Competition between Kardar–Parisi–Zhang and Berezinskii–Kosterlitz–Thouless kinetic roughening on (001) singular surface during steady crystal growth Authors: Noriko Akutsu, Yoshihiro Kangawa Published: 2024 ABSTRACT AbstractKinetic roughening of the (001) singular surface during steady crystal growth is studied on the basis of a lattice model using the Monte Carlo method. At a sufficiently low temperature, there are known to be two kinetic roughening points as the driving force for crystal growth Δμ increases. At a low driving force ΔμKPZ(001), there is the Karder–Parisi–Zhang (KPZ) roughening transition point. On the KPZ rough surface, elementary steps around islands are well defined though the surface is thermodynamically rough, with a roughness exponent α consistent with the KPZ universal value of 0.3869. Island-on-island structures were found to be crucial in forming the KPZ rough surface. To understand the effects of the atomical roughness of the (001) surface and the interplay of steps on long-period undulations on this surface, the dependence on the temperature T and driving force for crystal growth Δμ of surface quantities is investigated. At higher temperatures, additional Berezinskii–Kosterlitz–Thouless (BKT) rough and re-entrant KPZ regions are found for large Δμ, where the crystal surface grows adhesively. A T–Δμ kinetic roughening diagram is also presented.Subject terms: Nonlinear phenomena; Statistical physics; Surfaces, interfaces and thin films; Computational nanotechnology Introduction The “smooth” surface of a complete crystal is believed to grow by a two-dimensional (2D) nucleation process1–6. In the case where the surface grows in a 2D poly-nucleation process near equilibrium, researchers have recently clarified using the Monte Carlo method that the surface is thermodynamically rough even though it is atomically smooth7 and the step edges are well defined8. Here, thermodynamic roughness9–11 can be defined on the basis of the criteria at equilibrium where L is the linear size of the system and W=W(L)=[h(x→)-⟨h(x→)⟩]2, in which h(x→) is the height of the surface at site x→, and ⟨·⟩ represents the ensemble average. This thermodynamically rough but atomically smooth surface is a newly discovered kinetically roughened surface8 with the Kardar–Parisi–Zang (KPZ) roughness exponent12–14 (KPZ rough surface). This surface also maintains a strongly anisotropic surface growth rate8. This fact provides guidance toward explaining strongly anisotropic growth such as that for snow15 in the Mullins–Sekerka instability16. Whether this fact is unique to the surface at sufficiently low temperatures or is universally true at various temperatures is not yet clear. In the crystal-growth research field, a kinetically roughened surface refers to a surface that grows via an adhesive growth process1,6,17,18. The kinetic roughening point is designated as Δμc, where Δμ is the driving force for crystal growth defined by Δμ=μambient-μcrystal. Here, μambient is the chemical potential of the ambient phase and μcrystal is the chemical potential of the crystal phase. The kinetic roughening point can be defined in several ways. One definition is G∗/kBT=πa2γs2/(kBTΔμc)=14,5, where G∗ is the nucleation barrier for the critical nucleus, kB is the Boltzmann constant, T is the temperature, a is the lattice constant of a cubic lattice, and γs is the step free energy per length. Cuppen et al.19 have proposed an alternative definition based on the observation of n-C22H44 crystal growth from n-hexane solutions17 and a Monte Carlo calculation reported by van Veenendaal et al.20. They defined Δμc as the zero point of the non-equilibrium step free energy, similar to the thermal roughening that belongs to the Brezinskii–Kosterliz–Thouless21–23 universality class (BKT rough surface). In the BKT rough surface at equilibrium, W2(L) is expressed by the wavenumber q for slope fluctuations caused by a capillary wave as W2(L)=[kBT/(2πgdetf(2))]∫1/L1/aq/(m2+q2)dq24–26, where g is a geometrical factor, detf(2) is the determinant of the surface stiffness tensor, and m is a constant. When m is not zero, W2(L) converges to a constant value (smooth surface); by contrast, when m=0, W2(L) diverges logarithmically because of the lower cut-off of the integral (the rough surface). At T=TR, W2(L)=(1/π2)lnL, where TR is the thermal roughening temperature. The amplitude 1/π2 is specific to the BKT universal class25,27–29. The amplitude of the height–height correlation function g(r→) on the rough surface is twice the amplitude of W2(L). In statistical mechanics and non-linear physics, kinetic roughening refers to KPZ roughening30–32. When surface width W(L) is extended to a time-dependent surface width W(L, t) with h(x→,t), the fluctuating interface often obeys the Family–Vicsek scaling function30,31 where α, β, and z are the roughness, growth, and dynamical exponents, respectively. In the non-equilibrium steady state, the surface width is characterised by the roughness exponent α. On the basis of the symmetry principle, we derive a surface-growth KPZ equation that includes a non-linear term12 as where v0 is the constant surface velocity, ν>0 is a coefficient related to surface tension, λ is a coefficient for the non-linear term, and ηt is white noise in space and time. In the case of a 2D surface in a three-dimensional (3D) system, the values of these exponents are predicted to be α=0.3869, β=0.2398, and z=1.613113. The KPZ universality class is known to explain many non-equilibrium systems such as directed polymers and quantum systems14,33,34. However, in the case of crystal growth, the observed roughness exponents differ from those predicted by the KPZ model14,31,35, with the exception of several special surface systems36–38. Numerous studies have been conducted to determine why crystal surfaces do not become KPZ rough31. One of the reasons why KPZ behaviour cannot be observed in crystal growth is that surface diffusion occurs during growth using methods such as molecular beam epitaxy (MBE)31,33,39–42. In the case of epitaxial growth from the vapor phase, the Villain–Lai–Das Sarma (VLDS) class is known41,43,44. The exponents are α=(4-d)/3, β=(4-d)/(8+d) and z=(8+d)/3. When d=2, we know that α=2/3, β=1/5, and z=10/3. Notably, far from equilibrium, the Mullins–Sekerka16 interface instability is caused by surface or volume diffusion2,4. This instability leads to dendritic growth4,45–47 and formation of other spatiotemporal patterns4,48–51. These patterns are self-similar rather than self-affine. Another reason is step bunching or macrostep formation at equilibrium and near equilibrium52. Because of microscopic step–step attraction, macrosteps are formed at sufficiently low temperatures at equilibrium53–59. For 0≤Δμ<Δμcr, the terraces and side surfaces are atomically and thermodynamically smooth, where Δμcr is a crossover point between 2D single and poly-nucleation processes at the lower edge of the macrostep59. The mean height of the macrosteps decreases as Δμ increases until disassembly occurs at ΔμR60–62. For Δμcr<Δμ<ΔμR, though both terraces and side surfaces are atomically smooth, W calculated by the Monte Carlo method diverges as L→∞. The provisional roughness exponent changes as αJ≈0.55∼0.85 following the shift of ΔμR(L), and αJ converges to α=0.60 in the limit of L→∞52. In this article, focusing on the (001) singular surface, we will clarify whether the KPZ roughening point and the “separate” crossover point to a BKT rough surface is a universal phenomenon or a phenomenon specific to the special temperature for nucleation-limited steady growth. To this end, the surface growth velocity V and surface width W=W(L) are calculated for several temperatures using the Monte Carlo method as a non-conserved system. The microscopic model based on the Monte Carlo simulation is a restricted solid-on-solid (RSOS) model. Here, “restricted” means that the nearest-neighbour (nn) height difference is restricted to 0 and ±163–65. The RSOS model is equivalent to the 19-vertex model66,67. The model can be mapped to a 2D XY spin model68, the 1D quantum spin 1 chain model69, and the 1D delta function Bose gas model70 at equilibrium. Unfortunately, the model cannot be solved exactly using the Bethe ansatz71. Therefore, the Monte Carlo method is adopted so that the entropy effect is precisely taken into account. The model is more microscopic than the phase field model on a mesoscopic scale15 or the continuous model for crystal growth1,2,72–74, whereas it is a more coarse-grained model than the atomic model used in first-principles quantum calculations75. Results Model and calculation method RSOS model The energy of a surface with an orientation close to (001) and exhibiting (001) terrace roughness can be expressed by the discrete Hamiltonian76 where h(n, m) is the surface height at site (n, m) on a square lattice, N is the total number of lattice points, ϵsurf is the surface energy per unit cell on the planar (001) surface, and ϵ is the microscopic ledge energy associated with nn interactions. The summation with respect to (n, m) is over all sites on the square lattice. The RSOS condition, meaning that the height difference between nn sites is restricted to {0,±1}, is required implicitly. In this equation, Δμ is the driving force for crystal growth. To exclude diffusion effects, we assumed that the ambient phase is uniform. This assumption can be realised at the nanometre length scale near equilibrium. In the case where the ambient phase is an ideal solution, Δμ=kBTlnC/Ceq77, where kB is the Boltzmann constant, T is the temperature, C is the concentration of the solute, and Ceq is the concentration of the solute at saturation. If the ambient phase is an ideal gas, Δμ=kBTlnP/Peq78, where P is the gas pressure and Peq is the gas pressure at equilibrium. Parameters ϵ and Δμ in the present model correspond to ϕ and Δμ in van Veenendaal et al.’s work20. The RSOS model is a more restricted model than the Kossel model used by van Veenendaal et al.20. Because the RSOS model is a coarse-grained model used for first-principles quantum mechanical calculations, ϵsurf and ϵ are related to the surface free energy in the atomic model that includes the entropy for lattice vibrations and distortions75. Thus, these variables are affected by temperature. However, the present work assumes constant values for ϵsurf and ϵ in all of the calculations. On the other hand, the RSOS model is a microscopic model for continuous models or the mesoscopic phase field model. The step stiffness and surface stiffness in the RSOS model provide the values of the coefficient in the continuous1,2,72–74 or phase field model15. Monte Carlo calculations Crystal growth proceeds by the attachment or detachment of “atoms” at randomly chosen sites, where in this work “atom” means a specific unit or cube. The surface configuration was updated using the Metropolis algorithm. The transition probability was 1 for ΔE=Ef-Ei≤0 and was exp(-ΔE/kBT) for 0<ΔE, where Ei is the surface energy of the initial configuration and Ef is the surface energy of the proposed configuration. The surface energy was calculated on the basis of Eq. (4). The number of atoms in the crystal does not have to be conserved during the process, making the system a non-conserved system. The present work did not include atom exchange on the surface, i.e., surface diffusion was neglected. Note that the exponents for a fluctuating surface in a conserved system have different values than for non-conserved surfaces79. The first 2×108 Monte Carlo steps per site (MCS/site) were ignored, and each quantity was averaged over the subsequent 2×108 MCS/site. After 1×108 MCS/site, the calculated values were almost constant. For several parameter sets, we averaged over the next 2×108 MCS/site (4×108∼6×108 MCS/site) and confirmed that the latter values agree with the former values within an error of 0.2%. The surface slope p and the surface growth velocity V were calculated as macroscopic variables, such as where Nstep is the number of steps, which was fixed during the simulation, a=1 is a lattice constant, and τ~ is set to 2×108 MCS/site. Since the present study investigates the (001) surface of a simple cubic lattice, Nstep and p are zero. Periodic boundary conditions were adopted in the vertical ([1¯10]) and horizontal ([110]) directions at a rotation angle of 45 degrees around the ⟨001⟩ axis. This is to allow a comparison of the present results with the figures in Ref.8. In Ref.8, surfaces tilted towards the (111) direction were also studied, where steps run in the ⟨1¯10⟩ direction. Figure 1 shows an example of the simulated surfaces for several temperatures and driving forces for crystal growth. Examples of KPZ (panels (a) and (b)) and BKT (panels (c) and (d)) rough surfaces at 4×108 MCS/site. Upper figures show overhead views where the surface height is indicated by brightness. Lower figures show side views. In the side views, the lines indicate the surface height at the bottom line of the overhead views. L=3202. The squared surface width was calculated as where W is the surface width normal to the surface, x~ and y~ are the [110] and [1¯10] directions, respectively, and ⟨·⟩y~ and ⟨·⟩x~ are the averages over the y~ and x~ directions, respectively. The calculated W scaled by lnL or Lα as a function of Δμ is shown in Fig. 2 for several temperatures. Scaled surface widths as functions of Δμ. Upper figures: squared surface width W2 scaled by lnLwhere L is the size of the system. Lower figures: surface width W scaled by Lαwhere α is assumed to be a roughness exponent with the KPZ value (0.386913). Panel (a) is taken from Ref.8 licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0). Panels (b)–(d) show the present numerical results. Point A: ΔμKPZ(001). Point C: ΔμKtoB(001). Point D: ΔμBKT(001). Point E: ΔμBtoK(001). Though the RSOS model is simple, there are several external parameters: L, T, Δμ, and p (Nstep). In Ref.8, at the fixed temperature kBT/ϵ=0.4, L, Δμ, and p were systematically changed for a (001) singular surface, for which the KPZ roughening transition point ΔμKPZ(001) was identified. For larger Δμ, a KPZ rough to BKT rough crossover point ΔμKtoB(001) was also found. For ΔμKPZ(001)<Δμ<ΔμKtoB(001), the surface fluctuation width W is divergent as L increases to ∞ with the KPZ roughness exponent. An analysis of the surface growth velocity V on the (001) surface, showed that ΔμKPZ(001) agrees with the crossover point between 2D single nucleation to 2D poly-nucleation growth processes. However, whether ΔμKtoB(001) can be regarded as the kinetic roughening point Δμc is not clear. This article examines the ΔμKPZ(001) and ΔμKtoB(001) temperature dependence for a fixed surface slope p=0 (the (001) surface). For kBT/ϵ=0.4, islands are formed on (001) terraces by 2D nucleation. At higher temperatures, more thermally excited islands on (001) terraces are formed and the free energy of the steps decreases. The study systematically applies various driving forces for several typical temperatures, kBT/ϵ=0.63, 0.83, and 1.4, compared with the surface of kBT/ϵ=0.4. KPZ rough surface The driving-force dependence of the squared surface widths scaled by lnL are shown in the upper figures in Fig. 2 for several temperatures. The lower figures show the surface widths scaled by Lα with KPZ roughness exponent α assumed to be 0.3869. Panels (b)–(d) show the present numerical results. The surface with 0≤Δμ<ΔμKPZ(001) (point A in Fig. 2a–c, lower figures) has a small surface width and a constant value as the system size increases. Hence, the surface is atomically and thermodynamically smooth in addition to being at equilibrium. For ΔμKPZ(001)<Δμ, the surface width increases with increasing system size. Hence, the surfaces are thermodynamically rough according to the definition in Eq. (1). In particular, for ΔμKPZ(001)<Δμ<ΔμKtoB(001) (between points A and C in Fig. 2 (lower)), the surface width increases as W∝Lα, where α=0.3869. From the Monte Carlo results, the values of ΔμKtoB(001) are determined to be the largest Δμ for the KPZ rough region. The values of ΔμKPZ(001) and ΔμKtoB(001) are listed in Table 1. The KPZ roughening transition point ΔμKPZ(001) decreases with increasing temperature because the step tension decreases as the temperature increases. This KPZ roughening point agrees with the crossover point between the 2D single and poly-nucleation growth modes (refer to the “Surface growth rate” subsection). Table 1Characteristic driving forces.ΔμKPZ(001)/ϵΔμpoly∗(001)/ϵΔμKtoB(001)/ϵΔμBKT(001)/ϵΔμBtoK(001)/ϵΔμKPZ,2(001)/ϵkBT/ϵAA′CDEF0.40.55±0.050.58±0.042.00 ±0.10>2.6––0.630.25±0.030.24±0.031.30 ±0.102.10±0.142.50±0.14–0.830.101±0.0150.06±0.060.6 ±0.21.30 ±0.142.30±0.14–1.00.03±0.03–0.07 ±0.071.00±0.142.10 ±0.142.70 ±0.141.2–––>02.00 ±0.142.60±0.101.4–––>01.90 ±0.14>2.6 Characteristic driving forces. For a temperature kBT/ϵ=1.4 (Fig. 2d), it is higher than the phase transition point of the 2D nn Ising model kBTc/ϵ=1/ln(1+2)≈1.135, whereas the temperature is lower than the thermal roughening transition temperature TR(001)/ϵ=1.57856,57,66, where TR(001) is the thermal roughening temperature for the (001) surface. Clear ΔμKPZ(001) and ΔμKtoB(001) cannot be obtained. In Fig. 1a and b, the overhead and side views of the simulated surfaces for KPZ roughened surfaces are shown at kBT/ϵ=0.63 and 0.83, respectively. 2D island-on-island (multilayered island) structures are clearly observed. Islands smaller than the critical nucleus are also observed and are formed more frequently than the KPZ roughened surface at kBT/ϵ=0.48. These small islands are generated by thermal fluctuations. It is clear that the step edges around the islands are well defined. These observations indicate that the surfaces in panels (a) and (b) are atomically smooth even though they are thermodynamically rough. The surface grows via a 2D poly-nucleation process based on the well-defined surface steps. Notably, the growing surfaces are not always thermodynamically rough. For Δμ<ΔμKPZ(001), surfaces are atomically and thermodynamically smooth (Fig. 2). BKT rough surface In the present work, a BKT rough region has been newly found for a large Δμ at low temperatures. In Fig. 2 (upper figures), a BKT rough region is observed between ΔμBKT(001) (points D in Fig. 2 (upper figures)) and ΔμBtoK(001) (points E in Fig. 2 (upper figures)) for kBT/ϵ=0.63 and 0.83. For ΔμBKT(001)≤Δμ<ΔμBtoK(001), the W2 is logarithmically divergent with L, similar to the thermal roughening of the BKT universality class. The values of ΔμBKT(001) and ΔμBtoK(001) are listed in Table 1. In the region from point D to point E (Fig. 2 (upper figure)), the values of W2/lnL (amplitude) are larger than the universal value of 1/π2∼0.1013, which is the amplitude at TR(001)25,28. The surface structures in the regions are shown in Fig. 1c and d. Small islands and negative islands (clusters of holes) are so abundant that surface steps are not discernible. Therefore, the surfaces are not only thermodynamically rough but also atomically rough except for the ramified-step region near equilibrium. The ramified-step region is the region with ΔμKtoB(001)<Δμ<Δμc,MC(001), where Δμc,MC(001) is the smallest Δμ in the region with a linear surface growth rate. We will discuss Δμc,MC(001) in the next subsection “Surface growth rate”. ΔμBKT(001) decreases as the temperature increases because the population of islands and negative islands generated by thermal fluctuations increases as the temperature increases. At temperatures higher than about Tc for the 2D nn Ising model, the region 0<Δμ<ΔμBtoK(001) (point E) is BKT rough. The KPZ rough region and the adjacent crossover region disappear. At thermal equilibrium at kBT/ϵ=1.4, Fig. 2d shows that W2/lnL at Δμ=0 is smaller than the universal value 1/π2 at TR(001). Consequently, the surface is thermodynamically smooth at Δμ=0. However, in the non-equilibrium steady state, W for Δμ≠0 is logarithmically divergent for a wide range of Δμ (Fig. 2d (upper figure)). To show the details around Δμ=0, a figure similar to Fig. 2d but with a scale magnified near Δμ=0 is presented in Fig. 3. At equilibrium (Δμ=0), the data in both Fig. 3a and b are scattered. As the system size increases, W2 converges to a constant value. Then, W2/lnL decreases as L increases (Fig. 3a). This is consistent with a thermodynamically smooth surface. Magnified figures of scaled surface widths and surface velocity as functions of Δμ at kBT/ϵ=1.4. (a) W2 scaled by lnL. (b) W scaled by Lα with α=0.3869. (c) Surface growth velocity V. Line: V=aΔμ/ϵ+b where a=0.0684 and b=-2.74×10-4. C′ at Δμc,MC(001) indicates the smallest Δμ within the linear growth region. For non-equilibrium steady growth, W2 increases steeply up to Δμ/ϵ=0.05 and then increases gradually. W2/lnL with different L agrees well for all the data in Fig. 3a except for Δμ=0. It should be noted that ∂V/∂Δμ decreases for Δμ→0. This feature is the same as that for the surface growth rate with a slightly lower temperature than TR shown by Gilmer and Bennema3. The model of Gilmer and Bennema adopted a more realistic transition probability than the present model, and surface diffusion was taken into consideration. However, the features of surface growth velocity are the same as in the present study. Since the different models show the same behaviour, this phenomenon is universal and independent of the details of the system. Therefore, we conclude that the surface is BKT rough for non-equilibrium though the surface is thermodynamically smooth at equilibrium at kBT/ϵ=1.4. Surface growth rate The dependence of the surface growth rate V on the driving force for the crystal growth, Δμ, is shown for several temperatures in Fig. 4a. A plot of lnVvs.ϵ/Δμ is shown in Fig. 4b. The system size dependence of Δμpoly(001) at A′ in Fig. 4b is shown in Fig. 4c. Surface growth rates for several temperatures at the (001) surface (p=0) as functions of Δμ. The unit of surface growth rate is a/τ, where a (=1) is the unit height and τ is the time interval for 1 MCS/site. L=3202a. (a) V vs. Δμ/ϵ. C′ at Δμc,MC(001) indicates the smallest Δμ within the linear growth region. Lines: V=kΔμ/ϵ-b. Values of k and C′ are listed in Table 2. (b) lnV vs. ϵ/Δμ Lines: lnV=-gMC∗ϵ/Δμ+b′. Lines with lighter colour: lnV=-gMC′ϵ/Δμ+b′′. Values of gMC∗ and gMC′ are listed in Table 2. A′ at Δμpoly(001) indicates the crossover point from single to poly-nucleation. (c) System size dependence of Δμpoly(001) at kBT/ϵ=0.63. Line: ln(Δμpoly(001)/ϵ-Δμpoly∗(001)/ϵ)=88.0/ln(L/a)-19.0, where Δμpoly∗(001)/ϵ=0.24. The converged values Δμpoly∗(001) are listed in Table 1. Near equilibrium, the surfaces grow in a 2D single nucleation process except for the surface at temperatures 1.0∼