betaFeedback

WorksPhysics and Astronomy

Stability and molecular pathways to the formation of spin defects in silicon carbide

Abstract

Spin defects in wide-bandgap semiconductors provide a promising platform to create qubits for quantum technologies. Their synthesis, however, presents considerable challenges, and the mechanisms responsible for their generation or annihilation are poorly understood. Here, we elucidate spin defect formation processes in a binary crystal for a key qubit candidate–the divacancy complex (VV) in silicon carbide (SiC). Using atomistic models, enhanced sampling simulations, and density functional theory calculations, we find that VV formation is a thermally activated process that competes with the conversion of silicon (VSiV_{Si}) to carbon monovacancies (VCV_{C}), and that VV reorientation can occur without dissociation. We also find that increasing the concentration of VSiV_{Si} relative to VCV_{C} favors the formation of divacancies. Moreover, we identify pathways to create spin defects consisting of antisite-double vacancy complexes and determine their electronic properties. The detailed view of the mechanisms that underpin the formation and dynamics of spin defects presented here may facilitate the realization of qubits in an industrially relevant material.

Corresponding authors:

*Juan J. de Pablo Phone: (773) 702-7791 E-mail: [email protected]

*Giulia Galli Phone: (773) 702-0515 E-mail: [email protected]

Main Text

Introduction

Solid-state spin defects are an emerging platform with applications in quantum information science, sensing, and metrology[1]. The negatively-charged nitrogen-vacancy centers (NV−NV^{-}) in diamond and neutral divacancy defects (VVVV) in silicon carbide (SiCSiC) represent promising examples of defect complexes for use as spin qubits[2, 3, 4]. Divacancies in SiCSiC—a widely used material in the semiconductor industry—are particularly attractive, as they offer all-optical spin initialization and readout capabilities[5, 6], nuclear spin control[7], and a near-infrared high-fidelity spin-photon interface[8], in addition to long coherence times[9].

The optical and electronic properties of spin defects in SiCSiC, including the silicon vacancy (VSiV_{\mathrm{Si}})[10, 11], NV centers12,13, and carbon antisite-vacancy complexes (CSiVCC_{\mathrm{Si}}V_{\mathrm{C}})14 have been characterized using a variety of techniques. These include density functional theory (DFT) calculations, electron paramagnetic resonance spectroscopy (EPR), deep level transient spectroscopy (DLTS), and photoluminescence spectroscopy (PL)15–21. Little is known, however, of how to control the selective formation and spatial localization of defect complexes. Addressing this challenge is critical for their integration with optical and electric devices and nanostructures18.

Divacancies and other spin defects in SiC are generally generated by ion implantation or electron irradiation, followed by thermal annealing22–26. The vacancy-to-VV conversion efficiency, however, is low—a few percent or less22,27. VV localization is difficult to control, as samples need to be thermally annealed at high temperatures (above ∼700 °C) to generate the defect mobility necessary to create divacancies[8]. Furthermore, the spatial placement of VV appears to depend on the initial distribution of several intrinsic and extrinsic defect species (e.g., vacancies, interstitials, substitutional defects, and dopants) and the Fermi-level of the system27.

The dynamics of vacancy defect complexes in semiconductors, particularly SiC, present considerable challenges for theory and computation as well. In particular, the activation energies for vacancy migration in SiC can be on the order of several electronvolts19, thereby limiting the applicability of first-principles molecular dynamics (FPMD) simulations. As a result, prior DFT calculations in SiC have focused mainly on computing defect formation energies and migration barriers for monovacancies at T=0T=0 K28–33. Recent kinetic Monte Carlo simulations considered the dynamics of vacancies in SiC34, but they did so by invoking a priori mechanisms for defect mobilization.

Here, we study how spin defects hosted in vacancy defect complexes are formed in cubic, 3C-SiC, by coupling FPMD simulations with a neural-network-based enhanced sampling technique35,36 to efficiently probe phase space, and we compute the free energies and stabilities of several defects. In some cases, our FPMD results are augmented by classical MD simulations conducted with larger system sizes and over longer timescales. Additionally, we predict previously unidentified spin defects consisting of antisites and double vacancies. We focused on the 3C polytype, given the increasing number of low-cost synthesis strategies developed for high-purity 3C-SiC, e.g., heteroepitaxial growth on silicon substrates37,38. Furthermore, 3C-SiC contains a single configuration for each vacancy defect (VSiV_{\mathrm{Si}}, VCV_{\mathrm{C}}, and VVV_{\mathrm{V}}), and it is thus simpler than other polytypes (e.g., 4H-SiC) that can host multiple configurations. We find that the formation of divacancies is initiated by the migration and association of monovacancies, which occur in the same temperature regime as the crystallographic reorientation of VVV_{\mathrm{V}}. Our results reveal that a divacancy is a thermodynamically stable state, while VSiV_{\mathrm{Si}} represents a kinetically trapped state that readily transforms into an intermediate carbon antisite-vacancy (CSiVCC_{\mathrm{Si}}V_{\mathrm{C}}) defect. Our simulations also show that divacancy formation can be maximized by choosing initial conditions corresponding to a large fraction of VSiV_{\mathrm{Si}} in the sample to mobilize monovacancies prior to the destabilization of VSiV_{\mathrm{Si}}.

Results

Divacancy dynamics at high temperatures

The simulation of defect migration in covalently bonded materials, where activation energies may be as large as several electron volts, requires the collection of statistics for ~10 to 100 ns. Therefore, capturing the migration of vacancies in SiC using straightforward FPMD is at present prohibitively demanding from a computational standpoint. Hence, before conducting FPMD simulations coupled to enhanced sampling techniques, we explored the dynamics of vacancy defects in SiC using an empirical force field39, based on a widely-used Tersoff-type bond order formalism40,41. The force field was selected for its ability to yield temperature-dependent bulk densities and a decomposition temperature. in agreement with experiments42,43 (see Supplementary Figure S1). We found that classical MD using the empirical force field and FPMD simulations yield qualitatively similar energetics for vacancy migration processes (see Supplementary Note 1 for a detailed comparison between the two methods).

Density of 3C-SiC as a function of temperature for empirical interatomic potentials

Figure S1. Empirical interatomic potential benchmark study. Density of 3C-SiC at a given temperature was computed from a 1 ns-long classical MD simulation performed in the NPT ensemble at 1 atm using 4096 atom cells. The black line corresponds to the experimental data from Ref. 2. The following empirical force fields were used: “T89” (in purple) for the Tersoff potential[3, 4], “T-ZBL” (in green) for the Tersoff potential with the Ziegler-Biersack-Littmark (ZBL) screened nuclear repulsion[5], “EA05” (in light blue) for the Erhart-Albe potential[6], “GA01” (in orange) for the Gao-Weber potential[7], “GW-ZBL” (in yellow) for the Gao-Weber potential with the ZBL screened nuclear repulsion[8], “MEAM” (in dark blue) for the modified embedded-atom method[9], and “EDIP” (in red) for the environment-dependent interatomic potential[10]. Among these force fields, EDIP was chosen for this study based on its reasonable agreement with experimental data on the temperature-dependent density up to ∼2200\sim2200 K and the decomposition temperature, e.g., Tm∼2620T_{\mathrm{m}} \sim2620 K using EDIP and 2818 K from experiments[11].

To probe the dynamics of VV and its formation from single vacancy defects (see Figure 1), which are only present in low concentrations, we carried out 10-ns long classical MD simulations of a 4096-atom supercell of 3C-SiC containing a pair of VSiV_{\mathrm{Si}} and VCV_{\mathrm{C}} (see Methods). Analyses of the MD trajectories reveal one or more of the following phenomena at temperatures between 1000 K and 1500 K: (1) Monovacancy migration (Supplementary Movie 1), (2) VV formation from the pairing of two monovacancies (Figure 2a-Figure 2d, Supplementary Movie 3), (3) VV orientational changes (Figure 2e-Figure 2h, Supplementary Movie 4), and (4) conversion of VSiV_{\mathrm{Si}} into a carbon vacancy-antisite complex, CSiVC\mathrm{C}_{\mathrm{Si}}V_{\mathrm{C}} (Figure 2i-Figure 2j, Supplementary Movie 2). Below 1000 K, we do not observe any vacancy diffusion, even after continuing our simulations for 100 ns; above 1500 K, divacancies dissociate into VSiV_{\mathrm{Si}} and VCV_{\mathrm{C}}, and CSiVC\mathrm{C}_{\mathrm{Si}}V_{\mathrm{C}} also dissociates into CSi\mathrm{C}_{\mathrm{Si}} and VCV_{\mathrm{C}} (Figure 2j-Figure 2l). These results suggest that the formation of a VV is a temperature-activated process. VV generation occurs in the same temperature regime as the single vacancy diffusion and, surprisingly, by the destabilization of silicon vacancies via the conversion of VSiV_{\mathrm{Si}} into CSiVC\mathrm{C}_{\mathrm{Si}}V_{\mathrm{C}}.

Ball and stick representation of defects in cubic SiC

Figure 1. Ball and stick representation of defects in cubic SiC (3C-SiC). A divacancy defect complex (VVVV) consists of a carbon vacancy (VCV_\mathrm{C}) paired with a silicon vacancy (VSiV_\mathrm{Si}). Carbon and silicon atoms are shown as gray and blue spheres, respectively, whereas vacancies are depicted as white solid circles.

Dynamics of vacancy formation and migration from classical molecular dynamics simulations

Figure 2. Dynamics of vacancy formation and migration from classical molecular dynamics (MD) simulations. A series of snapshots from classical MD simulations at temperatures above 1000 K show mechanisms for a-d, the divacancy formation, e-h, divacancy reorientation, and i-l, conversion from a VSiV_{\mathrm{Si}} to a VCV_{\mathrm{C}} before the VSiV_{\mathrm{Si}} encounters an existing VCV_{\mathrm{C}} (not shown). Three-dimensional vacancy volumes colored in teal blue show the location, size, and shape of void volumes at the defect site. a, A carbon atom in between the VCV_{\mathrm{C}} and VSiV_{\mathrm{Si}} sites displaces the VCV_{\mathrm{C}} and forms a VVV_{\mathrm{V}}, following the path marked by a red arrow. b-c, During intermediate steps, the mobile carbon atom interacts with neighboring atoms as the local coordination changes from five- and three-fold. d, The divacancy consists of a carbon vacancy adjacent to a silicon vacancy. e, The orientation of VVV_{\mathrm{V}} changes as a carbon atom adjacent to VSiV_{\mathrm{Si}} migrates to the VCV_{\mathrm{C}} site. f-g, C-C bonds are formed and broken during intermediate steps. h, The VSiV_{\mathrm{Si}}-VCV_{\mathrm{C}} axis in the final VVV_{\mathrm{V}} configuration is rotated compared to that in the initial configuration, within the laboratory frame. i-j, VSiV_{\mathrm{Si}} converts into CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} as a nearest-neighbor carbon atom displaces it. Red circle indicates the CSiC_{\mathrm{Si}} site. k,l, CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} dissociates into VCV_{\mathrm{C}} and CSiC_{\mathrm{Si}} as VCV_{\mathrm{C}} exchanges position with a nearest carbon atom.

Our results between ~1000 K and 1500 K are consistent with recent measurements performed on irradiated samples of 4H-SiC44. These annealing experiments reported that with increasing temperature above ~600 K, there is a strong correlation between decreasing deep level transient spectroscopy (DLTS) intensities of VSiV_{\mathrm{Si}} and CSiVC\mathrm{C}_{\mathrm{Si}}V_{\mathrm{C}} and increasing DLTS and PL intensities of VCV_{\mathrm{C}} and VV. The experiments, therefore, are also consistent with the formation of VV and the formation of VCV_{\mathrm{C}} from VSiV_{\mathrm{Si}} and CSiVC\mathrm{C}_{\mathrm{Si}}V_{\mathrm{C}}.

Mechanism of VV formation and reorientation processes

Snapshots from our MD simulations above 1000 K are shown in Figure 2. We provide a 3D representation of the free volume at defect sites (see Methods) to investigate changes to the local structure proximal to each defect. As shown in Figure 2a-Figure 2b, the size and shape of the free volume surrounding VCV_{\mathrm{C}} and VSiV_{\mathrm{Si}} are different. In particular, VCV_{\mathrm{C}}-sites have tetrahedrally-shaped voids, while VSiV_{\mathrm{Si}}-sites have void volumes consisting of two overlapping tetrahedra centered at the VSiV_{\mathrm{Si}}-site. The void volume of the VSiV_{\mathrm{Si}} site is larger than that of the VCV_{\mathrm{C}} site, indicating that silicon and carbon atoms surrounding VSiV_{\mathrm{Si}} are more likely to be mobile than the atoms surrounding VCV_{\mathrm{C}}. When an atom in between VCV_{\mathrm{C}} and VSiV_{\mathrm{Si}} fills in a vacancy site, the two monovacancies join together to form a VV.

Our MD simulations show that the thermal annealing of SiC samples may also lead to a reorientation of the VV without dissociating the original VV, i.e., it leads to a change in the direction of the VSiV_{\mathrm{Si}}-VCV_{\mathrm{C}} axis within the laboratory frame. A recent photoluminescence confocal microscopy study of NV−NV^{-} in diamond reported a crystallographic reorientation process, where the NV defect symmetry axis in a thermally annealed sample was rotated from its starting configuration (before annealing) in the original reference frame45. Our results suggest that thermal annealing at temperatures between ~1000 K and 1500 K may be a route toward aligning the orientation of VVs, which would be desirable for sensing applications46,47.

Free energy landscape for divacancy formation and interconversions

To quantify the energetics of defect interconversion processes and identify possible transition pathways, we computed the free energy or the potential of mean force (PMF) for VV formation at 1500 K and compared the PMF with free energy profiles of other defect transformation processes that occur at the same temperature. In order to do so, we turned to DFT calculations, and we computed the PMFs by combining a neural-network based enhanced sampling method with FPMD,36 and we used the Cartesian coordinates of the mobile carbon or silicon atom as order parameters. We performed simulations at a Fermi-level known to favor the formation of neutral divacancies (see Methods). As expected, the computed 2D-PMFs for defect transformations in 3C-SiC reveal multiple minima and maxima that depend on the specific location of the defects in the crystal structure (Figure 3a-Figure 3c).

Free energy landscapes of vacancy conversion processes for VV and VSi

Figure 3. Free energy landscapes of vacancy conversion processes for VV and VSiV_{\mathrm{Si}}. Potentials of mean force (PMFs) were computed using enhanced sampling simulations with FPMD. Two-dimensional order parameters, ξi=(ξi,x,ξi,y)\xi_i = (\xi_{i,x}, \xi_{i,y}), are used to describe the position of a carbon atom migrating towards a vacancy site, which is the primary mechanism for VV formation (a, d), VV reorientation (b, e), and VSiV_{\mathrm{Si}} to CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} conversion processes (c, f). a-c, 2D-PMFs show free energy surfaces of vacancy conversion processes at 1500 K, whose minimum free energy pathways are marked by white dotted lines. d-f, Free energy profiles reveal intermediate (state MiM_i) and transition states (state TiT_i) along the reaction coordinates. The gray shaded regions denote the error in the PMF determined by block averaging. All three mechanisms are thermally activated processes with relatively high energy barriers (>1.3> 1.3 eV).

In the free energy profile shown in Figure 3a and Supplementary Figure 6, we identify a transition state (T) separating the free energy basins of VV and that of the two unassociated vacancies (VSi+VCV_{\mathrm{Si}} + V_{\mathrm{C}}). We find that the free energy of the VV is ~1.5 eV lower than that of the (VSi+VCV_{\mathrm{Si}} + V_{\mathrm{C}}), thus confirming that divacancies are stable species at 1500 K, consistent with high-temperature annealing experiments of as-grown 4H-SiC samples25,48. Additionally, the free energy difference between the VV and the transition state is comparable to that of the monovacancy migration barrier (~3.0 eV for the VSiV_{\mathrm{Si}} diffusion and ~3.9 eV for VCV_{\mathrm{C}}-diffusion, see Supplementary Figure S5) as both processes involve a next-nearest neighbor atom substituting a vacancy site. The predicted VCV_{\mathrm{C}} migration barrier obtained from FPMD (~3.9 eV) agrees with the barrier measured by DTLS on annealed SiC samples (~3.7 to 4.2 eV)19.

Free energy surfaces for monovacancy migration

Figure S5. Free energy surface for monovacancy migration from FPMD. a, 2D PMF of VCV_{\mathrm{C}} migration via diffusion of a nearest carbon atom, whose cartesian coordinates are used as the collective variables. b, 2D PMF of VSiV_{\mathrm{Si}} migration via diffusion of a nearest silicon atom, whose cartesian coordinates are used as the collective variables. The MFEPs are shown as white dotted lines. Free energy barriers for migration are ∼3.9 eV\sim3.9\ \mathrm{eV} and ∼3.0 eV\sim3.0\ \mathrm{eV} for VCV_{\mathrm{C}} and VSiV_{\mathrm{Si}}, respectively.

Free energy surfaces for VV formation and VV reorientation

Supplementary Figure 6. Free energy surface for VV formation and VV reorientation from FPMD. a, 2D PMF of VV formation by VCV_{\mathrm{C}} migration via diffusion of a nearest carbon atom, whose cartesian coordinates are used as the collective variables. b, 2D PMF of VV reorientation by VSiV_{\mathrm{Si}} migration via diffusion of a nearest silicon atom, whose cartesian coordinates are used as the collective variables. The MFEPs are shown as white dotted lines. Free energy barriers for VV formation and VV reorientation are ∼3.9\sim3.9 eV and ∼3.1\sim3.1 eV, respectively, which are higher than the other pathways shown in Figure 3a,Figure 3b in the main text.

Regarding the VV reorientation process, our simulations indicate that the diffusion free energy barriers of C-atom atoms (Figure 3d-Figure 3e) are lower than that of the Si-atom (Supplementary Figure 6) by ~0.4 eV. The PMF for the former process features two energetically equivalent local minima (M1M_1 and M2M_2) along the free energy profile (Figure 3d). By comparing the free energy profiles for VV formation and reorientation processes (Figure 3d-Figure 3e), we find that the barrier for VV dissociation (VV→VSi+VC\rightarrow V_{\mathrm{Si}} + V_{\mathrm{C}}) is ~0.5 eV higher than that required to change its orientation. This result suggests that the reorientation of divacancies can occur at a temperature lower than the temperature at which divacancies dissociate.

First-principles free energy calculations reveal that the height of the barrier for VSiV_{\mathrm{Si}} conversion into CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} (~1.3 eV) is similar to that of the barrier for VVVV formation (~1.5 eV). (see Figure 3d,Figure 3f.) Moreover, the activation energy for the formation of CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} from VSiV_{\mathrm{Si}} is lower than that of the migration of VSiV_{\mathrm{Si}} (~3.0 eV barrier; see Supplementary Figure S5). Thus, CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} formation is more likely to occur than silicon migration. Interestingly, the free energy of CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} is lower than that of VSiV_{\mathrm{Si}} by ~1.3 eV at 1500 K.

Optimal temperatures for divacancy stability in as-grown and irradiated SiC

To understand the behavior of divacancies at elevated temperatures, we first analyze the effect of annealing temperatures on the population of VVVV’s in our classical MD simulations. As shown in Figure 4a, the VVVV population decreases with increasing temperature above 1700 K, which is consistent with annealing experiments that measured the population of VVVV of as-grown SiC samples using electron paramagnetic resonance spectroscopy (EPR)48. We find that the kinetics of VVVV decay is faster in the presence of a higher vacancy concentration in the sample (see Supplementary Fig. 8). Hence, it is likely that the VVVV decay observed experimentally is slower than the one found here, since the VVVV concentrations of the experimental samples are much lower than those considered in our simulations (see Methods). By fitting an Arrhenius equation to the experimental data, i.e., y∝exp⁡(−ΔE/kBT)y \propto\exp(-\Delta E/k_{\mathrm{B}}T) (see Figure 4a), we estimate the energy difference between a bound and a dissociated VVVV state to be ~2.1 eV. This value is on a similar order of magnitude as the free energy difference between VVVV and VSi+VCV_{\mathrm{Si}}+V_{\mathrm{C}} obtained from FPMD (~1.5 eV).

Effects of temperature on the divacancy formation and destabilization

Figure 4. Effects of temperature on the divacancy formation and destabilization. a, Fraction of divacancies computed from classical MD simulations up to 10 ns, starting with only VV’s. The black circles are experimental data, showing normalized EPR intensity of VV measured after annealing a non-irradiated 4H-SiC sample for 30 min48. The dotted black line is a fit to experimental data (see descriptions in the text). Both the simulations and the experiment indicate that divacancies begin to dissociate at ∼1700\sim1700 K. b, Population of defects after VV dissociation. Solid and dotted lines indicate classical MD (10-ns long) and kinetic Monte Carlo (KMC) (1015k0−110^{15} k_0^{-1}-long) simulations, respectively. Both sets of data demonstrate that VCV_\mathrm{C}’s are the dominant defect species at high temperatures. At each temperature, the number of defects created from VV dissociation process (VCV_\mathrm{C}, VSiV_\mathrm{Si}, and CSiVCC_\mathrm{Si}V_\mathrm{C}) is normalized by the initial number of VV’s. c, Photoluminescence (PL) intensity of VV versus annealing temperature based on data from recent experimental studies using irradiated or ion-implanted samples of 4H-SiC[6] [44]. The PL intensity in each data set is normalized by the maximum signal measured. The type of implantation ion and the dose are shown in the legend. The dotted line is the fitted curve using a Gaussian Process model, which predicts an optimum annealing temperature of ∼1193\sim1193 K. The temperature regime for optimal VV formation from MD simulations is between ∼1000\sim1000 K and ∼1500\sim1500 K (blue shaded region).

After collecting classical MD trajectories for 10 ns starting with stable VVVV’s, we find that multiple vacancy defects are formed as divacancies dissociate (Figure 4b), leading to the presence of VCV_{\mathrm{C}}, VSiV_{\mathrm{Si}}, and CSiVCC_{\mathrm{Si}}V_{\mathrm{C}}. The dominant species detected at high temperatures are carbon vacancies. Other defects (VSiV_{\mathrm{Si}} and CSiVCC_{\mathrm{Si}}V_{\mathrm{C}}) are present in smaller fractions (less than ~10 % after 10 ns), as they can convert to VCV_{\mathrm{C}} via the following process: VSi→CSiVCV_{\mathrm{Si}} \rightarrow C_{\mathrm{Si}}V_{\mathrm{C}} →CSi+VC\to\mathrm{C}_{\mathrm{Si}}+\mathrm{V}_{\mathrm{C}}. In order to validate our classical MD results, we calculated energy differences using DFT, and then we computed defect populations via kinetic Monte Carlo (KMC) simulations (see Supplementary Note 2). Similar to classical MD simulations, our KMC model shows that, at high temperatures, carbon vacancies are the dominant species (Figure 4b). However, the KMC model predicts that the temperature required to dissociate CSiVC\mathrm{C}_{\mathrm{Si}}\mathrm{V}_{\mathrm{C}} is higher than that found in classical MD, because the VC\mathrm{V}_{\mathrm{C}} migration barrier is higher in FPMD (∼3.9\sim3.9 eV) than in classical MD (∼2.5\sim2.5 eV). The results from MD and KMC simulations of double vacancies and the computed free energy landscape of VV formation demonstrate that temperature facilitates both the creation and dissociation of VV. We expect the optimal temperature regime to anneal defects and generate VV to be between ∼1000\sim1000 K and ∼1500\sim1500 K or 1250±2501250 \pm250 K. This finding is consistent with recent annealing experiments carried out for irradiated or ion-implanted samples of 4H-SiC[6] [44], in which PL data of VV also suggests that the concentration of VV defects is maximized at ∼1200\sim1200 K (Figure 4c). More specifically, by fitting a Gaussian Process Model to experimental data, we estimate the optimal annealing temperature in experimentally irradiated samples to be ∼1193\sim1193 K.

Compositional dependence in divacancy formation

Our simulations and free energy calculations show that the kinetics of the VV formation is controlled by the diffusion of monovacancies, while the maximum number of divacancies is limited by the stability of VSi\mathrm{V}_{\mathrm{Si}}. These results suggest that the annealing temperature and relative concentration of VSi\mathrm{V}_{\mathrm{Si}} to VC\mathrm{V}_{\mathrm{C}} affect the mono-to-divacancy conversion efficiency. Experimentally, the composition of point defects in the sample can be changed by growing either carbon-rich or silicon-rich samples49 or by controlling the Fermi-level21. Thus, it is of interest to compute the mono-to-divacancy conversion efficiency at varying ratios of VSi\mathrm{V}_{\mathrm{Si}} to VC\mathrm{V}_{\mathrm{C}} (see Methods and Supplementary Note 2).

In the KMC simulations carried out in the temperature range where the conversion CSiVC→CSi+VCC_{\mathrm{Si}}V_{\mathrm{C}} \to C_{\mathrm{Si}} + V_{\mathrm{C}} occurs, we find that higher concentrations of VSiV_{\mathrm{Si}} lead to an increase of the probability of VVVV formation, even in the absence of VCV_{\mathrm{C}} at t=0t = 0. These results are in agreement with those of classical MD simulations at 1500 K (see Figure 5a) and are intriguing, as one would expect no stable VVVV when the system has predominantly one type of vacancy (VCV_{\mathrm{C}} or VSiV_{\mathrm{Si}}). The results obtained at lower temperature are more intuitive as the KMC simulations show that the maximum number of divacancies is formed when there is an equal concentration of VSiV_{\mathrm{Si}} and VCV_{\mathrm{C}} in the sample. These findings reveal that at high temperatures, the optimal condition to form VVVV would be to start with a large initial fraction of VSiV_{\mathrm{Si}} to mobilize monovacancy migration and CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} dissociation, so as to convert some VSiV_{\mathrm{Si}} into VCV_{\mathrm{C}} and then generate additional VCV_{\mathrm{C}}, as shown schematically in Figure 5b. A low initial concentration of VSiV_{\mathrm{Si}} reduces the VVVV conversion efficiency (see Figure 5c) because there are no energetically favored pathways to form VSiV_{\mathrm{Si}} from VCV_{\mathrm{C}}.

Dependence of VV conversion efficiency on the initial composition of monovacancies

Figure 5. Dependence of VV conversion efficiency on the initial composition of monovacancies. a, (top panel) KMC simulations showing changes in the long-time limit of VV population starting with varying concentrations of VSiV_{\mathrm{Si}}, i.e., the number of VSiV_{\mathrm{Si}} over the total number of vacancies, at two different temperatures TT (black line at TKMC=1500T_{\mathrm{KMC}} = 1500 K, red line at TKMC=2000T_{\mathrm{KMC}} = 2000 K). (bottom panel) Classical MD simulations showing the time-evolution of mono- to di-vacancy conversion up to 100 ns at TMD=1500T_{\mathrm{MD}}=1500 K, starting with different fractions of VSiV_{\mathrm{Si}}, i.e., the number of VSiV_{\mathrm{Si}} over the total number of VSiV_{\mathrm{Si}} and VCV_{\mathrm{C}}. The VV formation favors higher concentrations of VSiV_{\mathrm{Si}}, agreeing with the high-temperature behavior of the KMC model. b-c, Schematics for controlling divacancy formation by tuning the VSiV_{\mathrm{Si}}-to-VCV_{\mathrm{C}} ratio during thermal annealing at high temperature based on results from panel a. b, Samples having a greater number of VSiV_{\mathrm{Si}} than VCV_{\mathrm{C}} prior to annealing (left) producing more VV after annealing (right). c, Samples having a fewer number of VSiV_{\mathrm{Si}} than VCV_{\mathrm{C}} prior to annealing (left) producing fewer VV after annealing (right).

MD simulations identify potential spin defects

The generation of VSiV_{\mathrm{Si}} and CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} species during MD simulations of VVVV dissociation, validated by DFT-based KMC simulations (Figure 4b), pointed out species that, in appropriate charged states, are interesting spin-defects[10] [14]. Prior studies have detected and characterized spin defects based on impurities and antisite defects using EPR measurements and DFT calculations at T=0T=0 K50,51. Remarkably, we also identified potential spin defects from MD simulations of divacancies at 2000 K and above through the process: VV→VCCSiVC→CSiVC+VCVV \to V_{\mathrm{C}}C_{\mathrm{Si}}V_{\mathrm{C}} \to C_{\mathrm{Si}}V_{\mathrm{C}} + V_{\mathrm{C}} (see Supplementary Movie 5). Importantly, simulations show that these defects can be stabilized to room temperature (see Supplementary Note 3). Multiple structures of this vacancy complex are possible, depending on the relative orientation between CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} and VCV_{\mathrm{C}}, as well as the physical proximity between the two defects.

The results of hybrid DFT calculations52 of the electronic structure of the neutral antisite-vacancy complexes shown in Figure 6, indicate that VCCSiVCV_{\mathrm{C}}C_{\mathrm{Si}}V_{\mathrm{C}} (Figure 6b) and one of the structures for [CSiVC+VC][C_{\mathrm{Si}}V_{\mathrm{C}} + V_{\mathrm{C}}], [CSiVC+VC]n=3[C_{\mathrm{Si}}V_{\mathrm{C}} + V_{\mathrm{C}}]_{n=3} (Figure 6c), consist of defect levels with localized electron densities. The lowest electronic configuration of these defect levels is a triplet, as the NV−NV^{-} in diamond and VV0VV^{0} in SiC. However, we note that in the antisite-vacancy complexes, the two carbon vacancies are separated by a carbon antisite and, therefore, their ground state spin densities are more spatially delocalized than those of the VVVV state.

Electronic structures of VV0 and antisite-vacancy complexes

Figure 6. Electronic structures of VV0VV^{0} and antisite-vacancy complexes identified from MD simulations of VV dissociation. Defect structures (top) and defect energy level diagrams (bottom) are shown for neutral a, VV, b, VCCSiVCV_{\mathrm{C}}C_{\mathrm{Si}}V_{\mathrm{C}}, and c, [CSiVC+VC]n=3[C_{\mathrm{Si}}V_{\mathrm{C}}+V_{\mathrm{C}}]_{n=3}. For the [CSiVC+VC][C_{\mathrm{Si}}V_{\mathrm{C}}+V_{\mathrm{C}}] defect shown here, the two carbon vacancies are separated by at least n=3n=3 atoms. For each defect, the lowest electronic configuration is a triplet state. Shaded gray areas indicate energy levels below the valence band (VB) and above the conduction band (CB). The spin-majority (spin-minority) channel is denoted by upward- (downward-) pointing arrows.

In the case of the VCCSiVCV_{\mathrm{C}}C_{\mathrm{Si}}V_{\mathrm{C}} state, we find unoccupied defect levels within the band gap close to the conduction band, similar to neutral VVVV spin defect in 3C-SiC53 (see also Figure 6a). These results suggest that [CSiVC+VC]n=3[C_{\mathrm{Si}}V_{\mathrm{C}} + V_{\mathrm{C}}]_{n=3} and, especially, VCCSiVCV_{\mathrm{C}}C_{\mathrm{Si}}V_{\mathrm{C}} are possible spin defects. Since VCCSiVCV_{\mathrm{C}}C_{\mathrm{Si}}V_{\mathrm{C}} and [CSiVC+VC][C_{\mathrm{Si}}V_{\mathrm{C}} + V_{\mathrm{C}}] are created when VVVV dissociates (T>∼1700T > \sim1700 K), it is unlikely that these spin defects are simultaneously present with VVVVs in post-annealed samples. However, if present, high concentrations of these spin defects and/or isolated CSiC_{\mathrm{Si}} species with localized electronic spins54,55 may unfavorably decrease the VVVV coherence times.56

Notably, our simulations did not incorporate any prior information on the formation mechanism of the spin defects found here. This highlights the potential of an atomistic approach to identify not only novel spin defects when combined with electronic structure calculations but also microscopic mechanisms of their formation.

Discussion

Our simulations of the relative stability of monovacancies and divacancies in 3C-SiC and their dynamics have revealed that, to generate a high concentration of VVVV in the material, a large ratio of silicon to carbon vacancy concentration is needed. Several of the defect transformation mechanisms observed in our simulations are consistent with and help explain annealing experiments, both in as-grown samples and irradiated/ion-implanted samples.

Divacancies are found to undergo crystallographic reorientation without dissociating. These findings indicate that the optimal annealing temperature for reorientation, whose control is of interest for sensing applications, is lower than that required to generate the defects.

Beyond having delineated the formation mechanisms of VVV_\mathrm{V}, VSiV_\mathrm{Si}, and CSiVCC_\mathrm{Si}V_\mathrm{C} spin defects, which arise from the dissociation of VVV_\mathrm{V} into CSiVCC_\mathrm{Si}V_\mathrm{C} and another VCV_\mathrm{C}, we have also discovered plausible new spin defects and characterized their electronic structure using hybrid density functional theory. Such defects provide new opportunities for quantum technology applications.

Methods

Classical MD simulations for divacancy dynamics

Bulk 3C-SiC was modeled as a cubic supercell containing 4096 atoms with 3D periodic boundary conditions. Energy minimization and MD simulations were carried out using LAMMPS57 and an environment-dependent interatomic potential (EDIP) force field39, chosen after benchmarking several empirical potentials for SiC (Supplementary Figure S1).

We equilibrated a 3C-SiC structure at the target temperature TT and pressure of 1 atm in the NPT ensemble for 1 ns with a 1.0 fs time step. The system was further equilibrated for an additional 1 ns in the NVT ensemble. A Nose-Hoover thermostat and barostat were used to control temperature and pressure, respectively.

After the 2-ns long equilibration phase of the bulk crystal, three types of NVT production runs were conducted with supercells containing one of the following: (1) a pair of VCV_\mathrm{C} and VSiV_\mathrm{Si} and (2) a divacancy, and (3) 40 monovacancies with varying composition of monovacancies. In the first case, 20 independent 10-ns long trajectories were sampled at each TT between 500 K and 2500 K, every 500 K. In the second case, 100 independent 10 ns-long trajectories starting with a single divacancy were sampled at each temperature between 1500 K and 2500 K, every 100 K. In the last case, a high vacancy concentration (40/4096 = 1 %) was used to increase the probability of forming VV’s from monovacancies within the simulation time of 100 ns. At each composition (ratio of VSiV_{\mathrm{Si}} to VCV_{\mathrm{C}}), 25 independent 100-ns long trajectories were collected at 1500 K. The VV conversion efficiency at time tt was calculated by dividing the number of divacancies at time tt by half of the number of monovacancies at t=0t = 0. Overall, we performed over 2200 MD simulations of 10 to 100 ns each.

Void volume analysis for vacancy defects

To identify and track the position of vacancy defects, we constructed Voronoi cells around each atom in a bulk crystal using VORO++58. The location of the vacancy defect was defined by the centroid position of an empty Voronoi cell after aligning the atoms in the bulk crystal to those in the MD snapshot. We computed the 3D void volume at the vacancy defect site by calculating the distribution of voids on a discretized 3D grid with a spacing of 0.15 Å x 0.15 Å x 0.15 Å. The excluded volume of each atom was approximated as a sphere with a radius RR (1.6 Å for carbon and 2.0 Å for silicon), where the ratio of the C and Si radii was the same as the ratio of their respective van der Waals radii.

Determination of defect energy levels using DFT calculations

Electronic structure calculations were carried out using DFT as implemented in the Quantum Espresso59 and Qbox code60,61. Defects were modeled using 4x4x4 or 512-atom supercells of 3C-SiC. The interaction between core and valence electrons was described using Optimized Norm-Conserving Vanderbilt pseudopotentials from the SG15 library62 and we used a plane-wave basis with a kinetic energy cutoff of 55 Ry; the Brillouin zone was sampled with the Γ\Gamma-point. For each defect calculations, a smearing with a width of 0.001 Ry was used in ground state electronic structure calculation, using the Marzari-Vanderbilt procedure63; geometries were optimized at the Generalized Gradient Approximation (GGA) level of DFT using the Perdew-Burke-Ernzerhof (PBE) functional64 in Quantum Espresso. The defect-level energy diagrams were obtained using the dielectric-dependent hybrid (DDH) functional with a self-consistent Hartree-Fock mixing parameter of α=0.15\alpha=0.15 for SiC52 at the ground state triplet configuration using the recursive bisection method65 as implemented in the Qbox code.

Free energy calculations by coupling FPMD simulations with enhanced sampling

The free energy landscape66–69 for defect migration and formation was computed using the combined force-frequency method or CFF-FPMD36 as implemented in the software package SSAGES70 coupled to the FPMD software Qbox. Unbiased FPMD and enhanced sampling simulations were performed in the NVT ensemble using the Bussi–Donadio–Parrinello thermostat71 with a time step of 0.967 fs at T=1500T = 1500 K. To evaluate energy and forces, DFT calculations were carried out using the PBE functional using plane waves with a 40 Ry kinetic energy cutoff and the Γ\Gamma-point to sample the Brillouin zone. We used the same pseudopotential as in the defect energy level calculations. For computational efficiency, the system was modeled with a 216-atom cell (see Supplementary Note 4 for the finite size effect). Defect dynamical properties were simulated for the ground state triplet configuration. The total charge of the supercell was neutral; however, charge rearrangements (i.e., charges near individual vacancy sites) during vacancy migration are allowed to occur in FPMD simulations. The xx- and yy-coordinates of a mobile atom displacing a vacancy site were used as the collective variables (CV), e.g., ξi,x=XCi\xi_{i,x}=X_{C_i} and ξi,y=YCi\xi_{i,y}=Y_{C_i} for the VC_{\mathrm{C}} migration. During CFF-FPMD simulations, running average of forces and frequencies were recorded on a 2D grid with a grid spacing of 0.21 Å along ξi,x\xi_{i,x} and ξi,y\xi_{i,y}. Bayesian regularized artificial neural networks (BRANN) with two hidden layers (8 and 6 nodes in the first and the second layers, respectively) were used. For each free energy calculation, up to 63 walkers/replicas were employed, and simulations were carried out up to 30 ps per walker. To compute the mean free energy path (MFEP), we fitted the 2D-PMF to a BRANNs to enable a smooth interpolation of free energies from a discretized 2D-CV grid36. The nudge elastic band method72 with a spring constant of 5 eV/Å was applied to the BRANN-fitted PMF to find the MFEP between the initial and final states.

Data Availability

Data that support the findings of this study will be available through the Qresp73 curator at https://paperstack.uchicago.edu/explorer.

Acknowledgments

We thank Chris Anderson, Nazar Delegan, and He Ma for insightful comments and discussions, Joe Heremans for a careful reading of the manuscript and helpful discussions, and François Gygi for help in setting up FPMD calculations using the Qbox code. We gratefully acknowledge the use of Bebop in the Laboratory Computing Resource Center at Argonne National Laboratory; the computational resources at the University of Chicago Research Computing Center; and the Argonne Leadership Computing Facility, a DOE Office of Science User Facility supported under Contract DE-AC02-06CH11357, provided by the Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program. This work was supported by the Midwest Integrated Center for Computational Materials (MICCoM) as part of the Computational Materials Science Program funded by the US Department of Energy, Office of Science, Basic Energy Sciences, Materials Sciences and Engineering Division, through Argonne National Laboratory, under Contract No. DE-AC02-06CH11357. A.Y. gratefully acknowledges support from the National Institute of Allergy and Infectious Diseases of the NIH under grant F32 AI150208.

Author information

Contributions: E.M.Y.L., J.J.d.P., and G.G. designed research. E.M.Y.L. performed research. E.M.Y.L. and A.Y. contributed methods, simulation code, and analytic tools. E.M.Y.L. and A.Y. analyzed data. E.M.Y.L., A.Y., J.J.d.P., and G.G. wrote the paper.

Ethics declarations

Competing interests: The authors declare no competing interests.

References

Comparisons between density functional theory (DFT) and classical empirical force field results

Finite-size and thermal effects on free energies of defect migration in classical molecular dynamics (MD)

Based on enhanced sampling simulations with two different supercell sizes (see Supplementary Table 1), we find that the free energy of migration for a single vacancy can vary from ∼0.1\sim0.1-0.20.2 eV, which is comparable to the thermal energy (kBT∼0.12k_{\mathrm{B}}T \sim0.12 K at T=1500T = 1500 K).

Estimated migration barriers at 0 K indicate that the thermal effect may be as large as ∼0.3\sim0.3 eV, nevertheless smaller than the difference in free energy barrier between the VSiV_{\mathrm{Si}} and VCV_{\mathrm{C}} migration processes (∼1.0\sim1.0 eV) (see Supplementary Table 1).

ProcessInitial StateFinal StateClassical MD free energy barrier at 1500 K for a 4096-atom supercell (eV)Classical MD potential energy barrier at 0 K for a 4096-atom supercell* (eV)Classical MD free energy barrier at 1500 K for a 216-atom supercell (eV)
VCV_{\mathrm{C}} migrationVCV_{\mathrm{C}}VC′V_{\mathrm{C}}'2.52.782.6
VSiV_{\mathrm{Si}} migrationVSiV_{\mathrm{Si}}VSi′V_{\mathrm{Si}}'3.53.673.7

Table 1. Size and thermal effects in potential of mean force (PMF) calculations with classical MD

  • Computed using the climbing-image nudge elastic band method as implemented in the LAMMPS software package.

Free energy landscapes of defect transformation mechanisms from first principles molecular dynamics (FPMD) and classical MD using enhanced sampling simulations

ProcessInitial stateFinal stateClassical MD free energy barrier, forward (eV)Classical MD free energy difference (final – initial state), (eV)FPMD free barrier, forward (eV)FPMD free energy difference (final – initial state), (eV)
VCV_{\mathrm{C}} migrationVCV_{\mathrm{C}}VC′V_{\mathrm{C}}'2.503.90
VSiV_{\mathrm{Si}} migrationVSiV_{\mathrm{Si}}VSi′V_{\mathrm{Si}}'3.503.00
Vacancy-antisite conversionVSiV_{\mathrm{Si}}CSiVCC_{\mathrm{Si}}V_{\mathrm{C}}1.3-0.61.3-1.3
dissociation via VCV_{\mathrm{C}}VVV_{\mathrm{V}}VSi+VCV_{\mathrm{Si}} + V_{\mathrm{C}}2.20.93.81.5

Table 2. Comparisons between FPMD and classical MD results

migration
VV dissociation via VSiV_{\mathrm{Si}} migrationVVVSi+VCV_{\mathrm{Si}} + V_{\mathrm{C}}--3.01.5
VV reorientation via VCV_{\mathrm{C}} migrationVVVV′1.702.50
VV reorientation via VSiV_{\mathrm{Si}} migrationVVVV′--3.10

Table 5.

We compared the energetics of vacancy defects using two different approaches: empirical force fields and DFT. In order to do so, we performed enhanced sampling simulations with classical MD and first principles MD (FPMD) based on DFT (with the PBE functional) at 1500 K, and we computed the free energy surfaces for the following processes: monovacancy migration, VV formation, and VV reorientation. The barriers and energy differences between the initial and the final states are summarized in Supplementary Table 2.

We find quantitative differences between the results of classical MD and FPMD, but overall the results are in qualitative agreement. Specifically, VV has a lower free energy than the dissociated states (i.e., those of the VSiV_{\mathrm{Si}} and VCV_{\mathrm{C}}); CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} has a lower free energy than VSiV_{\mathrm{Si}}; and the VSiV_{\mathrm{Si}}-to-CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} conversion process has a lower free energy barrier than the single vacancy (VSiV_{\mathrm{Si}} or VCV_{\mathrm{C}}) migration processes. Also, the free energy barrier for VV reorientation is lower than that of VV dissociation into VSiV_{\mathrm{Si}} and VCV_{\mathrm{C}}. Therefore, the key findings of our work—(1) VV formation is limited by VSiV_{\mathrm{Si}} stability and (2) VV reorientation occurs without dissociation—are now also confirmed by using FPMD simulations.

A notable difference between classical and FPMD results is that the empirical force field predicts VCV_{\mathrm{C}} migration barrier to be lower than that of VSiV_{\mathrm{Si}}, while the opposite is true using DFT.

Kinetic Monte Carlo simulations

A minimal model to simulate the population dynamics of vacancies based on the energetics from DFT was constructed using a kinetic Monte Carlo (KMC) algorithm. To simulate vacancy migration processes occurring at T≥1500T \ge1500 K (Figure 4b and Figure 5a in the main text), we considered the following processes (P1 through P8):

  • P1. VV→VSi+VC\mathrm{V}_{\mathrm{V}} \rightarrow\mathrm{V}_{\mathrm{Si}} + \mathrm{V}_{\mathrm{C}} via VSi\mathrm{V}_{\mathrm{Si}} migration

  • P2. VSi+VC→VV\mathrm{V}_{\mathrm{Si}} + \mathrm{V}_{\mathrm{C}} \rightarrow\mathrm{V}_{\mathrm{V}} via VSi\mathrm{V}_{\mathrm{Si}} migration

  • P3. CSiVC→VSi\mathrm{C}_{\mathrm{Si}}\mathrm{V}_{\mathrm{C}} \rightarrow\mathrm{V}_{\mathrm{Si}}

  • P4. VSi→CSiVC\mathrm{V}_{\mathrm{Si}} \rightarrow\mathrm{C}_{\mathrm{Si}}\mathrm{V}_{\mathrm{C}}

  • P5. CSiVC→CSi+VC\mathrm{C}_{\mathrm{Si}}\mathrm{V}_{\mathrm{C}} \rightarrow\mathrm{C}_{\mathrm{Si}} + \mathrm{V}_{\mathrm{C}}

  • P6. CSi+VC→CSiVC\mathrm{C}_{\mathrm{Si}} + \mathrm{V}_{\mathrm{C}} \rightarrow\mathrm{C}_{\mathrm{Si}}\mathrm{V}_{\mathrm{C}}

  • P7. VV→VSi+VC\mathrm{V}_{\mathrm{V}} \rightarrow\mathrm{V}_{\mathrm{Si}} + \mathrm{V}_{\mathrm{C}} via VC\mathrm{V}_{\mathrm{C}} migration

  • P8. VSi+VC→VV\mathrm{V}_{\mathrm{Si}} + \mathrm{V}_{\mathrm{C}} \rightarrow\mathrm{V}_{\mathrm{V}} via VC\mathrm{V}_{\mathrm{C}} migration

Processes P2, P4, P6, and P8 are the reverse of P1, P3, P5, and P7, respectively.

For each Monte Carlo trajectory, all possible transition pathways are computed from the starting state. The rate expression for each process, from states ii to jj, is given by rij=kijcir_{ij}=k_{ij}c_i where cic_i is the concentration of species ii; kij=k0,ijexp⁡[−ΔEij‡/kBT]k_{ij}=k_{0,ij}\exp[-\Delta E_{ij}^{\ddagger}/k_{\mathrm{B}}T] is the rate constant; k0,ijk_{0,ij} is the rate constant prefactor; ΔEij‡\Delta E_{ij}^{\ddagger} the energy barrier; kBk_{\mathrm{B}} the Boltzmann factor; and TT the temperature. We assumed that the prefactor is the same for all processes: k0,ij=k0k_{0,ij}=k_0. Once rates are tabulated, the nn-th process is randomly selected by finding nn for which Rn−1<u1RN≤RnR_{n-1}<u_1R_N\le R_n, where Rn=∑j=1nrjR_n=\sum_{j=1}^{n}r_j is the cumulative function; RN=∑j=18rjR_N=\sum_{j=1}^{8}r_j is the total reaction propensity; and u1∈(0,1]u_1\in(0,1] is a random number drawn from a uniform distribution. Selected process from states ii to jj is carried out by decreasing the count of each species ii by 1 and increasing the count of each species jj by 1. Time tt is updated by t=t+Δtt=t+\Delta t where Δt=−ln⁡u2/RN\Delta t=-\ln u_2/R_N and u2∈(0,1]u_2\in(0,1] is another random number drawn from a uniform distribution. This process is repeated until t=1017k0−1t=10^{17}k_0^{-1}, unless told otherwise. Each trajectory began with 2000 vacancies with a total vacancy concentration of 1e-3 % (or one vacancy for every 10,000 atoms). 50,000 independent trajectories were sampled.

In this simple KMC model, processes involving carbon atom migration (P3/P4, P5/P6, and P7/P8) have the same energy barrier as that of VC\mathrm{V}_{\mathrm{C}} migration (3.9 eV), while those involving silicon atom migration (P1/P2) have the same barrier as that of VSi\mathrm{V}_{\mathrm{Si}} migration (3.0 eV), as computed from FPMD simulations (Supplementary Table 2). The relative energy differences between the initial and final states for P1/P2, P3/P4, and P7/P8 are also parameterized based on our FPMD results (Supplementary Table 2). For P5/P6, the energy difference between the CSiVC\mathrm{C}_{\mathrm{Si}}\mathrm{V}_{\mathrm{C}} and CSi+VC\mathrm{C}_{\mathrm{Si}}+\mathrm{V}_{\mathrm{C}} states is estimated to be ~0.2 eV from the deep level transient spectroscopy study[1] of CSiVC\mathrm{C}_{\mathrm{Si}}\mathrm{V}_{\mathrm{C}}, which measures the decay of bound CSiVC\mathrm{C}_{\mathrm{Si}}\mathrm{V}_{\mathrm{C}} with increasing annealing temperature (see Supplementary Fig. 8). KMC input parameters are summarized in Supplementary Table 3.

ProcessRate expression, rir_iEnergy barrier, ΔE≠\Delta E^{\ne} (eV)
P1r1=k1CVVr_1 = k_1 C_{\mathrm{VV}}3.0
P2r2=k2CVSiCVCr_2 = k_2 C_{\mathrm{V_{Si}}}C_{\mathrm{V_C}}3.0−[E(VSi+VC)−E(VV)]=1.53.0 - [E(\mathrm{V_{Si}} + \mathrm{V_C}) - E(\mathrm{V_V})] = 1.5
P3r3=k3CCSiVCr_3 = k_3 C_{\mathrm{C_{Si}V_C}}3.9
P4r4=k4CVSir_4 = k_4 C_{\mathrm{V_{Si}}}3.9−[E(VSi)−E(CSiVC)]=2.63.9 - [E(\mathrm{V_{Si}}) - E(\mathrm{C_{Si}V_C})] = 2.6
P5r5=k5CCSiVCr_5 = k_5 C_{\mathrm{C_{Si}V_C}}3.9
P6r6=k6CCSiCVCr_6 = k_6 C_{\mathrm{C_{Si}}}C_{\mathrm{V_C}}3.9−[E(CSi+VC)−E(CSiVC)]=4.13.9 - [E(\mathrm{C_{Si}} + \mathrm{V_C}) - E(\mathrm{C_{Si}V_C})] = 4.1
P7r7=k7CVVr_7 = k_7 C_{\mathrm{VV}}3.9
P8r8=k8CVSiCVCr_8 = k_8 C_{\mathrm{V_{Si}}}C_{\mathrm{V_C}}3.9−[E(VSi+VC)−E(VV)]=2.43.9 - [E(\mathrm{V_{Si}} + \mathrm{V_C}) - E(\mathrm{V_V})] = 2.4

Table 3. Supplementary Table 3: KMC input parameters

Thermal quenching of new spin defects consisting of antistites and vacancies using classical MD simulations

At temperatures above ∼\sim1800 K, VV\mathrm{V_V} dissociated into [CSiVC[\mathrm{C_{Si}V_C} and VC]n\mathrm{V_C}]_n for n=1,2,…n=1,2,\ldots in classical MD simulations. To determine if these antisite-vacancy complexes could be stabilized by thermal quenching, we performed additional classical MD simulations starting from [CSiVC[\mathrm{C_{Si}V_C} and VC]n=1\mathrm{V_C}]_{n=1} and [CSiVC[\mathrm{C_{Si}V_C} and VC]n=3\mathrm{V_C}]_{n=3}, in which temperature was continuously decreased from 1800 K to 300 K at the NVT ensemble with a Berendsen thermostat using a temperature damping parameter of 100 fs. For a range of cooling rates (0.1, 0.5, and 1 K/ps), [CSiVC[\mathrm{C_{Si}V_C} and VC]\mathrm{V_C}] species remained stationary during the cooling process, suggesting that these defects can be stabilized by thermal quenching.

Size-dependence of DFT calculations

To ascertain the finite size effect in migration energies obtained from DFT calculations, we evaluated the total energy difference between divacancy and monovacancies that are separated by a single Si-C bond (VSi+VC\mathrm{V_{Si}} + \mathrm{V_C}). For each defect state, the DFT total energy was minimized for a ground state triplet configuration that is neutrally charged. The result is summarized in Supplementary Table 4.

Supercell sizeE(VV)−E(VSi+VC)E(\mathrm{V_V}) - E(\mathrm{V_{Si}} + \mathrm{V_C})
216-atom1.98
512-atom2.04

Table 4. Supplementary Table 4: DFT-PBE size dependence

Supplementary Movie 1: MD simulation trajectory showing the VCV_{\mathrm{C}} migration process via a carbon atom diffusing to the VCV_{\mathrm{C}} site at 1500 K

Supplementary Movie 2: MD simulation trajectory showing VSi→CSiVCV_{\mathrm{Si}} \rightarrow C_{\mathrm{Si}}V_{\mathrm{C}} by a nearest-neighbor carbon atom hopping to the VSiV_{\mathrm{Si}}-site, followed by reorientation of CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} at 1500 K

Supplementary Movie 3: MD simulation trajectory showing VSi+VC→VVV_{\mathrm{Si}} + V_{\mathrm{C}} \rightarrow VV via a carbon atom diffusing to the VCV_{\mathrm{C}} site at 1000 K.

Supplementary Movie 4: MD simulation trajectory showing (1) VV→VSi+VCVV \rightarrow V_{\mathrm{Si}} + V_{\mathrm{C}}, (2) VSi+VC→VVV_{\mathrm{Si}} + V_{\mathrm{C}} \rightarrow VV, and (3) the reorientation of VVVV at 1500 K.

Supplementary Movie 5: MD simulation trajectory showing (1) VV→VCCSiVCVV \rightarrow V_{\mathrm{C}}C_{\mathrm{Si}}V_{\mathrm{C}}, (2) VCCSiVC→CSiVC+VCV_{\mathrm{C}}C_{\mathrm{Si}}V_{\mathrm{C}} \rightarrow C_{\mathrm{Si}}V_{\mathrm{C}} + V_{\mathrm{C}}, and (3) the reorientation of CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} at 1800 K

Free energy surfaces for monovacancy migration

Supplementary Figure 2. Free energy surface for monovacancy migration from classical MD. a, Two dimensional (2D) potential of mean force (PMF) of VCV_{\mathrm{C}} migration via diffusion of a nearest carbon atom, whose cartesian coordinates are used as the collective variables. b, 2D PMF of VSiV_{\mathrm{Si}} migration via diffusion of a nearest silicon atom, whose cartesian coordinates are used as the collective variables. The mean free energy paths (MFEPs) are shown as white dotted lines. Free energy barriers for migration are ∼\sim2.5 eV and ∼\sim3.5 eV for VCV_{\mathrm{C}} and VSiV_{\mathrm{Si}}, respectively.

Free energy landscapes of vacancy conversion processes for VV and VSi from classical MD

Supplementary Figure 3. Supplementary Figure 3: Free energy landscapes of vacancy conversion processes for VV and VSi_{\mathrm{Si}} from classical MD. Two-dimensional order parameters, ξi=(ξi,x,ξi,y)\xi_i = (\xi_{i,x}, \xi_{i,y}), are used to describe the position of a carbon atom migrating towards a vacancy site, which is the primary mechanism for VV formation (a, d), VV reorientation (b, e), and VSi_{\mathrm{Si}} to CSi_{\mathrm{Si}}VC_{\mathrm{C}} conversion processes (c, f). a-c, two dimensional potentials of mean force (PMF) show free energy surfaces of vacancy conversion processes at 1500 K, whose minimum free energy pathways are marked by white dotted lines. d-f, Free energy profiles reveal intermediate (state Mi_i) and transition states (state Ti_i) along the reaction coordinates. The gray shaded regions denote the error in the PMF determined by block averaging. All three mechanisms are thermally activated processes with relatively high energy barriers (> 1.5 eV).

Supplementary Figure 4. Electronic structures of VV0^{0} and antisite-vacancy complexes obtained using a 216-atom supercell and hybrid-DFT calculations. Defect structures (top) and defect energy level diagrams (bottom) are shown for neutrally charged a, VV, b, VC_{\mathrm{C}}CSi_{\mathrm{Si}}VC_{\mathrm{C}}, and c, [CSi_{\mathrm{Si}}VC_{\mathrm{C}} + VC_{\mathrm{C}}]n=3_{n=3}. Same procedure as in 512-atom supercells (results plotted in Figure 6) was used to compute the electronic structure (see Methods in the main text). The band energy levels computed with smaller supercells also predict a spin triplet ground state.

Two-panel plot showing the effect of concentration on the kinetics of VV dissociation

Figure S7. Effect of concentration on the kinetics of VV dissociation. Kinetic Monte Carlo simulations for systems with total vacancy concentration of a, 1e−31\mathrm{e}{-3} % and b, 1e−41\mathrm{e}{-4} %.

Decay in CSiVC with increasing annealing temperature

Figure 8. Supplementary Figure 8: Decay in CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} with increasing annealing temperature. Open circles indicate deep level transient spectroscopy (DTLS) measurements of CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} species versus annealing temperature by R. Karsthof, et al.[1] Relative population of bound and unbound CSiVCC_{\mathrm{Si}}V_{\mathrm{C}} at T=1500T = 1500 K is estimated from the fit to the expression, y=11+Aexp⁡(−B/T)y = \frac{1}{1+A\exp(-B/T)}. The free energy difference is given by −kBTln⁡P(bound CSiVC)/P(unbound CSiVC)∼0.2-k_{\mathrm{B}}T\ln P(\mathrm{bound}\ C_{\mathrm{Si}}V_{\mathrm{C}})/P(\mathrm{unbound}\ C_{\mathrm{Si}}V_{\mathrm{C}}) \sim0.2 eV.

References

  1. [1]Karsthof, R., Bathen, M. E., Galeckas, A. & Vines, L. Conversion pathways of primary defects by annealing in proton-irradiated n-type 4H-SiC. Phys. Rev. B 102, 184111 (2020).DOI
  2. [2]Kern, E. L., Hamill, D. W., Deem, H. W. & Sheets, H. D. Thermal properties of β-silicon carbide from 20 to 2000°C. in Silicon Carbide–1968 S25–S32 (Elsevier, 1969). doi:10.1016/B978-0-08-006768-1.50007-3.DOI
  3. [3]Tersoff, J. Modeling solid-state chemistry: Interatomic potentials for multicomponent systems. Phys. Rev. B 39, 5566–5568 (1989).pubmed.ncbi.nlm.nih.gov/9948964
  4. [4]Tersoff, J. Erratum: Modeling solid-state chemistry: Interatomic potentials for multicomponent systems. Phys. Rev. B 41, 3248–3248 (1990).pubmed.ncbi.nlm.nih.gov/10021516
  5. [5]Devanathan, R., Diaz de la Rubia, T. & Weber, W. J. Displacement threshold energies in β-SiC. J. Nucl. Mater 253, 47–52 (1998).DOI
  6. [6]Erhart, P. & Albe, K. Analytical potential for atomistic simulations of silicon, carbon, and silicon carbide. Phys. Rev. B 71, 035211 (2005).DOI
  7. [7]Gao, F., Bylaska, E. J., Weber, W. J. & Corrales, L. R. Ab initio and empirical-potential studies of defect properties in 3C–SiC. Phys. Rev. B 64, 245208 (2001).
  8. [8]Samolyuk, G. D., Osetsky, Y. N. & Stoller, R. E. Molecular dynamics modeling of atomic displacement cascades in 3C–SiC: Comparison of interatomic potentials. J. Nucl. Mater 465, 83–88 (2015).DOI
  9. [9]Huang, H., Ghoniem, N. M., Wong, J. K. & Baskes, M. Molecular dynamics determination of defect energetics in beta -SiC using three representative empirical potentials. Modelling Simul. Mater. Sci. Eng. 3, 615–627 (1995).DOI
  10. [10]Lucas, G., Bertolus, M. & Pizzagalli, L. An environment-dependent interatomic potential for silicon carbide: calculation of bulk properties, high-pressure phases, point and extended defects, and amorphous structures. J. Phys.: Condens. Matter 22, 035802 (2010).
  11. [11]Pierson, H. O. Characteristics and Properties of Silicon Carbide and Boron Carbide. in Handbook of Refractory Carbides and Nitrides 137–155 (Elsevier, 1996). doi:10.1016/B978-081551392-6.50009-X.DOI

Paper details

vlnc:W_26bhrdcq
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 11

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents