level configurations (Supplementary Information Section\u00a0XII<\/a>), with Extended Data Fig. 1c<\/a> depicting the most likely scenario. In the most stable charge state, both electrons occupy the same defect site, yielding a S\u2009=\u20090\u2009GS and a metastable S\u2009=\u20091 state. Laser excitation can then generate a metastable spin-pair charge state by promoting transitions from the S\u2009=\u20091 to the S\u2009=\u20091\/2 manifold. The pronounced asymmetry in the Rabi oscillations, reflected by an increasing contrast baseline in both S\u2009=\u20091\/2 and S\u2009=\u20091 transitions, strongly supports the metastable nature of these spin manifolds.<\/p>\n In the presence of a 13C nuclear spin, the defect electron spins can couple to the nuclear spin through hyperfine interactions, which vary across different charge states. In the weakly coupled spin-pair state, the nuclear spin is primarily coupled to a single electron spin at defect A. This is consistent with our experimental observations, in which hyperfine coupling constants Azz of 130 and 300\u2009MHz were measured for group II and group III defects, respectively.<\/p>\n Given their distinct hyperfine features, group II and III defects are likely to have more well-defined and deterministic structures. By contrast, group I defects, lacking hyperfine splitting, may include a range of chemical configurations: either similar to group II and III defects (but involving 12C instead of 13C) or very different chemical structures. This structural variability may account for the broader range of stability observed in group I defects, whereas group II and III defects tend to exhibit more consistent and stable behaviour under experimental conditions.<\/p>\n Simulation of spin photodynamics<\/p>\n On the basis of the possible energy-level models, we numerically simulate the spin photodynamics using the Lindblad master equation:<\/p>\n $$\\dot{\\rho }=-\\,i[H,\\rho (t)]+\\sum _{k}{\\varGamma }_{k}\\left[{L}_{k}\\rho (t){L}_{k}^{\\dagger }-\\frac{1}{2}\\{{L}_{k}^{\\dagger }{L}_{k},\\rho (t)\\}\\right],$$<\/p>\n \n (4)\n <\/p>\n in which \u03c1(t) is the time-dependent density matrix, \u0393k represents transition rates and Lk are the associated Lindblad operators. This simulation allows us to predict PL signals in both continuous-wave ODMR and pulsed Rabi experiments (see details in\u00a0Supplementary Information<\/a>).<\/p>\n In the continuous-wave ODMR simulation, we use the full Hamiltonian, which includes the optical manifold, spin-pair states and spin-triplet states. For model 1, in which the GS is a spin singlet (S\u2009=\u20090), the Hamiltonian is written as:<\/p>\n $$H={H}_{{\\rm{pair}}}\\oplus {H}_{m,S1}\\oplus {H}_{{\\rm{eg}}}=\\left(\\begin{array}{ccc}{H}_{{\\rm{pair}}}^{8\\times 8} & & \\\\ & {H}_{m,S1}^{6\\times 6} & \\\\ & & {H}_{{\\rm{eg}}}^{4\\times 4}\\end{array}\\right)$$<\/p>\n \n (5)\n <\/p>\n