A kinetic model for anisotropic reactions in amorphous solids Wei Hong* Department of Aerospace Engineering, Iowa State University, Ames, IA 50011, USA ABSTRACT When mechanical constraints are present, solid-state reactions often induce deviatoric inelastic strains in addition to volume change. Existing models either attribute such deformation to the plastic flow driven by the stress exceeding a non-measurable kinetic-dependent yield strength, or need to introduce a deviatoric-stress-dependent chemical potential. By employing the transformation strain to characterize the state of reaction, this letter formulates a kinetic model via averaging the reaction rate at all possible orientations. The model is illustrated through the constrained lithiation-delithiation process of silicon as an example. With just one fitting parameter, the model can quantitatively capture the experimental results. The model only hypothesizes linear kinetics, and does not need to introduce kinetic-dependent plasticity or modify basic thermodynamic quantities. This approach can also be applied to other material systems, as well as extended to the nonlinear kinetics of far-from-equilibrium reactions. 1 Solid-state reactions are seldom regarded a new topic in material synthesis, but when it comes to the strains or stresses induced by reactions, little has been explored beyond the similarities with liquid- or gas-state reactions. With the acceptance of another reactant through diffusion or other means of transportation, a solid reactant may expand in volume and/or change in shape. The common approach of modeling such processes is to assign a stress- and kinetics-independent transformation strain to the resultant.1-4 The actual state of deformation is then calculated based on linear elasticity (or elasto-plasticity) as if the stress was applied after the reaction. Such a decoupled approach may be applicable to a reaction of which a transformation strain is clearly defined, e.g. one with a background crystalline reactant that retains its coherency throughout the process, as sketched in Figure 1a. However, this approach may be problematic if either the reactant or the resultant is amorphous, and thus the transformation strain or its orientation cannot be uniquely determined by the state of reaction. As illustrated by Figure 1b, without significantly rearranging the spatial distribution of the reactant atoms, the resultants can bear very different transformation strains, depending on the relative positions of the inserted atoms. Variants of the resultants as those sketched in Figure 1b are chemically identical, but mechanically different if the material is stressed or constrained. To model the deformation involved in such reactions, it has been assumed that the transformation strain of an amorphous reaction is purely volumetric, and a separate process of plastic flow generates deviatoric inelastic strain, when the equivalent stress exceeds a threshold – the yield strength. However, the yield strength needs to be taken as composition and kinetics dependent.3,4 Such an assumption is perhaps originated from the description of liquid reactions, in which the state of reaction can be fully described by a scalar variable. For solid reactants, the assumption is less convincing. The plentiful observations on reactions which transform 2 crystalline solids into amorphous resultants5-11, and complete within atomically sharp phase boundaries12, may serve as counter evidences of this assumption: the amorphization would require plastic flow that is non-affine at atomic level which defies the applicability of continuum notions in the first place. It is natural to believe that such deformation takes place right at the insertion of the second reactant, instead of after the reaction. a b Figure 1. Schematics of reaction-induced deformation. (a) For a coherent crystalline-to- crystalline reaction, the transformation strain is well-defined. (b) For an amorphous-to- amorphous reaction, the deviatoric strain is arbitrary from the chemical consideration only – with the same relative positions and average atomic distances, the resultants may differ by a deviatoric strain. 3 The ambiguity could be clarified if one allows additional variables for the state of reaction. To differentiate between resultant variants like those sketched in Figure 1b, one needs a state variable which contains the orientation information. While the choice is more or less arbitrary, in this letter we select the transformation strain tensor ε t as the state variable. The rate of the reaction, represented by the time derivative of the transformation strain, ε t , is a function of the electrochemical driving force for the reaction and the stress tensor σ . To simplify representation, we further decompose ε t into the volumetric and deviatoric parts: ijt v 3 ij ~ijt . (1) Although the isotropy of the material requires the volumetric strain rate to be dependent only on the electrochemical driving force and the hydrostatic stress, the deviatoric strain rate ~ε t could be ~ . In a simple case when the dependence is linear, dependent on the deviatoric stress σ ~ijt Tijkl~kl , (2) where T is a fourth rank tensor relating the two deviatoric tensors, and the repeated indices indicate a summation. The isotropy of the amorphous material requires T to contain only one scalar parameter , and the transformation strain rate to be parallel to the stress deviator: ~ijt ~ij . (3) Although the kinetic relation (3) shares the same form as that of a Newtonian fluid, here is not a material constant, and could be dependent on the concentration of the mobile reactant or the energetic driving force. Instead of driven by shear stresses as a viscous fluid, the transformation strain should be regarded as a part of the reaction process, and is nonzero only when the reaction is taking place. It should be noted that relations in the form of Eq. (3) also appear in most existing theories of related phenomena,1,3,4,13 but the difference lies in the 4 coefficient and the underlying physical interpretation. For example, the similar expression in the reactive flow theory is interpreted as the plastic flow driven by a stress exceeding the yield strength.1,2 In some other models, the reactant chemical potential needs to be modified to include a contribution from deviatoric stress,13,14 and thus the scalar reaction rate (or volumetric strain rate) is also affected. In contrast to the existing theories, the current model makes no hypothesis other than linear kinetics and linear elasticity, as detailed in the following discussion. At the atomistic scale, the resultants of solid-state reactions are seldom isotropic. In the extreme case when two atoms meet and react, the resultant group is non-spherical and always takes an orientation. For amorphous solids, atomistic studies also show that it is energetically more favorable for a group of few resultant atoms to have an orientation than to be spherical or isotropic.15 In the absence of a directional field (e.g. electric field or stress), the orientations of different atomic groups are randomly taken and thus the average properties often exhibit macroscopic isotropy. Here we will focus on the case when non-negligible deviatoric stresses are present, so that the orientation distribution could be affected. We will use the Li-Si system as a timely example, but the methodology is generally applicable to almost all solid-state systems with amorphous reactants or resultants. For ease of description, we imagine a group of very few Si atoms and introduce an elemental deformation gradient tensor to represent the deformation caused by reaction 1t 0 F t 0 t2 0 0 0 0, t2 (4) where 1t and t2 are the principal stretches of transformation, functions of the nominal Li concentration C . The components of F t are written in the local coordinate system which aligns with the principal directions, and it is also assumed that the transformation of the Li-Si group has 5 transverse isotropy, which represents the case of 1:1 Li-Si atomic ratio as suggested by atomistic simulations15. The transformation F t can take general anisotropy and the resulting macroscopic rate equation still takes the form as in Eq. (3), but the coefficient will depend on more parameters. On the other hand, the available experimental results are mostly unidirectional and insufficient to distinguish these anisotropies. The transverse anisotropy is thus taken as a timely approximate. In the context of finite deformation, the local inelastic strain rate due to lithiation is given by 1t t 1 1 ε l F t F t 0 0 0 t2 t2 0 d1t 0 t 1dC 0 C 0 t2 0 t2 0 dt2 t2 dC 0 0 0 . dt2 t2dC (5) In general, the derivatives d1t dC and dt2 dC are dependent on the composition as well as the direction of the reaction (e.g. lithiation or delithiation), and may be identified through atomistic simulations.15,16 Here for simplicity, we take an approximation by assuming both d1t 1t dC and dt2 t2dC to be material constants and write them in terms of the volumetric contribution and deviatoric contribution : dt2 d1t . and t t 2dC 3 2 1dC 3 (6) The free-energy landscape of the reaction is illustrated schematically by Figure 2, whereas the actual reaction space should be three dimensional. In the absence of deviatoric stress or any other directional driving force, the energy profile is isotropic, as sketched in Figure 2a. When ~ , the free energy of the resultant is orientation the material is under a non-zero deviatoric stress σ dependent, as sketched in Figure 2b. Here we assume the driving forces to be much lower than 6 thermal fluctuation, and employ linear kinetics for the rate of reaction. Locally for a small atomic group, the transformation strain within unit time is given by Eq. (5), and its orientation by the azimuth and zenith angles , of the local x1 -axis in the global coordinate system. The linear kinetics assumption dictates the rate of reaction and thus the probability density of transformation in orientation , to be proportional to its electro-chemo-mechanical driving force, Pr , C det F t σl , : ε l , in which is the electrochemical driving force in the absence of stress, and σ l is the stress tensor σ transformed into the local coordinates through the rotation matrix Q , , σl Q σ QT . It should be pointed out that the driving force is for reaction only,3,4 which differs from the electrochemical potential for the migration of reactants. The factor det F t accounts for the finite volume change during the reaction, as the energy terms are written in the reference configuration, i.e. with respect to the number of background Si atoms. After normalization, the probability density could be written explicitly as a b B2 B A A B1 ~0 σ ~0 σ Figure 2. Sketches of the energy landscape of a solid-state reaction. a) In a hydrostatic stress state, the energy profile is isotropic and the material undergoes a pure volumetric transformation. 7 ~ , the free energy is lowered for anisotropic reactants b) Under non-negligible deviatoric stress σ at particular orientations, due to the contribution from elastic strain energy. ijl l 1 v ij C 1 v 1 C , Pr , 4C 1 v ijl ijl sindd 4 1 v kk 3 l l ij ij (7) where v det F t 1 is the volumetric strain due to reaction. With the aid of the probability distribution (7), we can calculate the expectation of the macroscopic inelastic strain rate due to reaction l ijt Qkikm Qmj Pr , sindd . (8) Substituting Eq. (7) into (8) and carrying out the integrals over the surface of a unit sphere, we arrive at ~ 2 1 3 v ij ijt C ij , 10 1 3 v nn 3 (9) As expected, the statistical analysis of the reactions in different orientations recovers the form from symmetry consideration, Eqs. (1) and (3), in which the unknown coefficient is now given explicitly by 3 2 1 v C , 10 (10) where 1 v kk 3 is the total (electro-chemo-mechanical) driving force for the reaction. The total driving force is related to, but should not be confused with the electrochemical potential for the diffusion of reactants. It could be easier to understand it by imagining a virtual source that directly injects the reactants to the place of reaction without any 8 transportation process.3,4 Different from a Newtonian flow, the deviatoric strain is only activated in a reaction process, when C 0 . On the other hand, the volumetric strain rate v C or the commonly invoked isotropic reaction rate C is independent of the deviatoric stress state. As a kinetic equation, Eq. (9) should be regarded as an extension to the commonly used linear kinetic relation of isotropic reactions. Unlike existing models which result in similar mathematical forms from different asumptions,13 Eq. (9) is only a consequence of linear kinetics and no modification on any thermodynamic quantities (e.g. the chemical potential) is needed. In the extreme case of an equilibrium or quasi-static state, the model would have no effect over that governs by thermodynamics. To complete the mathematical formulation, a stress-strain relation of linear elasticity (or elasto-plasticity) is needed in addition to Eq. (9). In the case when is not explicitly rate dependent, Eq. (9) can be reduced to a rateindependent form d ijt dC 3 ij 3 2 1 v ~ij . 10 (11) Furthermore, if the kinetics of the system is limited by a constant reaction rate (e.g. during galvanostatic lithiation of silicon), the total driving force may be regarded as constant. As neither nor could be directly measured, it is more convenient to leave the ratio 2 as a single fitting parameter to be determined by comparing to experiments. As a simple example to illustrate the model and compare with existing experiments, we calculate the lithiation process of a Si film constrained laterally by the underlying rigid substrate. For simplicity, we will start with the thin-film limit, in which both the deformation and the Li concentration are homogeneous throughout the film. The principle values of inelastic strain are the through thickness components 1t and the lateral component 2t . Here for simplicity, we 9 neglect plasticity or any inelastic deformation other than that induced by reaction. Due to the lateral constraint of the rigid substrate, the lateral elastic strain 2e 2t . Therefore, for relatively small 2t , the lateral stress component E 2t , 2 1 (12) where E is Young’s modulus and Poisson’s ratio. Combining Eqs. (11) and (12), we arrive at the differential equation for 2 : d 2 E dC 1 1 v 2 2 . 10 3 (13) In the case of small volumetric strain, v 1 , this simple dynamic system has an attractor at 2 103 2 . Depending on the sign of , or whether the system is under a lithiation or delithiation process, 2 takes a negative or positive value, respectively, corresponding to the plateau stresses observed in experiments. At any substantial Li concentration level, the volume expansion due to lithiation is actually non-negligible. However, if the stress change due to the geometric effect of volume expansion is much smaller than that due to constraint, i.e. the dimensionless combination E 2 10 , the attractor for stress 2 103 2 1 v still exists, just the value would tapper off with volume expansion at higher Li concentrations. Equation (13) contains 6 parameters, most of which are directly measurable. Here, we take the representative values for the Young’s modulus E 80 GPa and Poisson’s ratio 0.22 of amorphous silicon.17 The parameter is related to the volumetric transformation strain v C . From the commonly used linear relation,18-20 we obtain approximately 0.5 and 10 v exp x 1 . (For direct comparison, we have adopted the dimensionless measure of concentration, the x in Li xSi .) The only parameter that is not directly measurable is the ratio 2 . By taking the value 2 0.8 GPa-1 and numerically integrating Eq. (13), we reached a good agreement with the experimental data19 for both lithiation and delithiation processes, as plotted in Figure 3. If we take the value of the deviatoric transformation strain, 2.4 , from the result of atomistic simulations between the composition 0 x 1 ,15 the fitting parameter 2 0.8 GPa-1 corresponds to a driving force of approximately 0.9eV per Si atom from Si to LiSi, which is within the reasonable range for lithiation reactions. The reaction is thought to be accomplished through numerous small steps between Si and LiSi, each of which driven by a fraction of the total energy drop (0.9eV), and the linear kinetics assumption is not violated. It should also be noted that the driving force changes its sign from lithiation to delithiation, and consequently 2 is also negative during delithiation. Theory Experiment 2 2 (GPa) 1 0 -1 -2 0 0.5 1 x in Lix Si 1.5 2 Figure 3. Calculated stress-composition loop during a galvanostatic lithiation-delithiation cycle of a lateral constrained Si sample with homogeneous deformation. The experimental data (dots) 11 are extracted from Ref. 17. With only one fitting parameter, the model recovers the stress evolution during both lithiation and delithiation processes. Here for the purpose of illustration, we have taken many parameters to be composition independent. It is believed that if these parameters, such as E C , C , and v C are properly measured or calculated,15,16 even better agreement with experiments could be achieved. Compared to the model of reactive plastic flow,19 these composition-dependent material parameters only depends on the thermodynamic state rather than the kinetics or rate of reaction, and could thus be much easily measured ex-situ. After all, the change in the overall shape of the curve will be small. Although the current model of anisotropic reaction needs much less fitting parameters, it recovers the experimental observation of 1D constrained lithiation of amorphous Si equally well as the model of reactive flow.3,4,19 Moreover, it is expected that the two models will give similar predictions in other stress states, as the transformation strain in the current model and the plastic strain in flow theories are taken to be proportional to the stress deviator. However, as the two models have different physical origins, it is still possible to distinguish them experimentally. Let us still consider the lithiation process as an example. The reactive flow model considers the plateau of stress to be a consequence of plastic yielding. In other words, every point along the plateau of the curves in Figure 3 corresponds to a state in which the solid has already yielded. In the current model, on the other hand, the material is deformed elastically throughout the entire process (if the stress is below the yield strength of the solids). Despite the assumption that the yield strength is reaction dependent,19 a yielded solid is easily distinguishable from one in elastic regime. For example, one can apply an instantaneous mechanical load in situ to the sample during lithiation / delithiation (e.g. by bending the cantilever backward). The additional load 12 will result in a permanent plastic deformation according to the reactive flow theory, while the current model predicts a full recovery upon unloading if the additional load is not enough to yield the sample mechanically. Another major difference lies in the initial steep part of a lithiation or delithiation curve which is thought to be elastic in the reactive flow model, while in the current model the inelastic deviatoric deformation takes place whenever a stress deviator is present. Complicated by the formation of the solid electrolyte interface layer, the initial lithiation stage during the first cycle of a virgin sample may not be the best candidate for comparison. If we look at the lithiation/delithiation curves of subsequent cycles,20 the initial part tends to be more curved as predicted by the current model (Figure 4a). Moreover, we suggest measurements to be carried out over a small lithiation-delithiation cycle, in which the composition change is so small that the resultant stress is much lower than the plateau level of a complete cycle. As shown in Figure 4b, the current model predicts a stress-composition hysteresis loop that gradually shifts in stress values, due to the reaction-induced deviatoric deformation. In contrast, in the absence of plastic flow, the reactive flow model simply predicts overlapping lithiation and delithiation curves (or straight lines), and no shift (or relaxation) in subsequent cycles. Once this prediction is verified, such experiments can be used to determine some of the composition-dependent parameters or kinetic parameters, so that no fitting parameter will be needed in the model, and more accurate prediction could be achieved. 13 a b 2 1.5 1 2 (GPa) 1 2 (GPa) 2 0 -1 0.5 0 -0.5 -1 -2 0 0.5 1 x in Lix Si 1.5 2 -1.5 0.45 0.5 0.55 x in Lix Si 0.6 0.65 Figure 4. Stress evolution in repeated lithiation-delithiation cycles. a) Relatively large cycles of Li concentration x 0 ~ 2 , with slight shifted state of charge between each cycle. b) Small cycles around x 0.55 , in which the stress clearly exhibits inelastic hysteresis and relaxation. In summary, this letter presents a kinetic model for the anisotropic deformation induced by reactions in amorphous solids. The model uses the inelastic transformation strain to characterize the state of reaction, including the orientation information of the resultant, and makes no hypothesis beyond linear kinetics. The model carries much fewer parameters than existing theories describing similar phenomena, and most parameters can be directly measured or calculated through atomistic simulations. The model is a natural extension to the linear kinetic law of reactions, and needs no modification on thermodynamics or linear elasticity. The model clearly differentiates the inelastic transformation strain due to reaction and the plastic strain due to stress. The model is illustrated through the example of constrained lithiation and delithiation of Si thin film, and compared to published experiments. With only one fitting parameter, the prediction of the model reaches a satisfactory agreement with experimental measurements. Some further experiments have been suggested to validate the current model and distinguish it from existing reactive flow theories. We are eagerly waiting for experimental verifications. 14 Although Li-Si is used as a sample system here to illustrate the theory, it is believe that the current model can also be applied to other material systems of amorphous solid-state reactions, especially those with large volumetric changes or under large deviatoric stresses. The current approach may be further extended to model the nonlinear kinetics of far-from equilibrium reactions, although the resulting kinetic relations is expected to be more complex than that given herein. AUTHOR INFORMATION Corresponding Author *E-mail: whong@iastate.edu. ACKNOWLEDGMENT WH acknowledges the National Science Foundation for the support through grant CMMI1000748. REFERENCES (1) Bower, A. F., Guduru, P. R., Sethuraman, V. A., J. Mech. Phys. Solids 2011, 59 (4), 804828. (2) Liu, X. H. et al, Nano Lett. 2011, 11, 3312-3318. (3) Brassart, L.; Suo, Z. Int. J. Appl. Mech. 2012, 4 (03), 1250023. (4) Brassart, L; Suo, Z. J. Mech. Phys. Solids 2013, 61 (1), 61-77. (5) Li, F.; Zhang, L.; Metzger, R. M. Chem. Mater. 1998, 10 (9), 2470-2480. 15 (6) Li, H.; Huang, X. J.; Chen, L. Q.; Zhou, G. W.; Zhang, Z.; Yu, D. P.; Mo, Y. J.; Pei, N. Solid State Ionics 2000, 135 (1–4) 181-191. (7) Gong, D.; Grimes, C.; Varghese, O. K.; Hu, W.; Singh, R. S.; Chen, Z.; Dickey, E. C. J. Mater. Res. 2001, 16 (12), 3331-3334. (8) Obrovac, M. N.; Christensen, L. Electrochem. Solid-State Lett. 2004, 7 (5) A93-A96. (9) Chan, C. K.; Peng, H. L.; Liu, G.; McIlwrath, K.; Zhang, X. F.; Huggins, R. A.; Cui, Y. Nat. Nanotechnol. 2008, 3 (1) 31-35. (10) Key, B.; Bhattacharyya, R.; Morcrette, M.; Seznec, V.; Tarascon, J. M.; Grey, C. P. J. Am. Chem. Soc. 2009, 131 (26) 9239-9249. (11) Wan, W. H.; Zhang, Q. F.; Cui, Y.; Wang, E. G. J. Phys.: Condens. Matter 2010, 22 (41) 415501. (12) Chon, M. J.; Sethuraman, V. A.; McCormick, A.; Srinivasan, V.; Guduru, P. R. Phys. Rev. Lett. 2011, 107 (4) 045503. (13) Levitas, V. I.; Attariani, H. Sci. Rep. 2013, 3, 1615. (14) Cui, Z.; Gao, F.; Qu, J. J. Mech. Phys. Solids 2012, 60 (7), 1280-1295. (15) Shenoy, V. B.; Johari, P.; Qi, Y. J. Power Sources 2010, 195 (19), 6825-6830. (16) Zhao, K. J.; Wang, W. L.; Gregoire, J.; Pharr, M.; Suo, Z. G.; Vlassak, J. J.; Kaxiras, E. Nano Lett. 2011, 11 (7) 2962-2967. 16 (17) Freund, L.B.; Suresh, S., Thin film materials, Cambridge University Press, Cambridge, UK, 2003, p. 96. (18) Sethuraman, V. A.; Chon, M.J.; Shimshak, M.; Van Winkle, N.; Guduru, P. R. Electrochem. Comm 2010, 12, 1614-1617. (19) Zhao, K; Tritsaris, G. A.; Pharr, M.; Wang, W. L.; Okeke, O.; Suo, Z.; Vlassak, J. J.; Kaxiras, E. Nano Lett. 2012, 12 (8), 4397-4403. (20) Nadimpalli, S. P. V.; Sethuraman, V. A.; Bucci, G.; Srinivasan, V.; Bower, A. F.; Guduru, P. R. J. Electrochem. Soc. 2013, 160, A1885-A1893. 17