Introduction

Nitrogen-vacancy (NV) centers in diamond1,2,3,4,5,6,7,8 are defects consisting of a nitrogen substitution and an adjacent carbon vacancy in the diamond lattice (Fig. 1a). The negatively charged NV centers (NV-) have attracted exponentially growing interest in recent years primarily as candidates for quantum sensing and communication1,3,9,10. However, despite the tremendous interest, a major gap remains in our ability to map out NV- centers at the atomic-scale and probe the effect of the atomic-scale electrostatic environment around NV- centers on their qubit properties11,12,13,14,15. Critically, manipulation of individual NV charge states at atomic length scales has continued to pose a challenge12,16,17,18. Successful manipulation of local charge states could for example be used to improve quantum sensing with simultaneous high sensitivity and spatial resolution by creating small nm-scale pockets with a high density of NV- surrounded by a region without NV-19,20.

Fig. 1: Crystal and defect structure, schematic for the device, STM topography and spectroscopy, photoluminescence at tip-sample junction.
Fig. 1: Crystal and defect structure, schematic for the device, STM topography and spectroscopy, photoluminescence at tip-sample junction.
Full size image

a Crystal structure of diamond, consisting of C atoms (red), along with an NV center, consisting of a N substitution (blue) for C and a neighboring C vacancy (green); NV centers in diamond are oriented along the [111] direction. b Schematic of measurement geometry. Monolayer graphene is transferred onto diamond (100) single crystals and the graphene/diamond structure is clamped down by a Mo plate. Bias voltage is applied through the Mo clamp with tunneling current measurement by an uncoated or Au-coated tungsten STM tip. c STM topography of graphene on diamond showing characteristic honeycomb lattice (VBias = -500 mV, ISetpoint = 70 pA). d Large energy-range dI/dV spectrum showing 5.5 eV diamond band gap from -3.2 to +2.3 eV (VBias = -5.00 V, ISetpoint = 70 pA). Inset shows smaller-scale spectrum with expected V-shape from graphene (VBias = -500 mV, ISetpoint = 70 pA). e Photoluminescence from sample in 5 \(\mu\) m diameter laser spot surrounding tip-sample junction (532 nm excitation with 75 \(\mu\) W power, 12s integration, T = 77 K). The expected photoluminescence from NV- centers, including the characteristic 637 nm zero-phonon line (shaded in red) can be seen clearly, indicating the presence of NV- centers in our STM field-of-view. The inset is a camera view of the tip-sample junction with the 532 nm laser focused on the tip apex.

Scanning tunneling microscopy (STM) is one of the ideal probes for atomic-scale imaging, mapping, and manipulation of individual defects and to map their response to their local environment. The study of NV- centers by STM has thus far not been possible due to the insulating characteristics of the diamond crystal host lattice. Here, we use a combination of characterization techniques to identify and manipulate individual NV- centers in diamond. Motivated by recent STM measurements on insulators21,22,23,24, we introduce a novel technique in which a monolayer of graphene is transferred onto the diamond surface, creating a conductive interface that enables stable tunneling measurements while preserving the electrostatic environment of the underlying atomic defects. The graphene layer is electronically transparent, allowing us to probe the low energy electronic states of NV-; the tunneling process and accessibility of subsurface defects are discussed further in Supplementary Note 1 and Supplementary Note 2. Finally, all STM experiments were carried out in a newly developed laser-STM system which allows us to use in-situ photoluminescence (PL) as an independent confirmation of the presence of NV- centers.

Results and discussion

The primary samples used in this study are type-1b diamond crystals with the (100) facet facing the surface normal direction. They were fine polished and strain-released7 before irradiation with an electron beam at 2 MeV, with a dose of 2\(\times\)1016/cm2 to create the high-density ensemble of NV- defects. As shown in the atomic force microscopy (AFM) images from Figure S1, this leads to very smooth and homogeneous surfaces over large length-scales. We then use a wet transfer technique25,26 (see Methods), to cover the diamond sample with monolayer graphene. Raman spectra taken after this procedure (Figure S2a) show the appearance of graphene G- and 2D-peaks in addition to the 1332 cm-1 zone-center phonon peak27,28 of diamond. The 2D-peak is significantly stronger than the G-peak, confirming the graphene to be monolayer. As seen in the AFM images in Figs. S2b-c, the surface remains flat, with the exception of occasional folds in the graphene layer.

For STM studies, after transferring monolayer graphene we secure the graphene/diamond onto our sample holder using a Mo clamp which serves as the bias electrode. A schematic of our measurement configuration is shown in Fig. 1b. STM images (Fig. 1c) show the honeycomb lattice of the surface graphene layer. A larger area scan of this region is shown in Fig S3. dI/dV spectroscopy taken over a large energy range displays the expected 5.5 eV band gap from diamond from the valence band edge at -3.2 eV to the conduction band edge at +2.3 eV (Fig. 1d). A smaller energy-range spectrum reveals the expected V-shape density of states feature from graphene (Fig. 1d inset). The ability to clearly see the diamond bands demonstrates that the graphene monolayer still allows access to the electronic states from the bulk substrate below.

We first obtain optical evidence of the presence of NV- defects within our STM scanning field-of-view. To do this, we measure PL spectra at our tip-sample junction using a 532 nm excitation laser with a 5 \(\mu\) m beam spot diameter. As shown in Fig. 1e, the characteristic PL signal from NV- defects is found, including the zero-phonon line at ~637 nm (inset shows a camera image of the beam focused at the tip-sample junction). Having confirmed the presence of NV- centers, we move on to atomic-scale STM characterization of individual defects. We note that all subsequent studies described in the main text were performed without laser illumination.

For clarity of presentation, we will hereafter label defects measured by STM sequentially, as Defect 1, Defect 2, etc. A prominent defect that is discerned in negative bias topographies, labeled Defect 1, is shown in Fig. 2a. dI/dV spectroscopy on Defect 1 shows a characteristic peak at approximately -300 meV (below the Fermi energy), as displayed in Fig. 2c. As shown in Fig. S4, the defect peak is robust at various tip-sample junction conditions and a large energy range dI/dV spectrum shows no additional peaks (Fig. S5). We note that the defect is not observed in the topography at positive biases (Fig. 2b). This indicates21,29,30,31,32,33,34 that the defect resides beneath the surface—rather than being an adsorbate on the graphene layer—as it does not affect the local topography and is predominantly visible at negative sample biases due to the density of states peak at -300 meV. In Fig. S6, we show examples of defects with different dI/dV peak heights likely arising from their positioning in distinct subsurface layers below the surface. The corresponding topographies also show a sharper contrast between the defect and the surrounding lattice for Defect 8, which has a stronger dI/dV peak compared to Defect 9. As discussed further in Supplementary Note 2, we estimate that the maximum depth of NV centers which we can detect is ~2.5 nm below the surface.

Fig. 2: STM characterization and energy level distribution of diamond defects.
Fig. 2: STM characterization and energy level distribution of diamond defects.
Full size image

a STM topography (VBias = -500 mV, ISetpoint = 100 pA) of a potential NV- defect, labeled Defect 1. b Topography from the same area as in a at a positive bias (VBias = +1.60 V, ISetpoint = 70 pA) showing no feature from Defect 1, indicating that the defect is likely located subsurface, in the diamond crystal below the graphene monolayer. c dI/dV spectra (VBias = -900 mV, ISetpoint = 100 pA) on and away from Defect 1, at the positions marked by the blue circle and red dot shown in a and b. d, e Energy level diagrams for NV- and P1 defects in diamond, showing the expected position of the defect states relative to the diamond valence and conduction bands (from Ref.35,36,37,38), as well as the position of the band edges relative to the Fermi energy EFermi (from Fig. 1d). f Histogram of dI/dV peak energies from measurements of 45+ defects on two different diamond samples by STM. The energies fall into two primary regions. One is centered at approximately -260 meV and the other is centered at +600 meV relative to EFermi. These two peak energies closely match previously reported energy levels of the NV- ground state35,36 (marked |g > , with excited state marked |e > ) and the P1 center37,38 (N atom substitution for a C atom) in diamond, respectively. The valence and conduction band positions relative to EFermi are also shown with sky-blue and green shades. g dI/dV spectrum on a representative P1 center (VBias = -1.40 V, ISetpoint = 70 pA). Inset shows topography of the P1 defect (VBias = +1.00 V, ISetpoint = 70 pA). h, dI/dV spectrum on a representative NV- defect (VBias = -800 mV, ISetpoint = 100 pA). Inset shows a topography of the NV- defect (VBias = -1.00 V, ISetpoint = 70 pA). All scale bars in this figure correspond to 0.5 nm.

To obtain statistical information on these and other defects seen in samples, we measured ~45 defects on two different type-1b diamond samples, with more than 5 different STM tips; the two type-1b samples are found to have nearly identical NV- densities based on photoluminescence characterization before measurements. The histogram of measured defect peak energies shows two maxima (Fig. 2f). We find that most defects show a peak at \(\approx\)-300 meV. Comparing the known NV- energy levels from the band edges35,36 and using the position of the Fermi energy relative to the conduction and valence bands from our spectroscopy data (Fig. 1d), we expect the NV- ground state to fall at approximately -260 meV ( | g> in Fig. 2d). The representative peak energy (-300 meV) of the majority of the defects is therefore consistent with the expected ground state energy level of an NV- center. The slightly extended tail of the NV- histogram is likely due to a few regions of the diamond samples with shifted band edges as shown in Fig. S7. In addition, a few defects with a peak at \(\approx\)+550  eV are also found. These defects are likely single N atoms substituting for C atoms in the diamond lattice (P1 centers). Bias-dependent topographies confirm that these defects are also subsurface as expected and typically have different topographic features compared to the NVs we image (Fig. S8). Experimental37 and theoretical38 work indicate that P1 centers should show an energy level 1.7 eV below conduction band edge which would be \(\approx\)+600  eV above the Fermi energy in our case (Fig. 2e). We note that our identification of these two types of defects as intrinsic to the diamond is supported by additional considerations. Based on prior work39,40,41, C vacancies and N adatoms in graphene are expected to show peaks at different energies. Furthermore, measurements of an electronic grade diamond sample with NV centers embedded 5 nm below the surface, outside the STM probing depth range, show an absence of these defects as presented in Fig. S9.

Having identified the most prominent defects in our data as NV- centers, we now proceed to map out the density-of-states around one such defect, labeled Defect 4. Figure 3a shows an NV- defect with its corresponding dI/dV spectrum (Fig. 3b). A slice at the Defect 4 peak energy (-300 meV) is shown in Fig. 3c. The electronic states associated with Defect 4 have a double-lobe structure with one lobe stronger than the other. Given the [111] orientation (see Fig. 1a) of NV- defects3,9,42,43,44,45 projected to the (100) surface of the diamond crystals along with the two-part structure of NV-centers, the structure seen in our dI/dV maps aligns with the expected dipole orientation of the NV- relative to our substrate orientation (see arrows in Fig. 3c and Fig. 3g). These features are consistently observed on multiple defects. Shown in Fig. 3e-h is another NV- defect, labeled Defect 5, probed with the same STM tip and on the same diamond sample as Defect 4. Defect 5 shows the same asymmetric two-lobe structure as Defect 4, but is oriented in another projected [111] direction, helping us rule out an STM tip-shape effect as an origin of the two-lobed structure.

Fig. 3: Topography, dI/dV spectroscopy, and spatial mapping of NV- defect wavefunctions in diamond.
Fig. 3: Topography, dI/dV spectroscopy, and spatial mapping of NV- defect wavefunctions in diamond.
Full size image

a–d Topography (VBias = -900 mV, ISetpoint = 100 pA), dI/dV spectrum (VBias = 900 mV, ISetpoint = 100 pA) and two dI/dV map (VBias = 900 mV, ISetpoint = 100 pA) slices at -300 mV and +300 mV of an NV- defect, labeled Defect 4. A clear asymmetric double-lobe structure can be seen in the density-of-states at the peak energy of -300 mV, while no structure is visible at +300 mV. e–h Topography (VBias = -1.00 V, ISetpoint = 70 pA), dI/dV spectrum (VBias = -1.20 V, ISetpoint = 70 pA) and two dI/dV map (VBias = -1.20 V, ISetpoint = 70 pA) slices at -360 mV and +360 mV of another NV- defect, labeled Defect 5, on the same sample and probed with the same STM tip as Defect 4. Again, an asymmetric double-lobe structure can be seen in the negative energy density-of-states at the peak energy of -360 mV but has a different directionality from Defect 4, showing that the spatial structures of the defect wavefunction are not due to tip-shape effects. The dI/dV slice at positive bias again shows no structure. We find that the defects we image are along the diamond [111] direction projected to the (100) surface ([011], [\(0\bar{1}1\)]), in agreement with the expected directionality of NV-centers. The diamond lattice directions are labeled in (a). All scale bars in this figure are 0.5 nm.

Motivated by previous work on STM tip-induced charge-state manipulation in other materials21,46,47,48, we next attempt to manipulate the charge state of NV defects via tip-induced gating. Figure 4a shows Defect 6, an example defect which is to be manipulated. We position our tip above the defect, retract the STM tip 0.5-1.0 nm out of tunneling, and apply a voltage of +5 V to the sample with respect to the grounded tip for 1 min. After going back into tunneling, STM images of the same area no longer show the bright feature associated with Defect 6, as displayed in Fig. 4b. Additionally, dI/dV spectra reveal a transition from the characteristic defect peak to a spectrum with only the background graphene density-of-states, as shown in Fig. 4c. This procedure can be consistently applied to many defects as well as with different types of STM tips, as shown in Fig. 4d-f and Figure S10. To ensure that the changes in the topography and spectra are not due to tip-changes, for each manipulated defect, we investigate a nearby reference defect within 20–50 nm of the manipulated defect, as shown in Figure S11. No changes are observed for the reference defects, confirming that our tip manipulation procedure is highly localized and tip-changes are not responsible for our observations. Our ability to control the charge state of NV defects without affecting neighboring defects as close as 20 nm away is an order of magnitude improvement on the localization of such manipulation previously reported, as discussed further in Supplementary Note 3.

Fig. 4: STM tip-induced charge-state manipulation of defects in diamond.
Fig. 4: STM tip-induced charge-state manipulation of defects in diamond.
Full size image

a, b STM topography (VBias = -1.00 V, ISetpoint = 70 pA) of an NV- defect (labeled Defect 6) (a) before and (b) after tip-induced gating manipulation of the charge state using a standard W tip by applying +5 V to the sample relative to the tip. c, dI/dV spectra (VBias = 1.00 V, ISetpoint = 70 pA) before and after manipulation of Defect 6 showing the disappearance of the strong defect peak after manipulation. d, e STM topography (VBias = -900 mV, ISetpoint = 70 pA) of another NV- defect, labeled Defect 7, (d) before and (e) after charge-state manipulation using an Au-coated W tip, again by applying +5 V to the sample relative to the tip. f dI/dV spectra (VBias = -1.20 V, ISetpoint = 70 pA) before and after manipulation of Defect 7, again showing the disappearance of the strong defect peak after manipulation. All scale bars in this figure correspond to 0.5 nm.

A plausible scenario for the physical process occurring during the tip-induced manipulation procedure is a charge-state conversion from NV- to NV0. When a large positive bias is applied to the sample relative to the tip, the electric field can push an electron out of the near-surface NV- defect under the tip, converting the charge state from NV- to NV0. The ground state of the resulting NV0 defect is expected to be more than 2 eV below the Fermi energy35,36,49,50 where the graphene density-of-states contribution is large compared to the defect density-of-states (as in Fig. S5), which is consistent with the absence of a distinct peak in dI/dV spectra after the tip-induced manipulation process. As shown in Fig. 4, we have found that this manipulation procedure works with both standard W tips as well as plasmon-resonant Au-coated W tips, opening the door to future studies combining STM and near-field optical spectroscopy. Finally, we note that while reversal of the charge state to its original configuration is in principle feasible, it is challenging to demonstrate in this context. The process requires the application of a large negative sample bias, which interacts with the graphene layer (as shown in Fig. S14f), thereby preventing clear observation of the reverse charge state transition. We note that we have also tried this procedure under laser illumination as discussed further in Supplementary Note 4; while the defect state peak doesn’t change under illumination (Fig. S13), the charge state could not be reversed within the stable bias range (Fig. S15).

In summary, we present the first STM imaging and characterization of individual NV--centers in diamond, measuring more than 40 NVs with our technique. We obtain supporting evidence that we are indeed imaging NV--centers using our in-situ photoluminescence set-up and by comparing samples known to have different NV- depths. Additionally, using STM tip-induced gating, we demonstrate the ability to consistently manipulate individual defect charge states. The manipulation was carried out with Au-coated tips which are known to have a plasmon-resonant response51,52,53,54,55. This is a statistically robust procedure which we have successfully carried out on more than 10 defects. Simultaneously correlating optical signals with individual NVs measured by STM remains a goal for follow-up work. New sample fabrication techniques are being developed currently which may allow for thin NV-enriched diamond membranes to be placed on plasmon-resonant metallic substrates. As a result, future STM tip-enhanced, near-field PL studies could potentially provide further confirmation of the individual defects and charge states studied. Given the promise that NV centers hold1,2,3,4,5,6,9,35,36,43,56, the findings described in this work mark a critical advance in the atomic-scale understanding and engineering of solid-state qubits. Future possibilities involve integration of optically-detected magnetic resonance capabilities with STM57,58,59,60 to characterize and engineer qubit performance of individual defects at the atomic-scale in technologically optimal diamond host samples.

Methods

STM measurements

Our STM measurements were performed in a Unisoku USM1200LL instrument at 77 K in ultrahigh vacuum with a Femto DLPCA-200 pre-amplifier. dI/dV measurements are taken with a Stanford Research Systems SR830 lock-in amplifier with a modulation frequency of 911.1 Hz. The sample orientation is determined visually, enabled by optical access to the sample when mounted in the Unisoku USM1200LL STM. STM tips are electrochemically etched in NaOH solutions; uncoated W tips are annealed in vacuum by electron-beam heating before measurements. To prepare Au-coated tips, we use electron-beam evaporation to coat 80 nm of Au onto etched W tips; the coated tips do not have to be annealed in vacuum due to the chemical inertness of Au. FE-SEM images in Fig. S16 show that the tips remain sharp after the Au coating procedure. All data shown in the manuscript, with the exception of the labeled plots in Figs. S12, S13, and S15, were taken without laser illumination.

Sample preparation

Type-1b diamond crystals are purchased from ElementSix and are subsequently fine polished and strain-released, yielding a roughness of <0.3 nm. The strain-release step is to remove surface residual strain after fine polishing and is performed due to the potential for strain to affect the excited-state structure of NV defects61,62,63. Samples are then irradiated with an electron beam at 2 MeV, with a dose of 2\(\times\)1016/cm2. Following irradiation, the samples are tri-acid cleaned63,64 and annealed at 1200 °C for 2 h. We estimate a defect density of one NV- detectable by STM (within ~2-2.5 nm of the surface) in every \(25\times 25{n}{m}^{2}\) area using these sample preparation parameters, within the approximate range expected based on the previously reported results65,66,67. The electronic-grade crystal is nitrogen-implanted with a dose of \(5\times {10}^{14}{ions}/c{m}^{2}\) of 15N+ ions with 3 keV energy and \(15^\circ\) incidence angle at room temperature. The crystals are implanted in a patterned grid yielding ≈16% coverage and a target depth of 5 nm. Our SRIM simulations and literature64,68,69,70,71,72 yield an estimated density of about one NV- every \(20\times 20{n}{m}^{2}\) laterally, with the implantation energy used embedding them ~5 nm below the surface. We note that the tri-acid cleaning process performed on this sample after implantation cleans off amorphous carbon64 at the surface, but likely does not etch away a sufficient portion of the diamond itself to bring the NV-dense region enough to the surface for STM imaging, based on our results presented in Fig. S9.

Monolayer graphene, grown by chemical vapor deposition on both sides of a Cu foil, is purchased from ACS Material, LLC. Polymethyl methylacrylate (PMMA, C5H8O2) is spin-coated onto one side of graphene/Cu, following which O2 plasma etching is used to remove graphene from the back side. The Cu is next etched away in ammonium persulfate for 8 h; the solution is then diluted and the PMMA/graphene film floating in the dilute ammonium persulfate is scooped onto diamond substrates. After baking out the sample at 120 °C for 1 hour, the PMMA is removed by immersion of the sample in acetone for 6 h. The sample is then annealed for 3 h at 600 °C in vacuum to remove residues. Following this, the graphene/diamond sample is loaded into the STM prep-chamber through air, after which it is again annealed at 450 °C for 3 h in ultrahigh vacuum before transferring into the STM for measurements.

Photoluminescence and Raman measurements

Room-temperature photoluminescence (PL) and Raman spectra are taken with a Nanophoton Raman11 confocal microscope and 532 nm excitation. A 100\(\times\) objective lens is used, yielding a diffraction-limited beam spot. PL and Raman spectra from in-situ measurements at the STM tip-sample junction are taken at 77 K, using a StellarNet HyperNOVA spectrometer. Excitation is at 532 nm with a Thorlabs DJ532-40 DPSS laser, and the beam diameter is approximately 5 \(\mu m\).