Understanding and engineering of strongly-coupled solid-state quantum spin systems is a key challenge for quantum technology. Such systems could be utilized to observe and generate collective quantum behavior, leading to highly sought-after applications ranging from quantum simulation of non-equilibrium phases1,2,3 to quantum-enhanced sensing applications beyond the classical limit4,5,6. In particular, recent work using nitrogen vacancy (NV) centers in diamond have addressed challenging problems such as the observation of critical thermalization in a three dimensional ensemble1, and a Discrete Time Crystal (DTC) state subject to a periodic drive in a disordered spin ensemble2. Realization of a strongly interacting, many-spin system was possible through the fabrication of dense NV ensembles in diamond samples with both large nitrogen density ([N]) and high [N] to [NV] conversion yield1,7. In this regime, NV-NV dipolar couplings are the dominant spin interactions1,8, with robust control possible for an ensemble of NV spins at ambient temperature.

To build on this progress and realize the aforementioned applications, it is crucial to have deterministic control of interactions between the strongly-coupled solid-state spins. To date, such control has not been possible in a scalable manner, due to variation in spin-spin separation from the stochastic process by which quantum defects are fabricated in the host solid9. Nanoscale spatial precision of defect formation has recently been demonstrated10,11, yet the generation of ensembles of solid-state spins still largely depends on the stochastic methods of ion implantation12,13 and chemical vapor deposition14, leading to an order of magnitude variation in the distribution of spin-spin interactions. The lack of control over such spin-spin interactions limits the utility of solid-state spin systems for many-body simulations and the generation of multipartite entanglement for quantum-enhanced sensing15.

In this work, we present a method to deterministically tune the dipolar coupling between strongly-coupled spin-1 qutrits through the manipulation of dressed states. We use a strongly-coupled pair of NV centers in diamond as a testbed to demonstrate the approach. The negatively charged NV center is effectively a two-electron system16, forming an S = 1 spin qutrit. The present study builds on past work where pairs of strongly interacting electronic spins in diamond were utilized to study coherent manipulation of an electronic dark spin17, as well as generation of a room temperature entangled state18. Here, we employ two Rabi driving fields to induce an oscillating NV spin population between both the \({m}_{s}\) = 0 and −1 level and the \({m}_{s}\) = 0 and +1 level, thereby generating spin qutrit dressed states. With such doubly-dressed states, an effective dipolar coupling between the two NV spins (labeled NVA and NVB) can be modulated by careful apportion of the relative Rabi driving field magnitudes. In future work, this technique may enable robust tuning of interactions within an NV ensemble spin system, allowing quantum- enhanced sensing, and engineering of the NV-NV coupling dynamics or local disorder amplitude to study transitions of non-equilibrium phases1,2.

Generating dressed states in an interacting spin-1/2 system via introducing driving fields to decouple dipolar interactions (also referred to as motional narrowing) has been extensively studied in diverse systems such as NMR19,20 and superconducting qubits21. For spin defects in diamond, a single Rabi driving field was applied to spin-1/2 nitrogen electronic spins (P1s) to suppress the overall dipolar field noise on NV spins22,23. However, more complicated dynamics can arise for strongly-coupled spin-1 qutrits24, thereby requiring further analysis and experimental investigation, as presented here.

To begin, we consider a simple two qutrit spin Hamiltonian with driving terms. With the spin flip-flop terms neglected for the dipolar interaction between the NV electronic spins1, the effective time-dependent Hamiltonian of the system with resonant driving fields is given by:

$$\begin{array}{l}{\bf{H}}\left(t\right)=D{\left({S}_{A}^{z}\right)}^{2}+\gamma {B}_{A}{S}_{A}^{z}+D{\left({S}_{B}^{z}\right)}^{2}+\gamma {B}_{B}{S}_{B}^{z}\\ \qquad\quad\,\, +\,\sum _{i\in \left\{+,-\right\}}{\varOmega }_{i}^{A}\,{\rm{cos }}\left({\omega }_{i}^{A}t\right){S}_{A,i}^{x}\\ \qquad\quad\,\, +\,\sum _{i\in \left\{+,-\right\}}{\varOmega }_{i}^{B}\,{\rm{cos }}\left({\omega }_{i}^{B}t\right){S}_{B,i}^{x}+{\nu }_{{\rm{dip}}}{S}_{A}^{z}{S}_{B}^{z}\end{array}$$

where \(D\) denotes the zero-field splitting; \({{\bf{S}}}_{A}\) and \({{\bf{S}}}_{B}\) are the NV electronic spin operators; \(\gamma {B}_{i}\) is the bias magnetic field Zeeman splitting for each NV spin with field projection \({B}_{i}\) on each NV and gyromagnetic ratio \(\gamma\); \({\nu }_{{\rm{dip}}}\) parameterizes the magnetic dipolar coupling between the two NV spins; and \({\varOmega }_{\pm }^{A,B}\), \({\omega }_{\pm }^{A,B}\) are the Rabi and carrier frequencies, respectively, of external microwave fields that resonantly drive the \(\left|{m}_{s}=0\right\rangle \leftrightarrow \left|{m}_{s}=+1\right\rangle\) and \(|{m}_{s}=0\rangle \leftrightarrow |{m}_{s}=-1\rangle\) transitions for NVA and NVB. Hyperfine and transverse Zeeman terms are neglected in the above Hamiltonian, as their effect is greatly suppressed given the large NV zero-field splitting (D = 2.87 GHz), as well as the bias magnetic field and Rabi drive frequency conditions used in the present experiment. One exception is energy level splitting of the NV spin sub-levels due to the non-secular hyperfine interaction, as observed in Ramsey measurements and discussed below. (See additional discussion in Supplementary Information).

In our experiment, we apply Rabi driving to the “control” spin NVB while detecting the resulting effect with the second “sensor” spin NVA, with the bias magnetic field aligned with NVA. By taking \({\varOmega }_{+,-}^{A}\) = 0 and \({\varOmega }_{+,-}^{B}={\varOmega }_{+,-}\), the overall Hamiltonian can be diagonalized in a double-rotating frame with the rotating wave approximation. The effective dipolar coupling term can then be analytically solved, under the condition of \({\nu }_{{\rm{dip}}}\ll {\varOmega }_{+,-}\), yielding (see Supplementary Information):

$${\nu }_{{\rm{eff}}}\approx \frac{1}{2}\frac{\left({\varOmega }_{+}^{2}-{\varOmega }_{-}^{2}\right)}{\left({\varOmega }_{+}^{2}+{\varOmega }_{-}^{2}\right)}{\nu }_{{\rm{dip}}}\in \left[-\frac{{\nu }_{{\rm{dip}}}}{2},\frac{{\nu }_{{\rm{dip}}}}{2}\right]$$

The resulting doubly-dressed states can be projected onto an effective spin-1/2 system, as follows:

$$\begin{array}{c}{\left|+\right\rangle }_{d}=\frac{1}{\sqrt{2}}\left(\frac{{\varOmega }_{+}}{\sqrt{{\varOmega }_{+}^{2}+{\varOmega }_{-}^{2}}}\left|+\right\rangle +\frac{{\varOmega }_{-}}{\sqrt{{\varOmega }_{+}^{2}+{\varOmega }_{-}^{2}}}\left|-\right\rangle +\left|0\right\rangle \right),\\ {\left|-\right\rangle }_{d}=\frac{1}{\sqrt{2}}\left(\frac{{\varOmega }_{+}}{\sqrt{{\varOmega }_{+}^{2}+{\varOmega }_{-}^{2}}}\left|+\right\rangle +\frac{{\varOmega }_{-}}{\sqrt{{\varOmega }_{+}^{2}+{\varOmega }_{-}^{2}}}\left|-\right\rangle -\left|0\right\rangle \right)\end{array}$$

In this framework, we define new eigenvectors, denoted as \(\frac{{\varOmega }_{+}}{\sqrt{{\varOmega }_{+}^{2}+{\varOmega }_{-}^{2}}}\left|+\right\rangle +\frac{{\varOmega }_{-}}{\sqrt{{\varOmega }_{+}^{2}+{\varOmega }_{-}^{2}}}\left|-\right\rangle\) and \(\left|0\right\rangle\), for the rotated Pauli operator \({S}_{R}^{z}\). The effective spin coupling can be interpreted as the projection of the lab frame interaction component, \({\nu }_{{\rm{dip}}}{S}_{A}^{z}\otimes {S}_{B}^{z}\), onto \({\nu }_{{\rm{eff}}}{S}_{A}^{z}\otimes {\left({S}_{B,R}^{z}\right)}^{2}\), where \({\nu }_{{dip}}\) is transformed to \({\nu }_{{\rm{eff}}}\) due to the projection factor \(\frac{1}{2}\frac{\left({\varOmega }_{+}^{2}-{\varOmega }_{-}^{2}\right)}{\left({\varOmega }_{+}^{2}+{\varOmega }_{-}^{2}\right)}\). Equation (2) indicates that by tuning \({\varOmega }_{+/-}\), we can deterministically vary the effective dipolar coupling strength between \(-{\nu }_{{\rm{dip}}}/2\) and \(+{\nu }_{{\rm{dip}}}/2\). Equation (2) is a generalized formula that provides different driving conditions; e.g., \({\varOmega }_{+}={\varOmega }_{-},{\nu }_{{\rm{eff}}}=0;{\varOmega }_{-}=0,{\nu }_{{\rm{eff}}}=\frac{{\nu }_{{\rm{dip}}}}{2};{\rm{and}}\) \({\varOmega }_{+}=0,{\nu }_{{\rm{eff}}}=-{\nu }_{{\rm{dip}}}/2\).


To demonstrate the tunable effective coupling via a doubly-dressed state scheme within the solid-state spins, we use two strongly-coupled NV qutrit spins as a simple model system (with \({\nu }_{{\rm{dip}}}\, > \,{\varDelta }_{{\rm{bath}}}\approx 1/{T}_{2}^{* }\), where \({\varDelta }_{{\rm{bath}}}\) is the effective coupling strength between the NV and bath spins). To realize an isolated two strongly-coupled NV spins, we use a molecular implantation technique17,25. More details on the diamond sample is discussed in the methods. We first deploy a double electron-electron resonance (DEER) measurement protocol to measure the intrinsic coupling strength between two neighboring and strongly-coupled NV spins. Due to DEER’s spin echo-based pulse scheme, the NVA sensor spin accumulates net phase only due to the repeated inversion of the NVB control spin, filtering out other possible magnetic signal sources at frequencies lower than the NVB spin modulation. We then perform Ramsey spectroscopy on NVA for different polarizations and dressed states of NVB: this measurement characterizes the effective NV-NV dipolar field under different dressed states. To directly observe the change of NV-NV interaction dynamics under different dressed states, NV-NV polarization transfer via a spin-lock pulse sequence is used (Fig. 4). Note that the roles of NVA and NVB can be interchanged in all the results presented here. Finally, we employ Monte Carlo simulations to estimate variations in the interaction dynamics of the ensemble spin system utilizing the doubly-dressed state method.

Measuring the dipolar coupling using DEER

We deploy a DEER pulse sequence to determine the strength of the bare dipolar coupling between two strongly interacting NV spins. We select neighboring NVs with different quantization axes (Fig. 1a), which allows us to distinguish the two NV spins in the ODMR frequency domain (Fig. 1b), thereby allowing individual NV spin control. The DEER technique measures the dynamic phase accumulated by the sensor spin (NVA) due to the dipolar magnetic field generated by the control spin (NVB). First, we performed a DEER measurement in the single quantum (SQ) basis of \(|0\rangle\) and \(\left|+1\right\rangle\) states. The NVA and NVB interaction is given by an Ising term with coupling strength \({\nu }_{{\rm{dip}}}\) (Eq. (1)). The accumulated phase is projected back to the \(|0\rangle\) spin state for NVA via a probability measurement \({P}_{{\rm{SQ}}}\propto \cos ({\nu }_{{\rm{dip}}}\tau /2)\). Meanwhile, under the double quantum (DQ) basis of \(\left|B\right\rangle =\frac{\left|+1\right\rangle +|-1\rangle }{\sqrt{2}}\) and \(\left|D\right\rangle =\frac{\left|+1\right\rangle -|-1\rangle }{\sqrt{2}}\) for NVA, and \(\left|-1\right\rangle ,\left|+1\right\rangle\) states for NVB, the accumulated phase projected back to \(|0\rangle\) for NVA is given by \({P}_{{\rm{DQ}}}\propto \cos (2{\nu }_{{\rm{dip}}}\tau )\). In the SQ basis of sensor spin NVA with the control spin NVB flipped to \(\left|+1\right\rangle\) after being initialized to \(|0\rangle\), we measure a DEER signal oscillation of \({\nu }_{{\rm{dip}}}/2\) = 0.125 ± 0.01 MHz (see Fig. 1c, left). In the DQ basis of sensor spin NVA with the control spin NVB flipped to \(\left|+1\right\rangle\) after being initialized to \(\left|-1\right\rangle\), we measure a DEER signal oscillation of \(2{\nu }_{{\rm{dip}}}\) = 0.495 ± 0.031 MHz (see Fig. 1c, right). From both these measurements, we extract the NVA-NVB dipolar coupling parameter \({\nu }_{{\rm{dip}}}\) = 0.250 ± 0.015 MHz.

Fig. 1: Characterization of two NV qutrit spin system.
figure 1

a Schematic of two strongly-coupled NVs inside the diamond lattice, separated by ~10 nm. External bias magnetic field \({{\bf{B}}}_{{\rm{ext}}}\) is aligned with the quantization axis of the sensor spin NVA (red); the control spin NVB (blue) has a different quantization axis. b Optically detected magnetic resonance (ODMR) measurement of NV pair system. With different Zeeman splittings due to different \({{\bf{B}}}_{{\rm{ext}}}\) field projections on each NV quantization axis, NVA and NVB are resolved in the frequency domain. c Double electron-electron resonance (DEER) measurement pulse sequences and measurement results. NVA is used as a sensor spin in both the single quantum (SQ, left) and double quantum (DQ, right) basis. NVB is used as a control spin, \(\pi\) flipped from \({m}_{s}\) = 0 to \({m}_{s}\) = +1 (left) and double \(\pi\) flipped from \({m}_{s}\) = −1 to \({m}_{s}\) = +1 (right). From the resulting modulation frequencies of the SQ and DQ DEER measurements, we extract the NVA-NVB dipolar coupling strength of \({\nu }_{{\rm{dip}}}\) = 0.250 ± 0.015 MHz.

Measuring the dipolar coupling using Ramsey

Similar to DEER, Ramsey interferometry also allows the sensor spin NVA to accumulate a dynamic phase due to the static dipolar field produced by the control spin NVB. We can further extend the Ramsey method to measure the change in effective coupling strength by transforming NVB into different dressed states. In past work, Ramsey spectroscopy has been used as a spectroscopic tool to measure the effective coupling strengths between an ensemble of NV spins and a bath of paramagnetic P1 spins26. In contrast to DEER, there is no \(\pi\) pulse applied to NVB during a Ramsey measurement; therefore, the dipolar field is constant for an initially prepared \({m}_{s}\) state of NVB during the phase accumulation of the NVA spin. For a Ramsey measurement on the NVA SQ basis of \(|0\rangle\) and \(\left|+1\right\rangle\), when NVB is prepared in \(|0\rangle\), no dynamic phase is accumulated on NVA due to the zero longitudinal dipolar coupling. However, for NVB prepared in \(\left|\pm 1\right\rangle\) with non-zero longitudinal dipolar coupling, NVA exhibits a phase modulation of \(\pm \gamma {\nu }_{{\rm{dip}}}\tau\) during a Ramsey measurement. When a similar Ramsey measurement is performed using the NVA DQ basis of \({|B}\rangle\) and \({|D}\rangle\), the effective magnetic moment of the NVA spin is doubled and there is thus a twice faster phase accumulation. A fast Fourier transformation (FFT) applied to the Ramsey signal then reveals the phase modulation frequency and hence the dipolar coupling magnitude between the two spins. Our NV spin is accompanied by a 14N (I = 1) host nuclear spin; and throughout the measurement analysis, we focus on a spin state projection measurement on the \({m}_{I}=0\) state, one of the three NV hyperfine peaks in the ODMR spectrum. In the NVA SQ basis, we find Ramsey resonance peak shifts of \(\pm {\nu }_{{\rm{dip}}}\) = 0.26 ± 0.02 MHz; see Fig. 2a. In the DQ basis, we determine peak shifts of \(\pm 2{\nu }_{{dip}}\) = 0.52 ± 0.02 MHz, relative to the Ramsey resonance peak for the non-interacting case; see Fig. 2b. Uncertainty here is given by the frequency resolution of the FFT. Note that the dipolar coupling strength extracted from SQ and DQ Ramsey spectroscopy agrees with the DEER measurements described above. This coupling strength implies a separation of the two NVs of about 6 nm.

Fig. 2: Ramsey spectroscopy for sensing dipolar coupling strength of two NV qutrit system.
figure 2

a Single quantum (SQ) Ramsey spectroscopy pulse sequence varies the evolution time τ of the sensor spin NVA with the control spin NVB initialized to \(|0\rangle\) (gray), \(\left|+1\right\rangle\) (blue) and \(\left|-1\right\rangle\) (red) states. Relative shifts in the peak of the power spectrum of the Ramsey signal as a function of the state of NVB give the dipolar coupling strength \({\nu }_{{\rm{dip}}}\) = 0.26 ± 0.02 MHz. b Repeat of the same measurement for the double quantum (DQ) basis of NVA. Due to doubling of the effective magnetic moment of spin NVA in the DQ basis, twice larger shifts are observed in the Ramsey signal power spectrum, yielding \(2{\nu }_{{\rm{dip}}}\) = 0.52 ± 0.02 MHz.

Doubly-dressed state control of effective dipolar coupling

We next introduce a driving field on the control spin NVB to generate a doubly-dressed state, as outlined above; and then perform Ramsey detection in the DQ basis of NVA to sense the resulting interaction dynamics. In a semi-classical spin picture, double driving of the NVB spin transitions with Rabi frequencies of \({\varOmega }_{+}\) and \({\varOmega }_{-}\) transfer population into each spin state \(\left|+1\right\rangle\) and \(\left|-1\right\rangle\) proportional to \({\varOmega }_{+}^{2}\) and \({\varOmega }_{-}^{2}\), respectively. In the fast-driving limit of \({\nu }_{{\rm{dip}}}\ll {\varOmega }_{\pm }\), each population is time-averaged to its half, and the overall net spin population becomes \(\frac{1}{2}({\varOmega }_{+}^{2}-{\varOmega }_{-}^{2})\). Normalizing to the total population \({\varOmega }_{+}^{2}+{\varOmega }_{-}^{2}\), we get a time-averaged effective spin number of \({m}_{s}^{{\rm{eff}}}\approx ({\varOmega }_{+}^{2}-{\varOmega }_{-}^{2})/2({\varOmega }_{+}^{2}+{\varOmega }_{-}^{2})\); hence, the effective coupling between NVA and NVB becomes \({\nu }_{{\rm{eff}}}={m}_{s}^{{\rm{eff}}}{\nu }_{{\rm{dip}}}\), which is in agreement with Eq. (2). In the simplified Hamiltonian framework (see Supplementary Information), the doubly-dressed NVB can be regarded as an effective spin-1/2 system with the dressed states denoted as \({\left|+\right\rangle }_{d},{\left|-\right\rangle }_{d}\); the coupling strength between NVA and NVB is given by Eq. (2), which includes the time-averaging effect. See Fig. 3a.

Fig. 3: Tuning effective coupling strength of two NV qutrit system via doubly-dressed state.
figure 3

a Left inset box shows the double quantum state energy level of the sensor spin NVA. Right inset box shows the ground state energy level of the control spin NVB. Microwave control fields at resonance frequencies \({\omega }_{+}\), (between \(|0\rangle\) and \(\left|+1\right\rangle\) states) and \({\omega }_{-}\) (between \(|0\rangle\) and \(\left|-1\right\rangle\) states) are driven with Rabi frequencies \({\varOmega }_{\pm }\), which results in an effective two-level doubly-dressed state \({\left|+\right\rangle }_{d},{\left|-\right\rangle }_{d}\) shown in the right inset box. Below is the pulse sequence for DQ Ramsey spectroscopy on the sensor spin NVA with NVB driven by microwave control fields at \({\varOmega }_{\pm }\). b Normalized power spectrum of the Ramsey signal for NVA. Gray dots are measurement and red solid lines are Lorentzian fits to the data. Dressed-state control parameter \(\alpha =\frac{({\varOmega }_{+}-{\varOmega }_{-})}{({\varOmega }_{+}+{\varOmega }_{-})}\). For NVB driven on only a single transition, i.e., for \(\alpha\) = ±1, a modulation peak is observed at the bare dipolar coupling strength between the two NVs (\(\pm {\nu }_{{\rm{dip}}}\)) because phase is accumulated in NVA’s DQ basis. For NVB driven on both transitions with the same Rabi frequency, i.e., for \(\alpha\) = 0, a peak appears at FFT frequency = 0. As \(\alpha\) is swept from 1 to +1 with \({\varOmega }_{\pm }\) > 2 MHz > \({\nu }_{{\rm{dip}}}\), the modulation peak continuously shifts from \({\nu }_{{\rm{eff}}}=+{\nu }_{{\rm{dip}}}\) to \({\nu }_{{\rm{eff}}}=-{\nu }_{{\rm{dip}}}\). Here, the twice larger variation in measured \({\nu }_{\rm{eff}}\), compared to Eq. (2), is due to the Ramsey measured in DQ basis. c Variation of \({\nu }_{{\rm{eff}}}\) with dressed-state control parameter \(\alpha\). Red dots are measurements, blue dots are from a numerical simulation, and black dashed line is from DQ corrected Eq. (2). Simulation is from numerically solving Eq. (1). Error bar is given as a \(\delta f\) frequency step size in the FFT plot.

For detection, we monitor the DQ Ramsey power spectrum of one of the NVA hyperfine peaks in the frequency domain as we change \({\varOmega }_{+}\) and \({\varOmega }_{-}\). To satisfy the \({\nu }_{{\rm{dip}}}\ll {\varOmega }_{\pm }\) condition, all measurements are done with 2 MHz < \({\varOmega }_{\pm }\). We define \(\alpha =\frac{({\varOmega }_{+}-{\varOmega }_{-})}{({\varOmega }_{+}+{\varOmega }_{-})}\) as a control parameter that indicates the degree of relative driving strengths. For example, \(\alpha\) = 0 for equal Rabi frequencies (\({\varOmega }_{+}={\varOmega }_{-}\)), and \(\alpha\) = +1 when driving only the single transition \(|0\rangle \leftrightarrow |+1\rangle\) (\({\varOmega }_{+}\ne\) \(0,{\varOmega }_{-}=0\)). As \(\alpha\) is swept from –1 to +1, the Ramsey power spectrum peak transitions from \({\nu }_{{\rm{eff}}}=+{\nu }_{{\rm{dip}}}\) = +0.26 MHz to \({\nu }_{{eff}}=-{\nu }_{{dip}}\) = –0.26 MHz (Fig. 3b). Note that the twice larger variation in measured \({\nu }_{{eff}}\), in contrast to Eq. (2), is due to the Ramsey spectrum being measured in the DQ basis. We confirm that the effective coupling extracted from Eq. (2) and our numerical simulation, based on Eq. (1), lie within the measurement error bound (Fig. 3c). The detuning between each NV spin’s resonance frequency ≈60 MHz; therefore, the cross-talk effect from NVB’s driving field on NVA is negligible (see further discussion below).

Polarization transfer in dressed states

To validate the applicability of the doubly-dressed state scheme for controlling spin-spin interactions deterministically, we employ a dressed state spin polarization measurement27 with a spin-lock (SL) pulse sequence by varying the \(\left|\pm 1\right\rangle\) transitions relative Rabi frequencies. Polarization transfer measurements are a useful tool for exploring interaction dynamics within strongly-coupled systems1. In a two-spin-1/2 system, the cross-polarization transfer rate is given by half the bare dipolar coupling strength when the Rabi frequencies are matched. For the doubly-dressed state in a two-spin qutrit system, the effective coupling strength can be tuned within the effective spin-1/2 framework (Fig. 4a) via the interaction term \(\frac{{\nu }_{{\rm{eff}}}}{2}{S}_{A}^{x}\otimes {S}_{B}^{x}\), which implies that one can control the transfer rate deterministically (see Supplementary Information Eqn. (30)).

Fig. 4: Spin polarization transfer measurement under singly- and doubly-dressed state Hartmann–Hahn conditions (SHH and DHH).
figure 4

a Effective spin-1/2 model for cross-polarization between singly-dressed NVA and doubly-dressed NVB. b Polarization transfer pulse sequence. NVB is first prepared in a \(|0\rangle ,\left|+1\right\rangle\) mixed state and then Rabi driven to dressed states, while NVA is spin-locked along the y-axis. c NVA spin-lock (SL) coherence signal measurement. While sweeping the NVA Rabi frequency. NVB is driven to SHH or DHH by controlling the NVB Rabi frequency \({\varOmega }_{\pm }^{B}\). Once either matching condition is satisfied, \({\varOmega }^{A}=\sqrt{{\left({\varOmega }_{+}^{B}\right)}^{2}+{\left({\varOmega }_{-}^{B}\right)}^{2}}\), polarization from NVA is lost, as shown as dips in the SL signal. SHH is when NVB is singly driven, and DHH is when NVB is doubly driven. d Polarization transfer dynamics measured via the NVA SL coherence signal over the duration of NVA spin-lock driving. NVA Rabi frequencies are fixed at the SHH or DHH matching frequencies with NVB. Error bars are 1σ photon shot-noise.

We choose NVA as the polarization delivering spin and NVB as the polarization target spin and observe polarization transfer within the \(|0\rangle \leftrightarrow \left|+1\right\rangle\) subspace. NVB is first initialized and prepared in a fully dephased state in the \(|0\rangle\), \(\left|+1\right\rangle\) basis by applying a \({\left(\frac{{\rm{\pi }}}{2}\right)}_{x}\) pulse and a wait time \({T}_{{\rm{wait}}}\gg {T}_{2}^{* }\). The purpose of using a dephased NVB as a target polarization transfer state was to showcase the potential of off-axis NVs as a tunable source of spin bath noise in ensemble spins for many-body simulations, utilizing the double driving scheme introduced in this study. Then NVB is driven with single or double transition driving fields by tuning \({\varOmega }_{+}^{B}\left(|0\rangle \leftrightarrow \left|+1\right\rangle \right)\) and \({\varOmega }_{-}^{B}(|0\rangle \leftrightarrow \left|-1\right\rangle )\), inducing a change in the effective dipolar coupling between the two NVs in a dressed state picture (Fig. 4b). NVA is initialized and spin-locked along the y-axis with driving field \({\varOmega }^{A}(|0\rangle \leftrightarrow \left|+1\right\rangle )\). Once the two NVs satisfy either the singly- or doubly-dressed state Hartmann–Hahn matching conditions (SHH or DHH)28, only the \(\frac{{\nu }_{{\rm{eff}}}}{2}{S}_{A}^{x}\otimes {S}_{B}^{x}\) term (see Supplementary Information Eqn. (30)) survives in the rotating frame Hamiltonian, inducing transfer of polarization from NVA to NVB. For a two-level spin picture, this can be understood as generating resonant energy levels between the two dressed-state spin qubits. The Rabi frequency for each spin corresponds to the energy level splitting in the rotating frame, and by tailoring the Rabi frequency matched dressed states between the two spins, polarization can be exchanged in the double-rotating frame. For an S = 1 spin system, a doubly-dressed state generates an effective two-level system with energy splitting of \(\sqrt{{\varOmega }_{+}^{2}+{\varOmega }_{-}^{2}}\). Therefore, energy-conserving polarization exchange can occur once the singly driven qutrit spin’s Rabi frequency matches the other qutrit spin’s doubly-driven Rabi frequency \(\sqrt{{\varOmega }_{+}^{2}+{\varOmega }_{-}^{2}}\). For a given NVB singly- or doubly-dressed state, the Rabi frequency of NVA and the SL duration determine the degree and rate of polarization loss(gain) of NVA (NVB).

With a fixed spin-lock time \(\tau\), Rabi frequency \({\varOmega }^{A}\) is swept to verify the Hartmann–Hahn (HH) matching condition (see Supplementary Information Eqn. (30)) for the dressed state scheme. We apply a \(|0\rangle \leftrightarrow \left|+1\right\rangle\) single transition driving field to NVB with Rabi frequency of \({\varOmega }_{+}^{B}\) = 7.56 MHz and sweep the NVA Rabi frequency between \(|0\rangle \leftrightarrow \left|+1\right\rangle\) from \({\varOmega }^{A}\) = 6 to 9 MHz. Here, the SL duration is set to be the inverse of the estimated effective dipolar coupling between the two NVs. A Lorentzian dip is observed in the NVA SL coherence measurement (Fig. 4c), indicating a loss of polarization from NVA; the dip is located at \({\varOmega }^{A}\) = 7.66 ± 0.1 MHz, which coincides with the expected SHH matching condition. Next, we apply a double driving field to NVB, with Rabi frequencies of \({\varOmega }_{+}^{B}\) = 9.59 MHz and \({\varOmega }_{-}^{B}\) = 4.13 MHz, to induce a change in polarization transfer dynamics. Again, the NVA \(|0\rangle \leftrightarrow \left|+1\right\rangle\) transition Rabi frequency is swept from \({\varOmega }^{A}\) = 9 to 12 MHz. The DHH matching condition is given by \({\varOmega }^{A}=\sqrt{{\left({\varOmega }_{+}^{B}\right)}^{2}+{\left({\varOmega }_{-}^{B}\right)}^{2}}\) and the measurement result shows a dip appearing at \({\varOmega }^{A}\) = 10.51 ± 0.1 MHz (Fig. 4c). This result matches well with the calculated DHH condition of \({\varOmega }^{A}\) = 10.44 MHz. Broadening of the DHH polarization transfer dip, compared to that of SHH, may be due to heating of the coplanar waveguide used to deliver microwave signals. Note that the Rabi drive cross-talk error between the NV transitions is on the order of \(O\left(\frac{{\varOmega }^{2}}{{\varDelta }_{A,B}^{2}}\right)\). For our experiments, \({\varDelta }_{A,B}\approx 60\) MHz is the frequency detuning between the two NV \(|0\rangle \leftrightarrow \left|+1\right\rangle\) resonance peaks; hence the cross-talk error ≈0.03 is negligible for the present study (see additional discussion in Supplementary Information).

Next, to investigate the dependence of the polarization transfer rate on different dressed states, we park the Rabi frequencies at the HH matching conditions and vary the spin-lock (SL) duration. First, without any driving field applied on NVB, the NVA SL signal is measured by sweeping the SL duration time as a reference. Under the SHH matching condition, driven by \({\varOmega }_{+}^{B}\) = 7.56 MHz, NVA SL coherence is drastically lost at a rate of 119 ± 10 kHz, which is extracted from a fit to the data (Fig. 4d). For \({\varOmega }^{{\rm{A}}}\gg {\nu }_{{dip}}\), the calculated effective dipolar coupling strength from Eq. (2) is \({\nu }_{{\rm{eff}}}\approx {\nu }_{{\rm{dip}}}/2\approx\) 130 kHz, which agrees well with our measurement. Under the DHH matching condition, driven by \({\varOmega }_{+}^{B}\) = 9.59 MHz and \({\varOmega }_{-}^{B}\) = 4.13 MHz, NVA SL coherence is lost with a reduced rate of 73 ± 10 kHz compared to the SHH condition. This result indicates a reduced effective coupling between NVA and NVB; the calculated effective coupling strength is \({\nu }_{{\rm{eff}}}\approx\)89 kHz, which is roughly consistent with our measurement. Note that the measured polarization transfer rates are somewhat lower than the calculated values; also polarization return back to NVA does not happen with full contrast. We suspect such non-ideal behavior is due to imperfect HH matching conditions resulting from coplanar waveguide heating and drift of the external bias magnetic field during the measurements.

Simulation of interaction dynamics in a spin ensemble

We simulate strongly-coupled NV spin ensemble dynamics for the doubly-dressed state, using a semi-classical model. The doubly-dressed state scheme induces a homogeneity of spin interaction strengths, which can enhance the fidelity of generating many-spin entangled states29. Here, we assume a 50 ppm NV concentration with no other defect spin species present; implying a mean distance between NV spins of 5 nm, which is in the strong coupling regime for a Carr-Purcell-Meiboom-Gill (CPMG) pulse enhanced decoherence rate \(1/{T}_{2} <\) 0.2 MHz30. In reality, this 50 ppm NV-diamond sample also carries significant concentrations of unwanted bath electronic defect spins causing short NV \({T}_{2} < 1\) \({\rm{\mu }}{\rm{s}}\). However, the doubly-dressed scheme, realized via the SL sequence, largely decouples the NV spins from the bath spins1, facilitating observation of NV-NV spin interaction dynamics. Note that NV many-body interactions can lead to inhomogeneous detuning of individual NV centers from the globally applied Rabi driving field’s central frequency, which can affect (reduce) the effective dipolar coupling strength. We find that this detuning effect is limited and dominated by the coupling strength distribution of the NV spin ensemble (see Supplementary Information). One way to understand the ensemble spin dynamics is to adopt a spin bath spectral density model, parameterized with \(\triangle\), the spin to bath coupling, and \({R}_{{dd}}\), the pairwise bath spin-spin coupling26. In the simulation, NV ensemble spins are simplified into a collection of two-spin pairs, similar to the approximation applied in a second order cluster correlation expansion (CCE) calculation31,32 (see Supplementary Information). Including higher-order spin correlations in the simulation could, in principle, better estimate the ensemble spin dynamics. However, due to the complexity of the problem, only pairwise clusters are considered here. Microscopic understanding of clusters of three or more electronic spins has been proposed using the CCE method33. However, the coherence function between spin pairs sometimes becomes 0, which causes a divergence in triple or higher spin correlations, making calculations very challenging. We note that an approximate treatment of higher-order spin correlations is as a mean-field detuning induced by long-range interactions on each spin pair33. With the doubly-dressed state scheme, this mean-field detuning can be easily adjusted, which may offer a promising avenue to study higher-order spin correlation dynamics in future work.

We use a central spin model and extract an overall effective dipolar interaction \({\varDelta }^{2}=\sum _{k}{\nu }_{{\rm{eff}},k}^{2}\) between the central NV and off-axis NV bath spins26. We only consider interactions with off-axis NV spins because for spins with the same axis, spin exchange is highly suppressed in the dressed state due to the NV’s spin-1 nature and the energy conservation in the rotating frame1. Different lattice configurations are simulated and contribute to a statistical distribution of \(\varDelta\) values. The effect of the doubly-dressed state scheme is included by adding driving field terms to the spin pair Hamiltonian. The effective dipolar interaction strength distribution follows a probability density function (PDF); and by assessing the resulting PDF peak position and full width half maximum (FWHM) values, we can estimate the overall spin interaction dynamics. With single transition driving of Rabi frequency \({\varOmega }_{+}\) = 10 MHz (\(|0\rangle \leftrightarrow \left|+1\right\rangle\)), both the PDF peak and FWHM are reduced by almost half compared to that for the not driven (ND) case (Fig. 5a). This trend continues as an additional second driving field \({\varOmega }_{-}\) (\(|0\rangle \leftrightarrow \left|-1\right\rangle\)) is introduced, pushing \(\varDelta\) from the strongly-coupled to weakly coupled regime. With \({\varOmega }_{+}\) = 10 MHz and \({\varOmega }_{-}\) = 8 MHz, \(\varDelta\) is peaked at 48 ± 10 kHz, compared to the expected effective coupling of 43 kHz deduced from the Eq. (2) using a bare dipolar coupling of 390 kHz extracted from the ND case (Fig. 5b). Next, we select pairs of off-axis NVs with the strongest coupling over the ensemble of spins and simulate the statistical distribution of pairwise dipole interaction strengths, \(\max ({\nu }_{{ij}})={R}_{{dd}}\). Among the four different NV crystalline axis classes, one class is fixed with no driving and the three other off-axis classes are driven. With single transition driving of \({\varOmega }_{+}\) = 10 MHz, both the PDF peak and FWHM are reduced almost in half (Fig. 5c); and the trend continues as a second driving field \({\varOmega }_{-}\) is applied. For example, the FWHM for doubly-dressed states is 80 ± 10 kHz with \({\varOmega }_{+}\) = 10 MHz, \({\varOmega }_{-}\) = 8 MHz; whereas the effective dipolar coupling value from Eq. (2) is \({\nu }_{{\rm{eff}}}\approx\) 107 kHz for an ND FWHM of 973 kHz, implying more uniformity in collective coupling strengths (Fig. 5d). For both \(\varDelta\) and \({R}_{{dd}}\), we also confirm the homogenization of coupling strengths through the convergence of FWHM/Peak values for doubly-dressed states at higher double transition Rabi frequencies (Fig. 5a, c). From the simulation results, the doubly-dressed state scheme is expected to both suppress interactions between the central NV spin and off-axis NV bath spins, and reduce variation of spin-spin coupling strengths, making the ensemble spin system more uniform for coherent manipulation. Such a homogenized interaction distribution, combined with the extended long spin coherence time due to the SL measurement, may improve the fidelity of non-classical state generation and extend spin entanglement lifetimes (see Supplementary Information).

Fig. 5: Semi-classical simulation of effective dipolar couplings for an NV spin ensemble in the doubly-dressed state.
figure 5

a Using a central spin model (inset), the effective coupling strength \(\varDelta\) between the central NV and the off-axis NV spin bath is calculated for multiple lattice configurations by varying the double transition driving field Rabi frequencies. The resulting \(\varDelta\) probability peak and full width at half maximum (FWHM) are then extracted and displayed (upper panel). Lower panel shows a \(\varDelta\) trend towards greater homogenization at higher double transition Rabi frequencies, as illustrated in the ratio of the FWHM/Peak for the doubly-dressed states. b Probability distribution function (PDF) for \(\varDelta\) when off-axis bath NV spins are not driven (ND); singly driven \({\varOmega }_{+}\) = 10 MHz, \({\varOmega }_{-}\) = 0; and doubly driven with \({\varOmega }_{+}\) = 10 MHz, \({\varOmega }_{-}\) = 8 MHz. c Upper panel: Extracted PDF properties for pairwise flip-flop rates \({R}_{{dd}}\) by varying \({\varOmega }_{\pm }\). As shown in the inset figure, among 4 different NV crystalline axis classes, one class (red) is fixed with no driving and 3 other off-axis classes (black) are driven. Lower panel shows homogenization trend of \({R}_{{dd}}\)’s FWHM/Peak (green) for the doubly-dressed states. d PDF for \({R}_{{dd}}\) when pairwise off-axis NV spins are not driven (ND); singly driven with \({\varOmega }_{+}\) = 10 MHz, \({\varOmega }_{-}\) = 0; and doubly driven with \({\varOmega }_{+}\) = 10 MHz, \({\varOmega }_{-}\) = 8 MHz.


To summarize, we demonstrate experimentally the use of dressed-state techniques to control the effective dipolar interaction in a strongly-coupled, solid-state electronic spin-1 system. Using a strongly-coupled pair of nitrogen vacancy (NV) centers in diamond as a model system, with \({\nu }_{{\rm{dip}}}\) being the bare NV-NV dipolar coupling strength, we induce Rabi driving between different ground spin state sub-levels (\(|0\rangle \leftrightarrow \left|\pm 1\right\rangle\)) and employ a doubly-dressed state to tune the effective coupling strength between \(-{\nu }_{{\rm{dip}}}/2 < {\nu }_{{\rm{eff}}} < +{\nu }_{{\rm{dip}}}/2\), which is spectroscopically observed via Ramsey measurements. Other pulse schemes5,34 to manipulate or suppress effective couplings are comparatively more complicated, with the duration and fidelity of the engineered Hamiltonian typically limited by pulse errors. In contrast, the doubly-dressed state scheme provides a robust method to tune the effective coupling dynamics in a qutrit system once the driving strength is larger than the bare dipolar coupling strength \({\nu }_{{\rm{dip}}}\). Furthermore, since the effective coupling strength depends only on the ratio of driving field amplitudes, systematic effects due to inhomogeneous Rabi driving amplitudes over the spin ensemble are highly suppressed. In principle, the doubly-dressed state scheme should be applicable to a larger number of spin clusters, particularly when each NV spin can be selectively controlled. We envision tuning the higher-order interaction dynamics by utilizing different dressed states for different NV spins. This method could be used to control the order parameter in a disordered spin system to study the transition of non-equilibrium phases1. Furthermore, reducing the local distribution of spin couplings could be used to increase fidelity in the generation of collective non-classical states. For example, creating an emergent Greenberger–Horne–Zeilinger (GHZ) state via quantum domino dynamics35 in an Ising spin chain largely depends on the interaction uniformity29,36. Also, better fidelity generation of a many-spin Schrodinger cat state could be a valuable resource for enhanced quantum sensing or quantum information applications.


Experimental setup

Measurements are conducted using a home-built NV-diamond confocal microscope setup. An acousto-optic modulator (Isomet Corporation) allows time-gating of a 400 mW, 532 nm diode-pumped solid-state laser (Changchun New Industries). The laser beam is coupled to a single-mode fiber, then delivered to an oil-immersion objective (100x, 1.3 NA, Nikon CFI Plan Fluor), and focused onto a diamond sample. The diamond sample is fixed on a three-axis motorized stage (Micos GmbH) for precise position control. NV red fluorescence (FL) is collected back through the same objective, then passes through a dichroic filter (Semrock LP02–633RS-25). A pinhole (diameter 75 µm) is used with a f = 150 mm telescope to spatially filter the FL signal, which is detected with a silicon avalanche photodetector (Perkin Elmer SPCM-ARQH-12). A signal generator (SG, Agilent E4428C) provides the carrier microwave signal. A 1 G s−1 rate arbitrary waveform generator (AWG, Tektronix AWG 5014 C) phase and amplitude modulates the carrier signal via an IQ mixer (Marki IQ 1545 LMP). Two resulting output microwave signals are amplified (Mini-circuits ZHL-16W-43-S+), combined, and sent through a gold coplanar waveguide fabricated on a glass cover-slip by photo-lithography and mounted directly on the diamond sample in order to manipulate the NV spins.

Diamond sample

Creation of a strongly-coupled NV pair is done by molecular ion implantation (Innovion corp), with a 6 keV energy +28N molecular beam and an implantation dosage of 1×109 cm−2 applied to a diamond substrate. The diamond is CVD grown, 12C isotopically purified to 99.99%, and has dimensions of 2 mm × 2 mm × 0.5 mm. After ion implantation, the diamond is annealed at 800 °C for 8 h and at 1000 °C for 10 h to enhance conversion of N to NV and optimize NV optical and spin properties. Statistical measurement of NV FL intensity reveals that 5% of the NVs consist of proximal (few nm separation) NV pairs, limited by the conversion of N to NV. 6 keV ion implantation creates NV pairs with an average separation of 6 nm1; this corresponds to 0.2 MHz magnetic dipolar coupling strength between the two NV spins. Using the DEER measurement technique described in the main text, the NV pair coupling strength in the diamond sample ranges from about 0.050 MHz to 0.8 MHz, for several strongly-coupled NV pairs studied. The specific NV pair used in the reported measurements has spin lifetimes given in Table 1.

Table 1 Measured spin lifetimes for the two strongly-coupled off-axis NVs used in the reported experiments.


We applied a bias magnetic field of approximately 45 G, aligned with one of the NV spins, to ensure that the detuning between the two NV spins (\(\varDelta\, >\, 60\) MHz) was sufficiently large to suppress the cross-talk error effect. The driven Rabi frequency was carefully selected to satisfy the conditions \(\varDelta \gg {\varOmega }_{\pm }\gg {\nu }_{{\rm{dip}}},h\), where h denotes the hyperfine constant, so that the interaction dynamics could follow the simple picture provided in Eq. (1). Calibration of the Rabi frequencies for each NV spin was achieved by initializing one of the NV spins and fitting the Rabi nutation measurement to a simple Hamiltonian model with a 14N hyperfine interaction. To mitigate the temperature, rise issue during the continuous Rabi driving, we set the duty cycle of the microwave signal to 10%.