betaFeedback

WorksPhysics and Astronomy

QED theory of electron beam-induced electronic excitation and its effect on sputtering cross sections in 2D crystals

Abstract

Many computational models have been developed to predict the rates of atomic displacements in two-dimensional (2D) materials under electron beam irradiation. However, these models often drastically underestimate the displacement rates in 2D insulators, in which beam-induced electronic excitations can reduce the binding energies of the irradiated atoms. This bond softening leads to a qualitative disagreement between theory and experiment, in that substantial sputtering is experimentally observed at beam energies deemed far to small to drive atomic dislocation by many current models. To address these theoretical shortcomings, this paper develops a first-principles method to calculate the probability of beam-induced electronic excitations by coupling quantum electrodynamics (QED) scattering amplitudes to density functional theory (DFT) single-particle orbitals. The presented theory then explicitly considers the effect of these electronic excitations on the sputtering cross section. Applying this method to 2D hexagonal BN and MoS2_2 significantly increases their calculated sputtering cross sections and correctly yields appreciable sputtering rates at beam energies previously predicted to leave the crystals intact. The proposed QED-DFT approach can be easily extended to describe a rich variety of beam-driven phenomena in any crystalline material.

Introduction

The holy grail of materials engineering is atomic scale control of the material structure. Towards this aim, electron irradiation by transmission electron microscopy (TEM) can be an effective means of structural manipulation with spatial control [1, 2, 3, 4]. Structural changes under electron irradiation can arise via atomic displacement in which an incident electron nudges a material atom from its initial site. We call this atom the primary knock-on atom (PKA). Two-dimensional (2D) crystals provide an excellent platform to measure the rates of these PKA displacements. When electron irradiation is normal to a 2D crystal’s surface, a displacement likely propels the PKA away from the crystal. As such, these displaced atoms are often ejected from the crystal in a process called sputtering. Sputtering events leave behind vacancies, which can then be counted using TEM. Counting the number of vacancies for a given dosage and beam energy allows one to experimentally determine the sputtering cross section of that crystal.

Sputtering occurs when the energy transferred to the PKA is greater than the PKA’s displacement threshold EdE_d. This means that a displacement is possible only if the kinetic energy of the beam electron exceeds some critical energy ϵc\epsilon_c. Many computational models have been proposed to predict both EdE_d and ϵc\epsilon_c to calculate electron beam-induced sputtering rates in 2D crystals [4, 5, 6, 7]. However, the vast majority of current methods focus solely on interactions between the beam electrons and material nuclei, neglecting any coupling with the material’s electrons. Thus, while present-day models give reasonable predictions for conductors [5], where electronic relaxation is rapid, they often vastly underestimate the atomic displacement rates in insulators. For example, the critical energy for sputtering boron or nitrogen from hexagonal boron nitride (hBN) is predicted to be 80 keV [8]. However, sputtering has been observed in hBN under 30 keV irradiation [9]. Furthermore, selenium sputters from WSe2_2 and MoSe2_2 under irradiation energies of 60 and 80 keV, respectively. These energies are almost 150 keV below their predicted critical energies [10, 11]. Lastly, while the calculated critical energy for sulfur sputtering in MoS2_2 is about 90 keV, sulfur has been shown to sputter under 20 keV beams [12]. Discrepancies like these suggest that the displacement thresholds in insulating crystals are much smaller than what is predicted by ground-state theory. Lehnert et al. have proposed that the consideration of inelastic scattering, i.e., beam-induced electronic excitation, can lead to such a reduction in the displacement threshold [11]. This would increase the sputtering cross section for all beam energies and enable sputtering for energies well-below the ground state ϵc\epsilon_c.

To account for these effects, we combine quantum electrodynamics (QED) and density functional theory (DFT) to derive the probability of beam-induced electronic excitation in 2D insulating crystals. The basic idea is as follows: DFT can provide effective single-particle states that can be decomposed into a plane-wave basis [13–15], while QED is well-equipped to describe how each plane-wave evolves in time through interactions with an electromagnetic field [16–18]. Thus, a plane-wave decomposition of the Kohn-Sham orbitals can allow for a component-by-component treatment of the interactions between the beam and material electrons. This generalized QED-DFT approach enables, for the first time, a first-principles description of any beam-matter interaction process. The only limitations of this method are the order to which the time-evolution operator is expanded and the sophistication of the theory used to determine the material’s electronic structure. Additionally, while DFT is used here, our method is compatible with any firstprinciples formalism that can produce single-particle eigenstates and eigenvalues for a given material. This paper is divided into three sections. First, we describe the key physical processes that dictate the rate of beam-induced sputtering in a 2D insulating crystal. Second, we derive the probability of beam-induced electronic excitation in these materials as a function of beam energy. Finally, we show how this excitation probability can be used to predict sputtering cross sections in hBN and MoS2_2 that quantitatively agree with experiment.

Three interactions and three rates

The majority of present-day beam-damage models focus solely on the interaction between the beam electron and target nucleus. These models are thus centered on one process: energy transfer from the beam electron to the nucleus. For this, one defines a differential cross section dσ/dE(εb,E)d\sigma/dE(\varepsilon_b,E) providing the distribution of energy transfers EE for a given beam energy εb\varepsilon_b. The McKinley and Feshbach differential cross section can adequately describe dσ/dEd\sigma/dE for light nuclei (Z<20Z < 20) [19–22]. Setting ℏ=c=1\hbar= c = 1,

dσdE=π(Zα∣pb∣β)2[Emax⁡E2−β(πZα+β)1E+πZαβEmax⁡E3],(1)\frac{d\sigma}{dE} = \pi\left(\frac{Z\alpha}{\lvert p_b\rvert\beta}\right)^2 \left[\frac{E_{\max}}{E^2} - \beta(\pi Z\alpha+\beta)\frac{1}{E} + \pi Z\alpha\beta\sqrt{\frac{E_{\max}}{E^3}}\right], \tag*{(1)}

where pbp_b is the momentum of the beam electron, β\beta is its velocity, ZZ is the atomic number of the target nucleus, and α\alpha is the fine structure constant.

One can then calculate the displacement cross section by integrating dσ/dEd\sigma/dE over all EE large enough to cause a displacement, so that

σ0(εb,Ed)=θ(Emax⁡(εb)−Ed)∫EdEmax⁡(εb)dσdE(εb,E)dE,(2)\sigma_0(\varepsilon_b,E_d) = \theta(E_{\max}(\varepsilon_b)-E_d)\int_{E_d}^{E_{\max}(\varepsilon_b)}\frac{d\sigma}{dE}(\varepsilon_b,E)dE, \tag*{(2)}

where Emax⁡(εb)E_{\max}(\varepsilon_b) is the maximum possible energy transfer for a given beam energy, i.e, the energy transfer resulting from a direct collision, and the step function θ\theta enforces that the cross section is zero when Emax⁡<EdE_{\max} < E_d. We will leave the step function implicit going forward for the sake of compactness. The critical energy εc\varepsilon_c is then the beam energy for which Emax⁡=EdE_{\max} = E_d. Therefore, the observation of sputtering at energies well below εc\varepsilon_c is completely at odds with equation (2). With this in mind, a fair amount of work has been done to treat deviations from equation (2). Notably, several studies have explored the effects of temperature on displacement cross sections [5, 6, 7]. This consideration involves calculating the degree to which the pre-collision thermal motion of the nucleus increases the cross section. However, these techniques essentially amount to smearing the beam energy dependence of the cross section, so that the cross section only strays significantly from equation (2) for beam energies very close to εc\varepsilon_c. Thus, temperature-induced increases in the cross section cannot account for the disparities between the equation (2) model and experiment. This necessitates the consideration of additional phenomena that can reduce EdE_d.

To address the limitations of equation (2), this work introduces a third party: the material’s electrons. Doing so brings two new interactions into play: one between the beam and material electrons and another between material electrons and nuclei. This yields a total of three interactions between the three pairs of particles (Figure 1). Therefore, the rate of beam-induced sputtering hinges on the rates of three processes mediated by these interactions.

Three interactions between three pairs of particles involved in electron beam-induced sputtering

Figure 1. Three interactions between three pairs of particles involved in electron beam-induced sputtering. In insulators, the probability of material electronic excitation (PiP_i) and the probability that those excitations are substantially long-lived (PfP_f) can be large. Therefore, the effect of exciting the material electrons on the sputtering rate should not be ignored in these materials.

  1. Beam and material electrons: a beam electron can excite some number nin_i ground state electrons to the conduction band (ii denotes the initial interaction with the beam). The probability of this event for a given beam energy ϵb\epsilon_b is Pi(ϵb,ni)P_i(\epsilon_b,n_i).

  1. Material electrons and nuclei: Some number nfn_f beam-induced excitations can survive long enough for the target atom to leave its original site (ff denotes the final system at the completion of sputtering). This depends on the nuclear kinetic energy EE and the excitation lifetime τ\tau. The probability that nfn_f of the nin_i excitations survive is Pf(E,τ,ni,nf)P_f(E,\tau,n_i,n_f).

  1. Beam electrons and material nuclei: the energy transferred to a material nucleus by the beam electron can exceed the PKA’s displacement threshold Ed(nf)E_d(n_f), which depends on the number of surviving excitations nfn_f. We define {Ed}\{E_d\} as the set of all displacement thresholds for all possible nfn_f. Sputtering occurs when E>Ed(nf)E > E_d(n_f). The differential cross section for an energy transfer EE from the beam electron to material nucleus is dσ/dE(ϵb,E)d\sigma/dE(\epsilon_b,E).

The sputtering cross section can then be calculated by coupling dσ/dEd\sigma/dE to PiP_i and PfP_f for all possible nin_i and nfn_f. With the terms defined above, this excitation-sensitive sputtering cross section can be written as

σ(ϵb,{Ed},τ)=∑ni=0∞Pi(ϵb,ni)∑nf=0ni∫Ed(nf)Emax⁡(ϵb)Pf(E,τ,ni,nf)dσdE(ϵb,E)dE.(3)\sigma(\epsilon_b,\{E_d\},\tau)=\sum_{n_i=0}^{\infty}P_i(\epsilon_b,n_i)\sum_{n_f=0}^{n_i}\int_{E_d(n_f)}^{E_{\max}(\epsilon_b)}P_f(E,\tau,n_i,n_f)\frac{d\sigma}{dE}(\epsilon_b,E)dE. \tag*{(3)}

If PiP_i and PfP_f are non-negligible when nin_i and nfn_f are nonzero, and EdE_d depends strongly on nfn_f, then interactions with the material electrons must be considered. We will later show that this makes σ\sigma in equation (3) larger than σ0\sigma_0 in equation (2) for all beam energies, most prominently when ϵb<ϵc\epsilon_b < \epsilon_c.

The remainder of this paper focuses on the derivations of PiP_i, PfP_f, and {Ed}\{E_d\}. § 3 describes how to combine QED with DFT to obtain PiP_i. § 5 then considers the evolution of the excited states during the sputtering process to derive {Ed}\{E_d\} and PfP_f. It then demonstrates how our formalism significantly improves the prediction of sputtering rates in hBN and MoS2_2.

Probability of beam-induced excitation

For a crystal in its ground state, an occupied electron energy eigenstate has zero overlap with any unoccupied state. However, the collision of a beam electron can give an occupied state a momentum boost that breaks this orthogonality. Thus, the boosted ground state has a nonzero probability of being measured in an excited state. We can use this idea to derive Pi(ϵb,ni)P_i(\epsilon_b,n_i), the probability that a beam electron with kinetic energy ϵb\epsilon_b excites exactly nin_i material electrons. The derivation can be broken down into four steps: (i) determine the amplitude for a free electron to scatter from one momentum eigenstate into another after collision with another free electron; (ii) generalize the formalism to obtain the amplitude for scattering from one wave packet into another by summing over the amplitudes for each momentum component of one wave packet to scatter into each momentum component of the other; (iii) decompose a pair of occupied and unoccupied crystal states into a momentum basis and plug them in as incoming and outgoing wave packets respectively, then square the amplitude to obtain the corresponding excitation probability for a particular transition; (iv) Compute the sum of all transition probabilities and use combinatorics to determine Pi(ϵb,ni)P_i(\epsilon_b,n_i). The following subsections address each step (i-iv) in detail.

Scattering of free electrons

We first derive the scattering amplitude for momentum transfer between two free electrons via Møller scattering [23–25]. With this, the interaction between actual material states can be written as a linear combination of these free-particle interactions. Going forward, we label the 4-momenta of the incoming electrons as p1p_1 and p2p_2, while the outgoing electrons have momenta p3p_3 and p4p_4. We also choose to make p1p_1 and p2p_2 components of the initial beam and material states respectively. The 4-momentum of the nnth electron can be written as pn=(ϵn,pnx,pny,pnz)=(ϵn,pn)p_n=(\epsilon_n,p_n^x,p_n^y,p_n^z)=(\epsilon_n,\mathbf{p}_n), where ϵn\epsilon_n is the particle’s energy and

pnp_n is its 3-momentum. Dot products between 4-vectors are then taken over Minkowski space, so that pn⋅pm=gμνpnμpmν=ϵnϵm−pn⋅pmp_n \cdot p_m = g_{\mu\nu}p_n^\mu p_m^\nu= \epsilon_n\epsilon_m-\mathbf{p}_n\cdot\mathbf{p}_m, where pn⋅pm=pnxpmx+pnypmy+pnzpmz\mathbf{p}_n\cdot\mathbf{p}_m = p_n^x p_m^x+p_n^y p_m^y+p_n^z p_m^z.

To lowest order, the amplitude for free electron scattering can be represented by two tree-level diagrams, which we call the tt- and uu-channels (Figure 2). Using Feynman’s rules [17,18], we can write these diagrams in terms of Dirac spinors, yielding the invariant matrix element

Feynman diagrams for electron-electron scattering

Figure 2. The lowest order electron-electron scattering perturbation includes two Feynman diagrams called the (a) tt-channel and (b) uu-channel. The incoming and outgoing electron states are represented by Dirac spinors us(p)u^{s}(p) and uˉs(p)\bar{u}^{s}(p) respectively, where pp and ss are the electron’s 4-momentum and spin index respectively. Subscripts 1 and 2 label components of the initial beam and material states respectively. The virtual photon 4-momentum qq is the momentum transfer between the electrons.

M(p4p3←p2p1)=e22∑s1∑s2∑s3∑s4[uˉs4(p4)γμus1(p1)(1p3−p2)2uˉs3(p3)γμus2(p2)+uˉs3(p3)γμus1(p1)(1p4−p2)2uˉs4(p4)γμus2(p2)](4)\mathcal{M}(p_4p_3 \leftarrow p_2p_1) = \frac{e^2}{2}\sum_{s_1}\sum_{s_2}\sum_{s_3}\sum_{s_4}\left[\bar{u}^{s_4}(p_4)\gamma^\mu u^{s_1}(p_1)\left(\frac{1}{p_3-p_2}\right)^2\bar{u}^{s_3}(p_3)\gamma_\mu u^{s_2}(p_2)+\bar{u}^{s_3}(p_3)\gamma^\mu u^{s_1}(p_1)\left(\frac{1}{p_4-p_2}\right)^2\bar{u}^{s_4}(p_4)\gamma_\mu u^{s_2}(p_2)\right] \tag*{(4)}

where sn=1s_n=1 or 2 denotes the spin of the nnth electron, us(p)u^s(p) is a Dirac spinor, and uˉs(p)\bar{u}^s(p) is its conjugate (§ 11). The factor of 1/2 before the summation arises from the assumption that the incoming states are spin unpolarized.

The first term in brackets is the tt-channel describing momentum transfer p3−p2p_3-p_2 and the second is the uu-channel describing momentum transfer p4−p2p_4-p_2. Because the DFT cutoff energy is much smaller than the beam energy, it is always the case that ∣p2∣\lvert\mathbf{p}_2\rvert is much smaller than ∣p1∣\lvert\mathbf{p}_1\rvert. Furthermore, we need only consider outgoing momenta for which the kinetic energy associated with either ∣p3∣\lvert\mathbf{p}_3\rvert or ∣p4∣\lvert\mathbf{p}_4\rvert falls within the DFT cutoff energy. In these cases, the magnitude of one outgoing momentum is similar to ∣p2∣\lvert\mathbf{p}_2\rvert, while that of the other is much greater. This means that one channel’s momentum transfer is always much larger than the other’s. As the momentum transfers reside in the denominators of either channel in equation (4), it follows that one channel always contributes much more to M\mathcal{M} than the other. Thus, when the tt-channel is significant, the uu-channel is negligible, and vice versa. Additionally, when integrating over all possible outgoing momenta, the contribution of the tt-channel is equal to that of the uu-channel. Taking advantage of this along with the indistinguishably of the electrons, we calculate only the tt-channel and multiply the resulting amplitude by 2 instead of calculating both channels and adding them. We can then define the 4-momentum transfer between the electrons as that of the tt-channel: q≡p3−p2q \equiv p_3-p_2. Because the tt-channel has a q2q^2 in the denominator, the resulting scattering probability is proportional to q−4q^{-4}. This makes large momentum transfers statistically irrelevant, allowing us to only consider momentum transfers inside the first Brillouin zone (BZ). The evaluation of the tt-channel in terms of the components of the electrons’ 4-momenta is straightforward, though cumbersome, and is described in § 11.

We can use the resulting M\mathcal{M} to obtain the free electron scattering amplitude

⟨p4p3∣T^∣p2p1⟩=(2π)4δ(p1+p2−p3−p4)M(p4p3←p2p1),(5)\langle p_4p_3|\hat{T}|p_2p_1\rangle=(2\pi)^4\delta(p_1+p_2-p_3-p_4)\mathcal{M}(p_4p_3\leftarrow p_2p_1), \tag*{(5)}

where T^\hat{T} is the scattering operator [17,18]. This gives the amplitude for two free electrons with momenta p1p_1 and p2p_2 to scatter into p3p_3 and p4p_4. Equation (5) is used to derive the scattering amplitude between two arbitrary wave packets in the next subsection.

Scattering of wave packets

The free particle scattering amplitude in equation (5) can be used to determine the amplitude for the scattering of two arbitrary electron states ϕ1\phi_1 and ϕ2\phi_2 into ϕ3\phi_3 and ϕ4\phi_4. This is obtained by sandwiching the scattering operator between the initial and final 2-particle states, i.e.,

⟨ϕ4ϕ3∣T^∣ϕ2ϕ1⟩=∫d3p4d3p3d3p2d3p1(2π)1216ϵ4ϵ3ϵ2ϵ1⟨ϕ4ϕ3∣p4p3⟩⟨p4p3∣T^∣p2p1⟩⟨p2p1∣ϕ2ϕ1⟩.(6)\langle\phi_4\phi_3|\hat{T}|\phi_2\phi_1\rangle=\int\frac{d^3p_4d^3p_3d^3p_2d^3p_1}{(2\pi)^{12}16\epsilon_4\epsilon_3\epsilon_2\epsilon_1}\langle\phi_4\phi_3|p_4p_3\rangle\langle p_4p_3|\hat{T}|p_2p_1\rangle\langle p_2p_1|\phi_2\phi_1\rangle. \tag*{(6)}

On the right side, we have inserted two resolutions of the identity given in equation (38). Inserting equation (5) into the integrand, the amplitude can be written in terms of the invariant matrix element M\mathcal{M}, becoming

∫d3p4d3p3d3p2d3p1(2π)1216ϵ4ϵ3ϵ2ϵ1⟨ϕ4∣p4⟩⟨ϕ3∣p3⟩⟨p2∣ϕ2⟩⟨p1∣ϕ1⟩\int\frac{d^3p_4d^3p_3d^3p_2d^3p_1}{(2\pi)^{12}16\epsilon_4\epsilon_3\epsilon_2\epsilon_1}\langle\phi_4|p_4\rangle\langle\phi_3|p_3\rangle\langle p_2|\phi_2\rangle\langle p_1|\phi_1\rangle
×M(p4p3←p2p1)(2π)4δ(p1+p2−p3−p4).(7){}\times\mathcal{M}(p_4p_3\leftarrow p_2p_1)(2\pi)^4\delta(p_1+p_2-p_3-p_4). \tag*{(7)}

Using the delta function to integrate over p4p_4 and p3zp_3^z yields

∫d2p3⊥d3p2d3p1(2π)816ϵ2ϵ1M(p4p3←p2p1)∣p3zϵ4−p4zϵ3∣⟨ϕ4∣p4⟩⟨ϕ3∣p3⟩⟨p2∣ϕ2⟩⟨p1∣ϕ1⟩,(8)\int\frac{d^2p_3^\perp d^3p_2d^3p_1}{(2\pi)^8 16\epsilon_2\epsilon_1}\frac{\mathcal{M}(p_4p_3\leftarrow p_2p_1)}{|p_3^z\epsilon_4-p_4^z\epsilon_3|}\langle\phi_4|p_4\rangle\langle\phi_3|p_3\rangle\langle p_2|\phi_2\rangle\langle p_1|\phi_1\rangle, \tag*{(8)}

where it is understood that p4p_4 and p3p_3 satisfy p1+p2=p3+p4p_1+p_2=p_3+p_4 (section S3). The normalization of 4-momentum states ∣pn⟩|p_n\rangle in terms of 3-momentum states ∣pn⟩|p_n\rangle as defined in equation (39) allows us to rewrite the expression as

∫d2p3⊥d3p2d3p1(2π)8ϵ4ϵ3ϵ2ϵ1M(p4p3←p2p1)4∣p3zϵ4−p4zϵ3∣⟨ϕ4∣p4⟩⟨ϕ3∣p3⟩⟨p2∣ϕ2⟩⟨p1∣ϕ1⟩.(9)\int\frac{d^2p_3^\perp d^3p_2d^3p_1}{(2\pi)^8}\sqrt{\frac{\epsilon_4\epsilon_3}{\epsilon_2\epsilon_1}}\frac{\mathcal{M}(p_4p_3\leftarrow p_2p_1)}{4|p_3^z\epsilon_4-p_4^z\epsilon_3|}\langle\phi_4|p_4\rangle\langle\phi_3|p_3\rangle\langle p_2|\phi_2\rangle\langle p_1|\phi_1\rangle. \tag*{(9)}

We can then discretize the momenta by replacing d3pi/(2π)3d^3p_i/(2\pi)^3 with V−1V^{-1} and d2pi⊥/(2π)2d^2p_i^\perp/(2\pi)^2 with A−1A^{-1}, where VV and AA are the volume and cross sectional area of the simulated crystal, i.e., the volume and cross sectional area of the unit cell times the number of kk-points used to sample the BZ. With this, the amplitude for electron states ∣ϕ1⟩\lvert\phi_1\rangle and ∣ϕ2⟩\lvert\phi_2\rangle to scatter into ∣ϕ3⟩\lvert\phi_3\rangle and ∣ϕ4⟩\lvert\phi_4\rangle takes the form

1AV2∑p3∑p2∑p1ϵ4ϵ3ϵ2ϵ1M(p4p3←p2p1)4∣p3zϵ4−p4zϵ3∣⟨ϕ4∣p4⟩⟨ϕ3∣p3⟩⟨p2∣ϕ2⟩⟨p1∣ϕ1⟩.(10)\frac{1}{AV^2}\sum_{\mathbf{p}_3}\sum_{\mathbf{p}_2}\sum_{\mathbf{p}_1}\sqrt{\frac{\epsilon_4\epsilon_3}{\epsilon_2\epsilon_1}}\frac{\mathcal{M}(p_4p_3\leftarrow p_2p_1)}{4\lvert p_3^z\epsilon_4-p_4^z\epsilon_3\rvert}\langle\phi_4\lvert p_4\rangle\langle\phi_3\lvert p_3\rangle\langle p_2\lvert\phi_2\rangle\langle p_1\lvert\phi_1\rangle. \tag*{(10)}

In the next subsection, we replace ∣ϕ1…4⟩\lvert\phi_{1\ldots4}\rangle with states relevant to electron beam-induced excitation.

Probability of a crystal excitation

We now consider the specific case of beam-induced excitations to determine the form of the four electron states in equation (10). We assign ∣ϕ1⟩\lvert\phi_1\rangle and ∣ϕ4⟩\lvert\phi_4\rangle to the initial and final beam states ∣pb⟩\lvert p_b\rangle and ∣pb′⟩\lvert p_b'\rangle respectively. States ∣ϕ2⟩\lvert\phi_2\rangle and ∣ϕ3⟩\lvert\phi_3\rangle are then the ground and excited crystal states ∣nk⟩\lvert nk\rangle and ∣n′k′⟩\lvert n'k'\rangle respectively, where nn and n′n' are band indices and kk and k′k' are k-points. Substituting these specific states into expression (10), the amplitude for exciting ∣nk⟩\lvert nk\rangle to ∣n′k′⟩\lvert n'k'\rangle becomes

⟨pb′,n′k′∣T^∣nk,pb⟩=1AV2∑p3∑p2∑p1ϵ4ϵ3ϵ2ϵ1M(p4p3←p2p1)4∣p3zϵ4−p4zϵ3∣\langle p_b',n'k'\lvert\hat{T}\lvert nk,p_b\rangle=\frac{1}{AV^2}\sum_{\mathbf{p}_3}\sum_{\mathbf{p}_2}\sum_{\mathbf{p}_1}\sqrt{\frac{\epsilon_4\epsilon_3}{\epsilon_2\epsilon_1}}\frac{\mathcal{M}(p_4p_3\leftarrow p_2p_1)}{4\lvert p_3^z\epsilon_4-p_4^z\epsilon_3\rvert}
×⟨pb′∣p4⟩⟨n′k′∣p3⟩⟨p2∣nk⟩⟨p1∣pb⟩.(11){}\times\langle p_b'\lvert p_4\rangle\langle n'k'\lvert p_3\rangle\langle p_2\lvert nk\rangle\langle p_1\lvert p_b\rangle. \tag*{(11)}

The values of ϵ1…4\epsilon_{1\ldots4} need to be clarified before moving forward. The zeroth components of the initial and final beam momenta obey the free particle dispersion relations, so ϵ1=∣p1∣2+m2\epsilon_1=\sqrt{\lvert\mathbf{p}_1\rvert^2+m^2} and ϵ4=∣p4∣2+m2\epsilon_4=\sqrt{\lvert\mathbf{p}_4\rvert^2+m^2}. The beam energy that appears in equation (3) is then defined as the beam electron’s total energy minus its rest mass: ϵb≡ϵ1−m\epsilon_b\equiv\epsilon_1-m. Meanwhile, the momentum of the crystal states can be treated nonrelativistically. Thus, the zeroth components of the crystal state momenta are the energy eigenvalues of the crystal state plus the electron rest mass, i.e., ϵ2=ϵnk+m\epsilon_2=\epsilon_{nk}+m and ϵ3=ϵn′k′+m\epsilon_3=\epsilon_{n'k'}+m. For the remainder of this derivation, we continue to leave our expressions in terms of ϵ1…4\epsilon_{1\ldots4} for compactness.

Sputtering from a 2D crystal often requires that the beam electron is backscattered or nearly backscattered, in which case, its final trajectory after collision with the nucleus is nearly antiparallel to its initial trajectory and perpendicular to the crystal surface. Given that many 2D materials (including hBN and MoS2_2) possess inversion and/or reflection symmetry about the crystal plane, we assume that the likelihood of excitation before and after the collision are about equal. In light of this, we calculate the excitation probability during a sputtering event assuming the beam electron’s trajectory is not altered by its collision with the nucleus. That is, we impose that p1=∣p1∣z^\mathbf{p}_1=\lvert\mathbf{p}_1\rvert\hat{\mathbf{z}} until an electronic excitation is induced.

We can now evaluate the bra-ket products in equation (11). The initial beam state is highly localized on pbp_b, meaning that

∣⟨p1∣pb⟩∣2=Vδp1,pb⇒⟨p1∣pb⟩=V1/2δp1,pb.(12)\lvert\langle p_1\lvert p_b\rangle\rvert^2=V\delta_{p_1,p_b}\quad\Rightarrow\quad\langle p_1\lvert p_b\rangle=V^{1/2}\delta_{p_1,p_b}. \tag*{(12)}

Meanwhile, the ground and excited crystal states can be expanded into a plane-wave basis, so that

∣nk⟩=V−1/2∑GCG+kn∣G+k⟩|n\mathbf{k}\rangle= V^{-1/2}\sum_{\mathbf{G}} C^{n}_{\mathbf{G}+\mathbf{k}}|\mathbf{G}+\mathbf{k}\rangle
∣n′k′⟩=V−1/2∑GCG+k′n′∣G+k′⟩,(13)|n'\mathbf{k}'\rangle= V^{-1/2}\sum_{\mathbf{G}} C^{n'}_{\mathbf{G}+\mathbf{k}'}|\mathbf{G}+\mathbf{k}'\rangle, \tag*{(13)}

where each G\mathbf{G} is a reciprocal lattice vector. By re-expressing p2\mathbf{p}_2 and p3\mathbf{p}_3 as G2+k2\mathbf{G}_2+\mathbf{k}_2 and G3+k3\mathbf{G}_3+\mathbf{k}_3 respectively, we find

⟨p2∣nk⟩=⟨G2+k2∣(V−1/2∑GCG+kn∣G+k⟩)=V1/2CG2+knδk2,k.\langle\mathbf{p}_2|n\mathbf{k}\rangle= \langle\mathbf{G}_2+\mathbf{k}_2|\left(V^{-1/2}\sum_{\mathbf{G}} C^{n}_{\mathbf{G}+\mathbf{k}}|\mathbf{G}+\mathbf{k}\rangle\right) = V^{1/2}C^{n}_{\mathbf{G}_2+\mathbf{k}}\delta_{\mathbf{k}_2,\mathbf{k}}.
⟨p3∣n′k′⟩=⟨G3+k3∣(V−1/2∑GCG+k′n′∣G+k′⟩)=V1/2CG3+k′n′δk3,k′,(14)\langle\mathbf{p}_3|n'\mathbf{k}'\rangle= \langle\mathbf{G}_3+\mathbf{k}_3|\left(V^{-1/2}\sum_{\mathbf{G}} C^{n'}_{\mathbf{G}+\mathbf{k}'}|\mathbf{G}+\mathbf{k}'\rangle\right) = V^{1/2}C^{n'}_{\mathbf{G}_3+\mathbf{k}'}\delta_{\mathbf{k}_3,\mathbf{k}'}, \tag*{(14)}

Lastly, we do not care where the outgoing scattered electron ends up, so we wish for ∣pb′⟩|\mathbf{p}'_b\rangle to satisfy

∣⟨p4∣pb′⟩∣2=V⇒⟨p4∣pb′⟩=V1/2.(15)|\langle\mathbf{p}_4|\mathbf{p}'_b\rangle|^2=V \quad\Rightarrow\quad\langle\mathbf{p}_4|\mathbf{p}'_b\rangle=V^{1/2}. \tag*{(15)}

The excitation amplitude is then obtained by plugging in the bra-ket products from equations (12), (14), and (15) into equation (11). This gives us the excitation amplitude

⟨pb′,n′k′∣T^∣nk,pb⟩=1A∑G3∑G2ϵ4ϵ3ϵ1ϵ2M(p4p3←p2p1)∣p32ϵ4−p42ϵ3∣CG3+k′n′∗CG2+kn.(16)\langle\mathbf{p}'_b,n'\mathbf{k}'|\hat{T}|n\mathbf{k},\mathbf{p}_b\rangle= \frac{1}{A}\sum_{\mathbf{G}_3}\sum_{\mathbf{G}_2} \sqrt{\frac{\epsilon_4\epsilon_3}{\epsilon_1\epsilon_2}} \frac{\mathcal{M}(\mathbf{p}_4\mathbf{p}_3\leftarrow\mathbf{p}_2\mathbf{p}_1)} {\left|\mathbf{p}_3^2\epsilon_4-\mathbf{p}_4^2\epsilon_3\right|} C^{n'*}_{\mathbf{G}_3+\mathbf{k}'}C^{n}_{\mathbf{G}_2+\mathbf{k}}. \tag*{(16)}

where it is understood that p2=(m+ϵnk,G2+k)\mathbf{p}_2=(m+\epsilon_{n\mathbf{k}},\mathbf{G}_2+\mathbf{k}) and p3=(m+ϵn′k′,G3+k′)\mathbf{p}_3=(m+\epsilon_{n'\mathbf{k}'},\mathbf{G}_3+\mathbf{k}'), and p4\mathbf{p}_4 and pb′\mathbf{p}'_b satisfy p1+p2=p4+p3\mathbf{p}_1+\mathbf{p}_2=\mathbf{p}_4+\mathbf{p}_3. Squaring this amplitude yields the probability of a single electronic excitation from the valence band state ∣nk⟩|n\mathbf{k}\rangle to the conduction band state ∣n′k′⟩|n'\mathbf{k}'\rangle for a given beam energy ϵb\epsilon_b, that is,

P(n′k′←nk∣ϵb)=∣⟨pb′,n′k′∣T^∣nk,pb⟩∣2.(17)P(n'\mathbf{k}'\leftarrow n\mathbf{k}|\epsilon_b)=\left|\langle\mathbf{p}'_b,n'\mathbf{k}'|\hat{T}|n\mathbf{k},\mathbf{p}_b\rangle\right|^2. \tag*{(17)}

Probability of nin_i excitations

We are finally ready to derive Pi(ϵb,ni)P_i(\epsilon_b,n_i), the probability that a beam electron excites a particular number of electrons nin_i. First, we define the sum of all transition probabilities

S(ϵb)≡∑k∑k′∑n∑n′P(n′k′←nk∣ϵb)≡∑j=1NPj(ϵb),(18)S(\epsilon_b)\equiv\sum_{\mathbf{k}}\sum_{\mathbf{k}'}\sum_n\sum_{n'}P(n'\mathbf{k}'\leftarrow n\mathbf{k}|\epsilon_b)\equiv\sum_{j=1}^{N}P_j(\epsilon_b), \tag*{(18)}

where k\mathbf{k} and k′\mathbf{k}' run over all k-points, nn over the valence bands, and n′n' over the conduction bands (those possessing states with energy between the Fermi level and the work function). The index jj in the right-most expression labels the possible single-particle excitations (e.g., j=n′k′←nkj=n'\mathbf{k}'\leftarrow n\mathbf{k}).

Combinatorics tells us that the probability of exciting exactly one excitation is

Pi(ni=1)=∑j1Pj1∏j2≠j1(1−Pj2)=∑j1Pj1(1−∑j2≠j1Pj2+12∑j3≠j1,j2∑j2≠j1Pj2Pj3−⋯ ).(19)\begin{aligned} P_i(n_i=1)=\sum_{j_1}P_{j_1}\prod_{j_2\ne j_1}(1-P_{j_2}) \\ &=\sum_{j_1}P_{j_1}\left(1-\sum_{j_2\ne j_1}P_{j_2}+\frac{1}{2}\sum_{j_3\ne j_1,j_2}\sum_{j_2\ne j_1}P_{j_2}P_{j_3}-\cdots\right). \tag*{(19)} \end{aligned}

In the large-crystal limit, the number of states, and thus the number of transitions, is large so that the summations over jij_i in equation (19) are approximately equal to one another. That is,

∑j2≠j1Pj2∼∑j2Pj2=S.(20)\sum_{j_2\ne j_1}P_{j_2}\sim\sum_{j_2}P_{j_2}=S. \tag*{(20)}

In this limit, the probability of exactly one beam-induced excitation can be written as

Pi(ni=1)∼∑jPj(1−S+12S2−⋯ )=e−S∑jPj=Se−S.(21)P_i(n_i=1)\sim\sum_jP_j\left(1-S+\frac{1}{2}S^2-\cdots\right)=e^{-S}\sum_jP_j=Se^{-S}. \tag*{(21)}

In the same way, the probability of two excitations is

Pi(ni=2)=12∑j1∑j2≠j1Pj1Pj2∏j3≠j1,j2(1−Pj3)∼S22e−S.(22)P_i(n_i=2)=\frac{1}{2}\sum_{j_1}\sum_{j_2\ne j_1}P_{j_1}P_{j_2}\prod_{j_3\ne j_1,j_2}(1-P_{j_3})\sim\frac{S^2}{2}e^{-S}. \tag*{(22)}

In general, the probability of exactly nin_i beam-induced excitations is approximately

Pi(ϵb,ni)∼S(ϵb)nini!e−S(ϵb).(23)P_i(\epsilon_b,n_i)\sim\frac{S(\epsilon_b)^{n_i}}{n_i!}e^{-S(\epsilon_b)}. \tag*{(23)}

Thus, we see that the probability PiP_i can be written purely in terms of SS.

We can now use formula (23) to calculate excitation probabilities in hBN and MoS2_2 (Figure 3). DFT is used to obtain the plane-wave coefficients CG2+knC_{\mathbf{G}_2+\mathbf{k}}^{n} and CG3+k′n′C_{\mathbf{G}_3+\mathbf{k}'}^{n'} and eigenvalues ϵkn\epsilon_{\mathbf{k}}^{n} and ϵk′n′\epsilon_{\mathbf{k}'}^{n'} for the pristine unit cell of each material. These are plugged into equation (16) to obtain the amplitude for each transition. We sum over the squares of all resulting amplitudes to obtain SS, which is then plugged into formula (23) to obtain PiP_i for both materials. We emphasize that the only DFT calculation needed to determine PiP_i is the electronic structure relaxation of a pristine unit cell, a very inexpensive calculation.

Probabilities of initially exciting a certain number of electrons in hBN and MoS₂

Figure 3. Probabilities of initially exciting a certain number of electrons nin_i in (a) hBN and (b) MoS2_2 with a beam electron. The excitation probabilities tend to decreases with increasing beam energy. The excitation probabilities in MoS2_2 are larger than those in hBN because MoS2_2 has a significantly smaller band gap.

The probabilities plotted in Figure 3 reveal some notable trends. First, for sufficiently large beam energies, Pi(ϵb,ni)P_i(\epsilon_b,n_i) decreases with increasing ϵb\epsilon_b for all ni>0n_i>0. This is because a faster beam electron has less time to interact with the material and cause an excitation. In this regime, Pi(ϵb)P_i(\epsilon_b) is proportional to ϵb−1\epsilon_b^{-1}, a relationship originally predicted by Bethe [12, 26, 27]. This means that multiple excitations are more likely at low beam energies. Furthermore, the probability of remaining in the ground state P(ϵb,ni=0)P(\epsilon_b,n_i=0) vanishes as ϵb\epsilon_b goes to zero. This implies that a stationary electron in the vicinity of a material is guaranteed to interact with the material’s electrons and affect its electronic structure. However, the validity of formula (23) diminishes as ϵb\epsilon_b approaches zero. In the case of a slow beam electron, the interaction between the beam and material electrons can no longer be approximated by the single virtual photon transfer processes depicted in Figure 2, as the amplitudes for higher-order processes become more significant. The effects of processes beyond the tree-level should be the subject of future work. Lastly, the excitation probability is inversely proportional to the material’s band gap. This is because the zeroth component of the momentum transfer qq depicted in Figure 2 is the difference in energy eigenvalues between the occupied and unoccupied states (§ 3.3). Thus, the smallest possible denominator of the tt-channel in equation (4) is proportional to the difference in eigenvalues squared. The experimentally measured band gap of MoS2_2 about 1.9 eV [28] while that of hBN is about 6.1 eV [29]. This means MoS2_2 hosts transitions with smaller eigenvalue differences, making the summands in equation (16) larger.

With Pi(ϵb,ni)P_i(\epsilon_b,n_i) derived in formula (23) and dσ/dE(ϵb,E)d\sigma/dE(\epsilon_b,E) defined in equation (1), we now have two of the three functions depicted in Figure 1 needed to calculate the sputtering cross section in equation (3). The final ingredients are Pf(E,τ,ni,nf)P_f(E,\tau,n_i,n_f), the probability that nfn_f excitations survive the displacement event given nin_i initial beam-induced excitations, and {Ed}\{E_d\}, the set of all displacement thresholds for all nfn_f. The derivations of these objects are described in the next section.

Sputtering cross section

We have demonstrated how the interaction between the beam and material electrons can induce electronic excitations in the material. In this section, we show how these excitations bring about much larger sputtering cross sections than those predicted by a ground state theory. We start by showing how electronic excitations can reduce the displacement threshold EdE_d. We then show that longer excitation lifetimes increase PfP_f for nonzero nfn_f, giving the beam-induced excitations more opportunities to lessen EdE_d. This motivates us to write the sputtering cross section in equation (3) analytically in terms of the excitation lifetime τ\tau. Finally, the resulting equation is used to predict the sputtering rates of boron and nitrogen in hBN and sulfur in MoS2_2, which can be made to agree well with experiment for appropriate values of τ\tau for each material.

Effect of excitations on the displacement threshold

We begin by describing how beam-induced excitations can reduce the displacement threshold EdE_d in a process called bond softening. As EdE_d is the lower bound of integration over EE in equation (3), and the differential cross section in equation (1) behaves like E−2E^{-2} for small EE, reductions in EdE_d can greatly increase the sputtering cross section. Exactly how excitations change EdE_d is an ambitious study on its own, requiring a careful consideration of the excited electrons’ evolution and various relaxation pathways [30]. Here instead, we make three simplifying assumptions that allow us to calculate EdE_d with only ground state DFT.

Before describing these assumptions, we first define some important terms. Consider the moment immediately after the beam electron collides with a material nucleus. The nucleus now has a velocity corresponding to the kinetic energy EE transferred from the beam electron. The resulting nuclear motion away from its equilibrium position causes the energy of the system to increase. In this sense, the system climbs an energy surface from the bottom of its equilibrium well. Far away from the well’s bottom, the energy surface eventually plateaus. If the system reaches this plateau, the displaced PKA moves freely away from its initial site without deceleration. At this point, we consider the PKA to have sputtered. We call the energy at the well’s bottom E0E_0 and that at the plateau EsE_s. If the energy surface is static throughout the entire process, then the displacement threshold is simply Ed=Es−E0E_d = E_s - E_0. When E>EdE > E_d, the system has enough energy to climb out of the well, and the PKA sputters. Our task now is to determine how beam-induced excitations in the material electrons and their subsequent relaxation affect E0E_0 and EsE_s. To facilitate this, we make the following assumptions.

Assumption 1: the excited electrons and holes occupy the band edges. To justify this, we shift our focus to the material electrons immediately after the collision. There are now nin_i electrons in the conduction band and nin_i holes in the valence bands. Kretschmer et al. simulated the time-evolution of these excitations using Ehrenfest dynamics [12,31]. They found that the excited electrons relax nonradiatively to the conduction band minimum (CBM) in a few femtoseconds, while the holes in the valence band take a similar amount of time to relax to the valence band maximum (VBM). In contrast, the PKA takes several hundreds of femtoseconds to fully sputter [7]. Thus, the nonradiative relaxation of the electronic structure is essentially instantaneous, and we can assume that all excited electrons and holes occupy the CBM and VBM respectively before the PKA has been displaced by an appreciable amount.

These findings greatly simplify the calculation of E0E_0. Given nin_i excitations, E0E_0 is just the energy of the pristine system with nin_i electrons and holes in the CBM and VBM respectively. In the large crystal limit, this amounts to adding the pristine band gap EgE_g to the system’s ground state energy nin_i times. In other words, E0(ni)=E0(0)+niEgE_0(n_i) = E_0(0) + n_i E_g. This approximation of course ignores any binding energy between the electron and hole. However, we believe that this formalism allows for an efficient treatment of the lowest order effects of excitation on E0E_0.

Assumption 2: the sputtered system is in its ground state. Finding the plateau energy EsE_s can be a bit more involved. To calculate it properly, one must track how the excitations in the CBM and VBM evolve as the PKA moves away from the crystal. This would require Ehrenfest dynamics of a supercell over timescales of hundreds of femtoseconds, which is prohibitively expensive. Here we seek a much less costly set of DFT calculations that can still provide a reasonable approximation for EsE_s. To this end, we draw upon another finding of Kretschmer et al. [12]. After excitation, the displacement of the PKA causes the occupied and unoccupied CBM and VBM states to converge into the band gap and localize on the resulting defect. These converging states are the bonding and antibonding states that connect the PKA to the host crystal. Thus, we should consider how the beam-induced excitations affect the electronic structure on both the PKA and the remaining vacancy.

We start by considering how the ground state eigenvalues evolve as the PKA moves away from the crystal. For hBN, the evolution of eigenvalues differ depending on whether boron or nitrogen is sputtered (Figure 4). Boron is electropositive while nitrogen is electronegative. Thus, in an hBN crystal, the nitrogen atoms borrow negative charge from the neighboring boron atoms. This means that the pp-orbital states that would be occupied on an isolated boron atom are vacant, hovering in the conduction band of hBN. Conversely, the pp-orbital states of nitrogen lie occupied in the valence band. For both atoms, separation from the crystal causes those states to converge to degenerate pp-orbitals on their respective atoms. When boron is sputtered, these states come from the conduction band. This means that excited electrons residing in the CBM tend to transfer negative charge to the sputtered boron. However, this charge transfer is quite energetically unfavorable since boron is electropositive. The energy cost of this is much larger than the band gap of hBN, compelling any excess electrons on boron to relax to the host crystal. On the other hand, the states that localize on sputtered nitrogen must rise from the valence band. Thus, the holes at the VBM tend to transfer positive charge to the sputtered nitrogen, which is again energetically unfavorable for the electronegative atom. Thus, because charge transfer to the sputtered PKA has such a large energy cost, we presume that in most cases, the PKA becomes charge neutral by the time sputtering has occurred.

Eigenvalues and occupations of the eigenstates of armchair hBN throughout the sputtering process

Figure 4. Eigenvalues and occupations of the eigenstates of armchair hBN throughout the sputtering process with respect to the distance between sputtered (a) boron and (b) nitrogen and their pristine sites. The thickness of each curve is proportional to the localization of the corresponding state on the sputtered atom. In hBN, the electronegative nitrogen atoms borrow charge from the neighboring boron atoms. Thus, the pp-orbital states that would be occupied on an isolated boron atom reside in the conduction band of hBN and descend into the gap as boron sputters. On the other hand, most of the pp-orbital states of nitrogen lie in the valence band and rise into the gap during nitrogen sputtering.

Furthermore, because the states that localize on the PKA converge to the same degenerate p-orbital, the PKA assumes its ground state after it has sputtered, regardless of how many electrons the beam excites. We call the energy of the isolated ground state atom EaE_a. Meanwhile, the convergence of the bonding and antibonding states localized on the remaining vacancy during the displacement causes the excited electron and hole energy levels to cross multiples times. This means that there are plenty of chances for an excited electron to relax nonradiatively as the PKA separates from the material. Thus, we assume that the host crystal also relaxes to its ground state for most sputtering events.

Lastly, the sputtering threshold is always larger than the vacancy formation energy, because some of the energy transferred to the PKA disperses to the neighboring atoms. For this reason, we calculate the energy of the vacant system without relaxing the structure, as it has been shown that the free energy gained by leaving the structure unrelaxed roughly matches the energy dissipated to the surrounding material [32]. We call the energy of the unrelaxed ground state vacancy EvE_v. Thus, we calculate the plateau energy using Es=Ev+EaE_s = E_v + E_a, the sum of the ground state vacancy and isolated atom free energies.

Assumption 3: the PKA travels at a constant velocity. Each number of excitations creates its own energy surface. In this way, electronic excitation and relaxation cause the system to hop from surface to surface. In our problem, the initial beam-induced excitation perches the system on an elevated surface. Once there, electronic relaxation via spontaneous emission (SE) enables downward surface hopping at the cost of the emitted photon energy. In this framework, EdE_d is the energy gained along all energy surfaces that the system traverses during sputtering. For example, suppose the system emits one photon as it sputters. EdE_d is then the energy gained along the portion of the excited energy surface traversed before emission plus that gained along the relaxed surface after emission. This is how SE can change EdE_d. However, the effect of SE on EdE_d must diminish as the PKA moves further away from the crystal. This means there must be some distance dd beyond which SE no longer affects the forces on the sputtered atom, leaving the energy surface unchanged. We take d=4.5d = 4.5 Å in line with Kretschmer et al.’s study, though we acknowledge that the exact meaning of this distance is different in their work [12].

We wish to capture this idea while making the calculation of PfP_f as intuitive as possible. We do this by assuming that each energy surface is constant except for a step when the PKA reaches a distance dd away from its equilibrium site. This means that immediately after the collision that excites nin_i electrons, the surface assumes a constant energy E0(ni)E_0(n_i). Each relaxation via SE causes the system to drop to the surface beneath

it, decreasing its energy by EgE_g. We define nfn_f as the number of electronic excitations that survive the PKA’s traversal of distance dd to the energy step. Thus, the energy of the system just before the PKA reaches the step is E0(nf)E_0(n_f). Beyond the step, all energy surfaces have energy EsE_s, and SE no longer changes the energy surface. With this formalism, EdE_d is simply the height of the step,

Ed(nf)=Es−E0(nf)=Ev+Ea−E0(0)−nfEg.(24)E_d(n_f) = E_s - E_0(n_f) = E_v + E_a - E_0(0) - n_f E_g. \tag*{(24)}

Thus we have derived a simple relationship for how electronic excitations affect EdE_d. However, before we show how this relationship affects the sputtering cross section, we must first point out an important caveat. Equation (24) suggests that EdE_d can be negative for large nfn_f. In these cases, the affected atom accelerates away from its pristine site even without energy transfer from the beam electron. This would seem to suggest that the sputtering cross section is infinite, since the PKA would sputter for any energy transfer, no matter how small. This is corroborated by the fact that the integral in equation (3) diverges as EdE_d approaches zero, meaning that the sputtering cross section σ\sigma would approach infinity. However, this is nonsensical, as it would imply an infinite sputtering rate for a finite beam current. In reality, even though the beam electron does nudge all of the material nuclei to some extent, one nucleus always receives a bigger nudge than the rest. This is the atom onto which the beam-induced excitations localize. Thus, for each beam electron, only one atom’s displacement threshold is reduced by electronic excitations. For this reason, the displacement cross section should never exceed the cross sectional area occupied by a single target atom. This means that EdE_d, the lower integration bound in equation (3), must have a lower bound itself, which we call Emin⁡E_{\min}. When EdE_d falls below Emin⁡E_{\min}, we replace it with Emin⁡E_{\min}. We can approximate Emin⁡E_{\min} by setting the displacement cross section in equation (2) equal to the area occupied by the PKA and solving for the displacement threshold (§ 17). With this, we are finally ready to use equation (24) to determine the set of displacement thresholds {Ed}\{E_d\} to insert into equation (3).

All positive displacement thresholds computed for this work are listed in Table 1. Explanations for the calculations of EvE_v, EaE_a, E0(0)E_0(0), and EgE_g are given in subsections 4.3 and 4.4. The three largest displacement thresholds for MoS2_2 with nf=0n_f = 0, 11, and 22 excitations are similar to those calculated with DFT-based molecular dynamics simulations [12]. This provides some assurance that our simplified approach to calculating EdE_d yields reasonable results. Excitation numbers nfn_f greater than those listed in Table 1 make EdE_d negative, in which case the exact value of EdE_d is not important since it is replaced with Emin⁡E_{\min} in the calculation of σ\sigma. With that said, it is critical to consider large nfn_f, even if the resulting EdE_d is less than Emin⁡E_{\min}. These large nfn_f can make appreciable contributions to the total cross section, especially at small beam energies for which Pi(ϵb,ni)P_i(\epsilon_b,n_i) is significant for large nin_i (Figure 3). As nfn_f cannot exceed nin_i, this means that we must consider sufficiently large nin_i to acknowledge these contributions. Eventually, these contributions diminish as we increment nin_i, because the ni!n_i! in the denominator of PiP_i in formula (23) eventually outgrows the SniS^{n_i} in the numerator. We can therefore truncate the summation over nin_i in equation (3) at some adequately large nimax⁡n_i^{\max}. This means that nimax⁡n_i^{\max} must be converged for each material, as materials with greater SS require greater nimax⁡n_i^{\max}. In this work, we choose nimax⁡n_i^{\max} large enough so that including more excitations increases the sputtering cross section by less than 1% for the smallest experimental beam energy considered for each material (Figure S1).

displacement threshold (eV)
nfn_f0123
B from hBN12.858.784.710.64
N from hBN12.718.644.570.50
S from MoS2_26.925.043.161.28

Table 1. All computed positive displacement thresholds for sputtering from the hBN armchair edge and MoS2_2 surface. The thresholds decrease as the number of surviving excitations nfn_f increases.

Convergence plots of the sputtering cross section for B from hBN, N from hBN, and S from MoS₂

Figure S1. Convergence of the sputtering cross section with respect to the maximum number of beam-induced excitations nimax⁡n_i^{\max} considered. The simulated beam energies ϵb\epsilon_b are the lowest experimental beam energies used for each material. The excitation lifetimes τ\tau are those used to fit the experimental data in figures 5 and 6. Each cross section is deemed converged when any increase in nimax⁡n_i^{\max} increases the cross section by less than 1%.

Thus, we have shown how EdE_d depends only on the number of surviving excitations nfn_f. We must now determine the likelihood that nfn_f excitations survive given nin_i beam-induced excitations. This is done by comparing the excitation lifetime τ\tau to tst_s, the time it takes for the PKA to travel a distance dd. We handle this task in the next subsection, where we write the sputtering cross section in equation (3) in terms of τ\tau.

Sputtering cross section in terms of the excitation lifetime

In hBN and MoS2_2, occupation of the antibonding state localized on the PKA does not affect the free energy of the sputtered system, since the antibonding and bonding states converge to the same degenerate p-orbital in the limit that the PKA is isolated from the host material. Therefore, only the excitation lifetime of the host material needs be considered in calculating Pf(ni,nf)P_f(n_i,n_f), the probability that nfn_f of the nin_i beam-induced excitations survive long enough to reduce the displacement threshold.

We define the ratio of surviving excitations as

R(E,τ)=e−ts(E)/τ,(25)R(E,\tau)=e^{-t_s(E)/\tau}, \tag*{(25)}

where tst_s is the time it takes for the sputtered atom to travel a distance dd, and the excitation lifetime τ\tau is determined for each material by fitting the cross section to experimental data. Given nin_i beam-induced excitations, the probability that nfn_f excitations survive is

Pf(ni,nf)=(ninf)Rnf(1−R)ni−nf=(ninf)∑n=0ni−nf(ni−nfn)(−1)nRnf+n.(26)P_f(n_i,n_f)=\binom{n_i}{n_f}R^{n_f}(1-R)^{n_i-n_f}=\binom{n_i}{n_f}\sum_{n=0}^{n_i-n_f}\binom{n_i-n_f}{n}(-1)^nR^{n_f+n}. \tag*{(26)}

Equation (26) allows us to rewrite the integral in equation (3) as

∫EdEmax⁡PfdσdE dE=(ninf)∑n=0ni−nf(ni−nfn)(−1)n∫EdEmax⁡Rnf+ndσdE dE.(27)\int_{E_d}^{E_{\max}}P_f\frac{d\sigma}{dE}\,dE=\binom{n_i}{n_f}\sum_{n=0}^{n_i-n_f}\binom{n_i-n_f}{n}(-1)^n\int_{E_d}^{E_{\max}}R^{n_f+n}\frac{d\sigma}{dE}\,dE. \tag*{(27)}

Assumption 3 from the previous subsection again aids us here. Because the PKA travels at a constant velocity until it reaches the sputtering distance dd, we can write tst_s from

equation (25) quite simply as ts=dM/2Et_s = d\sqrt{M/2E}. This allows for the straightforward analytical integration of equation (27). Using dσ/dEd\sigma/dE defined in equation (1), the integral on the right can be written as

∫Rnf+ndσdEdE=π(Zα∣p∣β)2[4τ2Emax⁡d2Mμξ+1μ2e−μξ+2β(πZα+β)Ei⁡(−μξ)+πZαβτμd8Emax⁡Me−μξ]+const.,(28)\int_{R^{n_f+n}} \frac{d\sigma}{dE}dE = \pi\left(\frac{Z\alpha}{|\mathbf{p}|\beta}\right)^2 \left[\frac{4\tau^2 E_{\max}}{d^2M}\frac{\mu\xi+1}{\mu^2}e^{-\mu\xi} + 2\beta(\pi Z\alpha+\beta)\operatorname{Ei}(-\mu\xi) + \frac{\pi Z\alpha\beta\tau}{\mu d}\sqrt{\frac{8E_{\max}}{M}}e^{-\mu\xi}\right] + \mathrm{const.,} \tag*{(28)}

where we define

ξ≡−dτM2Eandμ≡nf+n,(29)\xi\equiv-\frac{d}{\tau}\sqrt{\frac{M}{2E}} \quad\text{and} \quad\mu\equiv n_f+n, \tag*{(29)}

and the function Ei is the exponential integral,

Ei⁡(x)=∫−∞xett dt.(30)\operatorname{Ei}(x) = \int_{-\infty}^{x}\frac{e^t}{t}\,dt. \tag*{(30)}

Looking back at equation (3), we now have everything we need to evaluate the sputtering cross section. We derived Pi(ϵb,ni)P_i(\epsilon_b,n_i) in formula (23), and we performed the integration over EE analytically in equations (27) and (28), setting {Ed}\{E_d\} in equation (24) as the lower integration bounds. We also have a criterion to truncate the summation over nin_i at nimax⁡n_i^{\max} for a given material, as described at the end of section 4.1. In the following subsections, we use these results to calculate the sputtering cross sections of hBN and MoS2_2.

Boron and nitrogen sputtering from hexagonal boron nitride

Electron beam irradiation has been shown to bore nanopores in monolayer hBN at beam energies far beneath the calculated ground state critical energy of ϵc∼80\epsilon_c \sim80 keV [8, 9] [33, 34]. In a pristine hBN layer, these beam-induced pores can be initialized from isolated boron and nitrogen vacancies. The atoms surrounding these vacancies have reduced coordination numbers, and thus, smaller displacement thresholds than those in the pristine material. Therefore, the atoms lining the defect are more likely to sputter than their surface counterparts for a given beam energy. As the edge atoms continue to sputter away at a high rate, the nanopore grows, eventually extending up to a few nm in diameter [9] [34, 35].

Cretu et al. measured the radial growth of these nanopores under electron irradiation for several temperatures ranging from 673 to 1473 K [9]. By dividing the radial growth rate by the beam current, they estimated the sputtering cross section to be around 25 barn under both 30 and 60 keV beams at temperatures of 1273 K and below (figure 5c). The cross sections were fairly temperature independent at these relatively low temperatures. Cretu et al. also found that the edges of these pores most often assume an armchair structure. Therefore, in an attempt to reproduce these measurements, we calculate the cross section of boron and nitrogen sputtering from an armchair edge at 1273 K (figure 5). To evaluate the displacement thresholds defined in equation (24) and listed in table 1, EvE_v is the free energy of the armchair edge supercell with a single boron or nitrogen vacancy, while EaE_a is that of an isolated boron or nitrogen atom. E0(0)E_0(0) and EgE_g are then the calculated free energy and band gap of the pristine armchair edge supercell. We plug the resulting set of displacement thresholds {Ed}\{E_d\} into equation (3) to calculate the sum of the boron and nitrogen sputtering cross sections. The excitation probabilities of hBN plotted in Figure 3a are then used for PiP_i. Strictly speaking, the calculation of PiP_i should consider the effects of the edge state orbitals. However, in our formalism, the beam electron is in a momentum eigenstate that is highly delocalized in real space. We therefore assume that the radius of the beam is much larger than that of the nanopore. This means that the majority of the beam-matter interactions occur in regions of pristine material, validating the use of the PiP_i calculated for pristine hBN. Future work should consider the effects of localized beam electron states to simulate beams with smaller focal points. Lastly, temperature effects on the cross section are considered in the manner described in our previous work [7].

Sputtering from the armchair edge of hBN

Figure 5. Sputtering from the armchair edge of hBN. The schematics in panels (a) and (b) depict the sputtering of nitrogen from out-of-plane and in-plane perspectives respectively. Panel (c) plots the sputtering cross sections of both boron and nitrogen in green and blue respectively. The black curves are the total sputtering cross section, i.e., the sum of their corresponding blue and green curves. The solid curves account for beam-induced excitations assuming an excitation lifetime of τ=240\tau= 240 fs (labelled “exc.” for excited). The dashed curves ignore the possibility of beam-induced excitation (labelled “gro.” for ground). The black squares are experimentally observed sputtering cross sections for hBN at 1273 K [9]. The consideration of beam-induced excitations reduces the disagreement between theory and experiment substantially.

Fitting our cross section curves to the data of Cretu et al. yields an excellent agreement if the predicted excitation lifetime is set to τ∼240\tau\sim240 fs. This predicted lifetime is much shorter than the reported excitation lifetime of ∼0.75\sim0.75 ns in pristine hBN [36], indicating that the sputtering process can significantly reduce the excitation lifetimes of hBN. We suspect that the atomic motion gives rise to non-radiative relaxation pathways that are not explicitly accounted for here. This motivates a closer investigation into the full electronic evolution of the hBN system post-collision. However, such a study is beyond the scope of this work. With that said, the electronic structure in the vicinity of a sputtering PKA differs significantly from that of the pristine room-temperature systems in which excitation lifetimes are experimentally measured. There is therefore no reason to expect the predicted lifetimes in this work to match those obtained by experiment.

We also see that the sputtering cross section, and thus the growth rate of the nanopore, is minimized for beam energies between 30 and 60 keV. Below these energies, the sputtering cross section begins to grow as the beam energy decreases. This is due to a non-negligible probability of final excitation numbers nf>3n_f > 3, for which Ed(nf)E_d(n_f) falls below Emin⁡E_{\min} (table 1). In these cases, the sputtering cross section is Ω∼5×108\Omega\sim5 \times10^{8} barn, seven orders of magnitude larger than the measured cross section at 30 keV. This suggests that one can expect the beam-induced nanopores in hBN to grow under beam energies as low as 1 keV. However, one must again be cautious of the predicted cross section at low beam energies in which the tree-level theory starts to break down. Nevertheless, figure 5c demonstrates a strong beam-energy dependence in the sputtering cross section across a wide range of experimentally relevant beam-energies. These findings could facilitate precise control of nanopore growth rates in hBN under electron irradiation.

Sulfur sputtering from molybdenum disulfide

Kretschmer et al. measured the sputtering cross section of sulfur from MoS2_2 for several beam energies ranging from 20 to 80 keV [12]. They found a peak in the cross section at 30 keV, much less than the predicted ground state ϵc∼90\epsilon_c \sim90 keV [12]. To help explain this unexpected peak, we calculate the cross section for sulfur sputtering from pristine MoS2_2 (figure 6). The vacant system free energy EvE_v for the calculation of {Ed}\{E_d\} is that of a MoS2_2 supercell with a single sulfur vacancy. EaE_a is then the free energy of an isolated sulfur atom, and E0(0)E_0(0) is that of the pristine MoS2_2 supercell. We then set EgE_g equal to the experimental band gap of 1.88 eV [28]. Using the resulting {Ed}\{E_d\}, we find that summing over the contributions of final excitation numbers nf>2n_f > 2 to the sputtering cross section produces a peak just below 30 keV, matching Kretschmer et al.’s findings remarkably well. In fitting to this peak, we predict an excitation lifetime of τ∼81\tau\sim81 fs. This is again much shorter than the excitation lifetime of pristine MoS2_2, which is on the order of a few picoseconds [37–39]. However, this is also much shorter than the fitted lifetime of hBN found in the previous subsection, consistent with the fact that the excitation lifetime of pristine hBN is much longer than that of MoS2_2.

The difference between the two materials’ lifetimes leads to markedly dissimilar cross section behaviors at low beam energies. Below beam energies of 30 keV, the cross section of MoS2_2 gradually drops to zero with decreasing ϵb\epsilon_b. In contrast, hBN’s total cross section has a minimum at around 40 keV and begins to increase as ϵb\epsilon_b decreases. Eventually, hBN’s sputtering cross section peaks before dropping quickly to zero as the beam energy goes to zero (figure S2). However, this peak occurs at around 0.5 keV, far below the lower energy bound of figure 5c. These distinct cross section behaviors can be explained by the amplified sensitivity of nfn_f to τ\tau at low beam energies. Equations 25 and 26 tell us that larger τ\tau makes large nfn_f more likely. At the same time, nfn_f cannot exceed nin_i. Thus, nfn_f is more sensitive to τ\tau at low beam energies for which nin_i is large. Accordingly, because τ\tau is greater in hBN than in MoS2_2, the expected values of nfn_f in hBN are much larger than those of MoS2_2 at low beam energies. This explains why the effect of considering excitations is much more pronounced in hBN than in MoS2_2. Furthermore, the difference in the cross section behavior is exacerbated by the fact that sulfur is heavier than both boron and nitrogen, so that its post-collision velocity is always smaller for a given energy transfer EE. It follows that tst_s is larger, and thus, PfP_f is smaller for a given EE in MoS2_2. This again increases the likelihood that hBN has more final excitations than MoS2_2, meaning that the effects of beam-induced excitation on hBN’s sputtering cross section are greater.

Schematic of sulfur sputtering and cross sections for MoS$_2$

Figure 6. Sputtering of sulfur from the outgoing MoS2_2 surface. The sputtering of molybdenum from MoS2_2 can be ignored because its displacement threshold is significantly larger than that of sulfur. The schematics in panels (a) and (b) depict the sputtering process from out-of-plane and in-plane perspectives respectively. Panel (c) plots the contributions of various numbers of final excitations nfn_f to the sputtering cross section assuming an excitation lifetime of τ=81\tau= 81 fs. The black dashed curve plots the predicted cross section ignoring the possibility of beam-induced excitation. The black squares are experimentally observed cross sections at 300 K [12]. The contribution from the sum of all nf>2n_f > 2 final excitations matches the experimental cross section remarkably well.

We also plot the contributions of nf=0n_f = 0, 1, and 2 excitations to MoS2_2’s sputtering cross section separately. In doing so, we see that the contributions of nf=1n_f = 1 or 2 would conceal the peak at 30 keV. Thus, it seems that the likelihoods of nf=1n_f = 1 or 2 are somehow suppressed. This suggests that the individual beam-induced excitations and subsequent relaxation are in some cases correlated. That is, the excitation and/or relaxation rates of a given electronic transition are affected by the coinciding distribution of electronic excitations. Thus, a proper treatment of these excitations and their effect on the sputtering cross section requires that the excitation probabilities of every possible transition are calculated for every possible excitation configuration. Such a nonlinear calculation is beyond the scope of this work, but should certainly be pursued in a future study. Nonetheless, the methods laid out here demonstrate that the consideration of beam-induced excitation can provide a quantitative justification for the sulfur sputtering rate to peak at a beam energy well-below the expected ground state critical energy.

Conclusion

In this paper, we developed a first-principles method to more accurately describe electron beam-induced sputtering cross sections in 2D insulating crystals by accounting for beam-induced electronic excitations and their subsequent relaxations. The method combines QED scattering theory with DFT electronic structure calculations to determine the likelihood of beam-induced excitation. The results show that the excitation probability is inversely proportional to both the material’s band gap and beam electron’s kinetic energy. We then show how these nonzero excitation probabilities increase the predicted sputtering cross sections of both hBN in MoS2\mathrm{MoS}_2. These cross sections can be made quantitatively similar to those obtained experimentally by treating the excitation lifetime as a fitting parameter. The methods laid out are computationally efficient, requiring only a few ground state electronic optimization calculations for each cross section curve. Thus, the formalism that we have developed can be easily applied to any 2D crystalline material to simulate the rates of atomic displacement under electron irradiation.

With that said, several questions naturally arise from our study. For example, why is the excitation lifetime reduced so drastically during sputtering? How might excitation and relaxation rates of different transitions be correlated? How might preexisting defects affect those rates? These questions urge follow up work to address the full breadth of physical processes involved in beam-matter interactions. Future studies should also consider additional electronic relaxation pathways to determine their effect on PfP_f. Correspondingly, other electronic responses such as ionization, core excitations, and second order electronic excitation effects such as Auger scattering can be incorporated into the calculation of PiP_i [38,40–42]. Moreover, spin polarization effects can also be examined by making M\mathcal{M} spin-dependent, i.e., not averaging over spins as is done in equation (4). The beam electron path can also change significantly after collision with the nucleus. Ensuing research should investigate how these altered trajectories generate new excitation probabilities PiP_i, which must certainly play a role in 3D bulk materials. Furthermore, the methods here can be modified to accommodate more advanced DFT techniques. For example, the calculation of PiP_i should be made compatible with ultrasoft and projector augmented wave pseudopotentials [43] in a manner similar to that employed in modern GW codes [44–46]. In addition, the plane-wave coefficients of excited electronic states can be calculated self-consistently using constrained DFT [47]. Perhaps most importantly, progress in this field requires much more experimental data. We therefore hope this paper encourages new experimental investigation into beam-induced sputtering for beam energies below the predicted ground state critical energy.

Clearly, the combined QED-DFT approach to modeling beam-induced excitations and their effect on the sputtering cross section opens up a rich and diverse field of physics for both theoretical and experimental exploration. We hope that this work and the work it may stimulate can eventually enable the use of electron beams for precise atomic-scale engineering of crystalline materials.

Methods

DFT [13,14] was used to determine all electronic and ionic structures from first-principles. Free energy calculations were carried out with the Vienna ab initio Simulation Package (VASP) [15, 48] implementing the projector augmented wave (PAW) method [43] along with the Perdew-Burke-Ernserhof (PBE) generalized gradient approximation (GGA) to the exchange correlation functional [49]. Van der Waals interactions were accounted for using the optB88-vdW density functional methods [50, 51]. All parameters were converged so that any increase in precision would change the total free energy by less than 1 meV per atom. The cutoff energies for hBN and MoS2_2 were set to 800 and 550 eV respectively, while the BZs of both materials’ pristine unit cells were sampled with a Γ\Gamma-centered 6×6×16 \times6 \times1 Monkhorst-Pack meshes [52], corresponding to 0.417 and 0.328 and k-points per inverse A˚\mathring{\mathrm{A}} for hBN and MoS2_2 respectively. To achieve the same k-point densities, surface vacancies in MoS2_2 and edge vacancies in hBN were placed in respective 6×6×16 \times6 \times1 and 4×1×14 \times1 \times1 supercells whose BZs were sampled with a single k-point on Γ\Gamma. The heights of the hBN and MoS2_2 cells were 12 A˚\mathring{\mathrm{A}} and 20 A˚\mathring{\mathrm{A}} respectively to provide sufficient separation from periodic images. Nanoribbon structures were used to simulate isolated armchair edges in hBN. These ribbons were more than 16 A˚\mathring{\mathrm{A}} across and placed in cells 28 A˚\mathring{\mathrm{A}} wide to avoid interactions between opposing edges and periodic images. Lastly, relaxation iterations of ionic positions and lattice constants persisted until the all Hellmann-Feynman forces settled below 1 meV/A˚\mathring{\mathrm{A}}.

Because equation (16) relies on the orthogonality of the Kohn-Sham orbitals, the optimized norm-conserving Vanderbilt pseudopotentials [53, 54] implemented in Quantum ESPRESSO [55] were used to determine all plane-wave coefficients CG+knC_{\mathbf{G}+\mathbf{k}}^{n}. The sum of all transition probabilities SS defined in equation (18) was used to gage the convergence of all parameters, which were deemed converged when any increase in precision changed SS by less than 5% (Figure S4). The parameter values that satisfy this criteria differed substantially from those needed for free energy convergence. The cutoff energies for hBN and MoS2_2 were set to 299 and 286 eV respectively, while their cell-heights were respectively set to 18 and 12 A˚\mathring{\mathrm{A}}. For both materials, the maximum virtual photon momentum ∣qmax⁡∣\lvert\mathbf{q}_{\max}\rvert required for convergence fell well within their first BZs. We therefore chose ∣qmax⁡∣\lvert\mathbf{q}_{\max}\rvert to be the magnitude of high-symmetry point KK in each respective BZ. The maximum number of initial excitations considered for hBN and MoS2_2 were nimax⁡=5n_i^{\max}=5 and 9 respectively (Figure S1). Finally, convergence of SS requires extreme care in density-dependent sampling of the BZ. This necessitates fitting a curve to SS calculated for various k-point mesh densities and extrapolating to an infinitely fine mesh to estimate the converged value of SS (§ 16). The most dense k-point meshes used to fit these curves had dimensions of 45×45×145 \times45 \times1 and 36×36×136 \times36 \times1 for hBN and MoS2_2 respectively.

Six convergence plots for hBN and MoS$_2$

Figure S4. Convergence of SS with respect to the (a and d) maximum magnitude of virtual photon momentum considered, (b and e) plane-wave DFT cutoff energy, and (c and f) height of the unit cell. A 6×6×16 \times6 \times1 k-point mesh and a beam energy of 60 keV is used to generate all six plots. The converged parameters were found to be insensitive to changes in k-point density and beam energy. SS is deemed converged when any increase in precision changes SS by less than 5%.

Acknowledgements

Calculations were performed at the Center for Computational Innovations at Rensselaer Polytechnic Institute, Livermore Computing Center at Lawrence Livermore National Laboratory (LLNL), and Compute and Data Environment for Science at Oak Ridge National Laboratory (ORNL), which is supported by the Office of Science of the U.S. Department of Energy (DOE) under Contract No. DE-AC05-00OR22725. This work was performed under the auspices of the U.S. Department of Energy by LLNL under Contract DE-AC52-07NA27344. This material is also based upon work supported by the U.S. DOE, Office of Science, Office of Workforce Development for Teachers and Scientists, Office of Science Graduate Student Research (SCGSR) program. The SCGSR program is administered by the Oak Ridge Institute for Science and Education for the DOE under contract number DE-SC0014664. Funding was provided by the National Science Foundation (Award 1608171). Work (BGS, PG, DL, JJ) was also supported by ORNL’s Center for Nanophase Materials Sciences, a U.S. DOE Office of Science User Facility.

Supplemental Material: QED theory of electron beam-induced electronic excitation and its effect on sputtering cross sections in 2D crystals

Anthony Yoshimura; Michael Lamparski; Joel Giedt; David Lingerfelt; Jacek Jakowski; Panchapakesan Ganesh; Tao Yu; Bobby Sumpter; Vincent Meunier

Invariant matrix element M\mathcal{M}

As the excitation amplitude in equation (16) contains the invariant matrix element M\mathcal{M} and not ∣M∣2\lvert\mathcal{M}\rvert^2, we are unable to use the spin sum identities typically used to derive many QED cross sections [1, 2]. Nonetheless, the evaluation of M\mathcal{M} in equation (4) is straightforward, though a bit cumbersome. In the Dirac basis [3],

γ0=(I00−I)andγi=(0σi−σi0),(31)\gamma^0 = \begin{pmatrix} I & 0 \\ 0 & -I \end{pmatrix} \quad\text{and}\quad \gamma^i = \begin{pmatrix} 0 & \sigma^i \\ -\sigma^i & 0 \end{pmatrix}, \tag*{(31)}

where

I=(1001)andσ⃗={(0110),(0−ii0),(100−1)}.(32)I = \begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix} \quad\text{and}\quad \vec{\sigma} = \left\{ \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}, \begin{pmatrix} 0 & -i \\ i & 0 \end{pmatrix}, \begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix} \right\}. \tag*{(32)}

The electron spinors in equation (4) can be written as

u1(p)=ϵ+m(10pzϵ+mpx+ipyϵ+m)andu2(p)=ϵ+m(01px−ipyϵ+m−pzϵ+m),(33)u^1(p) = \sqrt{\epsilon+ m} \begin{pmatrix} 1 \\ 0 \\ \frac{p^z}{\epsilon+ m} \\ \frac{p^x + i p^y}{\epsilon+ m} \end{pmatrix} \quad\text{and}\quad u^2(p) = \sqrt{\epsilon+ m} \begin{pmatrix} 0 \\ 1 \\ \frac{p^x - i p^y}{\epsilon+ m} \\ \frac{-p^z}{\epsilon+ m} \end{pmatrix}, \tag*{(33)}

while

uˉs(p)=us†(p)γ0.(34)\bar{u}^{s}(p)=u^{s\dagger}(p)\gamma^{0}. \tag*{(34)}

As justified in section 3.1, we need only evaluate the t-channel contribution to M\mathcal{M} and multiply the result by 2. That is,

M(p4p3←p2p1)∼e2∑s1,s2,s3,s4uˉs4(p4)γμus1(p1)(1p3−p2)2uˉs3(p3)γμus2(p2).(35)\mathcal{M}(p_4p_3\leftarrow p_2p_1)\sim e^2\sum_{s_1,s_2,s_3,s_4}\bar{u}^{s_4}(p_4)\gamma^\mu u^{s_1}(p_1)\left(\frac{1}{p_3-p_2}\right)^2\bar{u}^{s_3}(p_3)\gamma_\mu u^{s_2}(p_2). \tag*{(35)}

Substituting equations (31) through (S4) into (S5) yields M\mathcal{M} in terms of the components of the electrons’ 4-momenta. That is,

M(p4p3←p2p1)∼−2e2(p3−p2)2[(ϵ1+m)(ϵ2+m)(ϵ3+m)(ϵ4+m)]−1/2×{(ϵ1+m)p3x[(ϵ2+m)p3x+(ϵ3+m)p2x]+2(ϵ1+m)(ϵ3+m)(p2y+ip2z)(p4y−ip4z)+2(ϵ2+m)(ϵ4+m)ip1y(p3y−ip3z)−[(ϵ2+m)(ϵ3+m)+p2xp3x+(p2y+ip2z)(p3y−ip3z)]×[(ϵ1+m)(ϵ4+m)+ip1z(p4y−ip4z)]}.(36)\begin{aligned} \mathcal{M}(p_4p_3\leftarrow p_2p_1)\sim{}&-\frac{2e^2}{(p_3-p_2)^2}\left[(\epsilon_1+m)(\epsilon_2+m)(\epsilon_3+m)(\epsilon_4+m)\right]^{-1/2}\\ &\times\Bigl\{(\epsilon_1+m)p_3^x\left[(\epsilon_2+m)p_3^x+(\epsilon_3+m)p_2^x\right]\\ &\quad+2(\epsilon_1+m)(\epsilon_3+m)(p_2^y+ip_2^z)(p_4^y-ip_4^z)\\ &\quad+2(\epsilon_2+m)(\epsilon_4+m)ip_1^y(p_3^y-ip_3^z)\\ &\quad-\left[(\epsilon_2+m)(\epsilon_3+m)+p_2^xp_3^x+(p_2^y+ip_2^z)(p_3^y-ip_3^z)\right]\\ &\qquad\times\left[(\epsilon_1+m)(\epsilon_4+m)+ip_1^z(p_4^y-ip_4^z)\right]\Bigr\}. \tag*{(36)} \end{aligned}

where we let p1p_1 denote the momentum of the beam electron so that p1x=p1y=0p_1^x=p_1^y=0. See the Mathematica [4] notebook in the supplemental material for more details.

Normalization

When integrating over 4-momentum space as is done in § 3.2, Lorentz invariance constrains a particle’s 4-momentum to obey p2=m2p^2 = m^2. It follows that the 4-momentum integration measure d4pd^4p is always multiplied by a delta function δ(p2−m2)\delta(p^2 - m^2), i.e.,

d4p δ(p2−m2)θ(p0)=d3p2p0δ(p0−ϵp),(37)d^4p\,\delta(p^2-m^2)\theta(p^0)=\frac{d^3p}{2p^0}\delta(p^0-\epsilon_p), \tag*{(37)}

where the Heaviside step function restricts our consideration to particles of positive mass (we note that antiparticles are interpreted as positive mass particles that propagate backwards in time). With this integration measure, the identity operator can be written as

I^=∫d4p(2π)4(2π)δ(p2−m2)θ(p0)∣p⟩⟨p∣=∫d3p(2π)32ϵp∣p⟩⟨p∣=∫d3p(2π)3∣p⟩⟨p∣.(38)\begin{aligned} \hat{I} &= \int\frac{d^4p}{(2\pi)^4}(2\pi)\delta(p^2-m^2)\theta(p^0)\lvert p\rangle\langle p\rvert\\ &= \int\frac{d^3p}{(2\pi)^3 2\epsilon_p}\lvert p\rangle\langle p\rvert\\ &= \int\frac{d^3p}{(2\pi)^3}\lvert p\rangle\langle p\rvert. \tag*{(38)} \end{aligned}

The last equality implies that

∣p⟩=(2ϵp)1/2∣p⟩.(39)\lvert p\rangle=(2\epsilon_p)^{1/2}\lvert p\rangle. \tag*{(39)}

Lab frame p3zp_3^z

For the scattering of two free electrons, conservation of 4-momentum allows us to write

p1+p2=p3+p4.(40)p_1+p_2=p_3+p_4. \tag*{(40)}

This constrains four of the six components needed to specify the 3-momenta of the two outgoing particles, so that only two components are independent. We choose the independent components to be p3xp_3^x and p3yp_3^y. Thus, we wish to find p3zp_3^z as a function of p1p_1, p2p_2, p3xp_3^x, and p3yp_3^y. From this, p4=p1+p2−p3p_4=p_1+p_2-p_3 is easily obtained, and we have all six components of outgoing 3-momenta needed to calculate the scattering amplitude from equation (16). We start by setting the zz-direction parallel to p1\mathbf{p}_1. This means that

p4x=p2x−p3xp4y=p2y−p3yp4z=p1z+p2z−p3zϵ4=ϵ1+Δϵ(41)\begin{aligned} p_4^x&=p_2^x-p_3^x \\ p_4^y&=p_2^y-p_3^y \\ p_4^z&=p_1^z+p_2^z-p_3^z \\ \epsilon_4&=\epsilon_1+\Delta\epsilon \tag*{(41)} \end{aligned}

where

ϵi=ϵpi=pi2+m2,Δϵ=ϵnk−ϵn′k′,(42)\epsilon_i=\epsilon_{p_i}=\sqrt{p_i^2+m^2}, \qquad\Delta\epsilon=\epsilon_{n k}-\epsilon_{n'k'}, \tag*{(42)}

and ϵnk\epsilon_{n k} and ϵn′k′\epsilon_{n'k'} are the eigenvalues of the excited hole and electron states, respectively. We can also write ϵ1=γm\epsilon_1=\gamma m where γ=1/1−β2\gamma=1/\sqrt{1-\beta^2} is the Lorentz factor corresponding to the beam electron’s velocity β\beta. Squaring the last line in (S11) and subtracting m2m^2 then gives

p42=p12+2γmΔϵ+O(Δϵ2).(43)p_4^2=p_1^2+2\gamma m\Delta\epsilon+O(\Delta\epsilon^2). \tag*{(43)}

We can ignore terms of order O(Δϵ2)O(\Delta\epsilon^2) since Δϵ≪m\Delta\epsilon\ll m. Meanwhile, squaring and summing the first three equations in (S11) tells us that

p42=p12+p22+p32+2[p1zp2z−p2⊥⋅p3⊥−(p1z+p2z)p3z],(44)p_4^2=p_1^2+p_2^2+p_3^2+2\left[p_1^z p_2^z-\mathbf{p}_2^\perp\cdot\mathbf{p}_3^\perp-\left(p_1^z+p_2^z\right)p_3^z\right], \tag*{(44)}

where the ⊥\perp superscript denotes the projection perpendicular to z^\hat{z}. Subtracting equation (44) from equation (43) then yields

0∼p22+p32+2[p1zp2z−p2⊥⋅p3⊥−(p1z+p2z)p3z]−2γmΔϵ.(45)0\sim p_2^2+p_3^2+2\left[p_1^z p_2^z-\mathbf{p}_2^\perp\cdot\mathbf{p}_3^\perp-\left(p_1^z+p_2^z\right)p_3^z\right]-2\gamma m\Delta\epsilon. \tag*{(45)}

Asymptotic formula (S15) can then be solved for p3zp_3^z, so that

p3z∼p1z+p2z±(p1z)2−(p2⊥−p3⊥)2+2γmΔϵ.(46)p_3^z\sim p_1^z+p_2^z\pm\sqrt{(p_1^z)^2-\left(\mathbf{p}_2^\perp-\mathbf{p}_3^\perp\right)^2+2\gamma m\Delta\epsilon}. \tag*{(46)}

We choose the −- from ±\pm as we impose that p3\mathbf{p}_3 is a component of the outgoing crystal electron state, whose zz-momentum is much smaller than that of the beam electron.

Converging nimax⁡n_i^{\max}

Peaks in the sputtering cross section of hBN

Sputtering cross sections of boron and nitrogen from the hBN armchair edge

Figure S2. The sputtering cross sections of boron and nitrogen from the hBN armchair edge peak at beam energies much lower than those typically used for microscopy and defect engineering. However, as mentioned in the main text, the validity of our perturbative approximation to the scattering operator breaks down at low beam energies (§ 4). Thus, the values and positions of these peaks may change if higher order perturbation terms are considered.

Fitting and converging SS

For a given beam energy ϵb\epsilon_b, we calculate the sum of all transition probabilities SS defined in equation (18) for various k-point densities where the number of k-points in the Brillouin zone is NkN_k. We then fit the points to a curve of the form

S(Nk,ϵb)=aNk−be−cNk+S∞(47)S(N_k,\epsilon_b) = \frac{a}{N_k-b}e^{-cN_k} + S_\infty \tag*{(47)}

where aa, bb, cc, and S∞S_\infty are all fitting parameters that depend on ϵb\epsilon_b (Figure S3). We repeat this for multiple values of ϵb\epsilon_b ranging from 5 to 100 keV and record the best fit S∞S_\infty for each ϵb\epsilon_b. Based on the work of Bethe [5, 6, 7], we fit the resulting values of S∞(ϵb)S_\infty(\epsilon_b) to an inverse function,

Plots showing the dependence of $S$ on the number of k-points and beam energy

Figure S3. Finding the large crystal limit of S(ϵb)S(\epsilon_b) requires extrapolation. Panel (a) shows the dependence of SS on the number k-points NkN_k in the Brillouin zone of hBN under 40 keV irradiation. The simulated points are fitted well by equation (S17). The green dashed line denotes S∞S_\infty, the asymptotic limit of S(Nk)S(N_k) for large NkN_k. Panel (b) then shows the dependence of S∞S_\infty on the beam energy ϵb\epsilon_b, which is fitted well by equation (S18). The green arrow in panel (b) illustrates how S∞(ϵb=40 keV)S_\infty(\epsilon_b = 40\ \mathrm{keV}) is determined by the fit in panel (a).

S∞(ϵb)=Aϵb−B+C.(48)S_\infty(\epsilon_b) = \frac{A}{\epsilon_b-B} + C. \tag*{(48)}

where AA, BB, and CC are fitting parameters. The parameters for hBN and MoS2_2 are given in Table S1. The fitted curve can then be substituted for S(ϵb)S(\epsilon_b) in formula (23) to obtain Pi(ϵb,ni)P_i(\epsilon_b,n_i), the probability of exciting nin_i electrons in the large crystal limit.

materialAA (eV)BB (eV)CC
hBN7.655-0.7700.06526
MoS2_249.05-3.8670.3547

Table S1. Fitting parameters of equation (S18) for hBN and MoS2_2.

Finally, SS must also be converged with respect all parameters. These include the maximum virtual photon momentum, DFT cutoff energy, and height of the pristine unit cell (Figure S4). SS is considered converged with respect to a parameter when any increase in the parameter’s precision changes SS by less than 5%.

Calculating Emin⁡E_{\min}

Minimum energy transfer from the beam electron to the target nucleus for MoS$_2$ and hBN

Figure S5. The minimum energy transfer from the beam electron to the target nucleus peaks at very low beam energies and is always much smaller than the nuclei’s average thermal kinetic energy of ∼39\sim39 meV at room temperature.

In hBN, a unit cell contains a boron and nitrogen atom. As these atoms have similar masses and displacement thresholds (table 1), the sputtering of both atoms should be considered for a given beam energy. Thus, we approximate the maximum cross sectional area σmax⁡\sigma_{\max} of these atoms to be ΩhBN/2\Omega_{\mathrm{hBN}}/2, where ΩhBN\Omega_{\mathrm{hBN}} is the area of the hBN unit cell. On the other hand, molybdenum is much heavier than sulfur and has a much larger displacement threshold in MoS2_2 [8]. This means that only the sputtering of sulfur needs to be considered. Additionally, of the two sulfur atoms in the MoS2_2 unit cell, only the atom on the outgoing surface is eligible to sputter from a pristine system [8]. Therefore, σmax⁡\sigma_{\max} for sulfur sputtering from MoS2_2 is ΩMoS2\Omega_{\mathrm{MoS}_2}, the area of the MoS2_2 unit cell.

To approximate Emin⁡E_{\min}, we use the Rutherford displacement cross section [9, 10, 11] as an approximation to equation (2),

σR=π(Zα∣p∣β)2(Emax⁡Ed−1).(49)\sigma_R = \pi\left(\frac{Z\alpha}{\lvert p\rvert\beta}\right)^2 \left(\frac{E_{\max}}{E_d}-1\right). \tag*{(49)}

Setting σR=σmax⁡\sigma_R=\sigma_{\max} and Ed=Emin⁡E_d=E_{\min} and solving for Emin⁡E_{\min} yields

Emin⁡(ϵb)=Emax⁡[Ωπ(∣pb∣βZα)2+1]−1.(50)E_{\min}(\epsilon_b)=E_{\max}\left[\frac{\Omega}{\pi}\left(\frac{\lvert p_b\rvert\beta}{Z\alpha}\right)^2+1\right]^{-1}. \tag*{(50)}

Using the Rutherford cross section instead of the McKinley-Feshbach cross section should be accurate for small beam energies for which β≪1\beta\ll1. This makes it well-suited for finding Emin⁡E_{\min}, since Ed(nf)<Emin⁡E_d(n_f)<E_{\min} only for large nfn_f, and large nin_i (and thus nfn_f) are much more likely for small beam energies (Figure 3).

With that said, the sputtering cross section is fairly insensitive to the exact value of Emin⁡E_{\min} when Emin⁡≪Emax⁡E_{\min} \ll E_{\max}. This is because the post-collision velocity of the PKA diminishes when the energy transfer EE shrinks. The beam-induced excitations therefore have more time to relax before the PKA reaches the step in the energy surface (§ 6). This makes PfP_f small for small EE, so that contributions to the integral in equation (3) are extremely tiny for EE near Emin⁡E_{\min}. As a result, changes in Emin⁡E_{\min} are essentially immeasurable for beam energies greater than 1 keV. Nonetheless, the use of Emin⁡E_{\min} is necessary for the calculation of a finite sputtering cross section.

References

  1. [1]M. E. Peskin and D. V. Schroeder, An Introduction to Quantum Field Theory. Reading, MA: Perseus Books Publishing, L.L.C., first ed., 1995.
  2. [2]T. Lancaster and S. J. Blundell, Quantum Field Theory for the Gifted Amateur. New York, NY, United States of America: Oxford University Press, first ed., 2014.
  3. [3]J. D. Bjorken and S. D. Drell, Relativistic Quantum Mechanics. International series in pure and applied physics, McGraw-Hill, 1964.
  4. [4]W. R. Inc., “Mathematica, Version 12.3.”
  5. [5]H. Bethe, “Zur Theorie des Durchgangs schneller Korpuskularstrahlen durch Materie,” Ann. Phys., vol. 397, no. 3, pp. 325–400, 1930.DOI
  6. [6]T. Susi, J. C. Meyer, and J. Kotakoski, “Quantifying transmission electron microscopy irradiation effects using two-dimensional materials,” Nat. Rev. Phys., vol. 1, pp. 397–405, may 2019.DOI
  7. [7]S. Kretschmer, T. Lehnert, U. Kaiser, A. V. Krasheninnikov, and A. V. Krasheninnikov, “Formation of defects in two-dimensional MoS₂ in transmission electron microscope at electron energies below the knock-on threshold : the role of electronic excitations,” Nano Lett., vol. 20, pp. 2865–2870, mar 2020.DOI
  8. [8]H. P. Komsa, J. Kotakoski, S. Kurasch, O. Lehtinen, U. Kaiser, and A. V. Krasheninnikov, “Two-dimensional transition metal dichalcogenides under electron irradiation: Defect production and doping,” Phys. Rev. Lett., vol. 109, pp. 035503–1–5, jul 2012.DOI
  9. [9]S. T. Thornton and J. B. Marion, Classical Dynamics of Particles and Systems. Boston, MA: Brooks/Cole, 5th ed., 2004.
  10. [10]J. J. Sakurai and J. Napolitano, Modern Quantum Mechanics. San Francisco, CA: Pearson Education, 2nd ed., 2011.
  11. [11]A. Yoshimura, M. Lamparski, N. Kharche, and V. Meunier, “First-principles simulation of local response in transition metal dichalcogenides under electron irradiation,” Nanoscale, vol. 10, pp. 2388–2397, jan 2018.DOI

Paper details

vlnc:W_566tbc34
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

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