1. Interface Dynamics in Epitaxial Growth Russel Caflisch Mathematics Department, UCLA

2. Collaborators UCLA: Anderson, Connell, Fedkiw, Gibou, Kang, Merriman, Osher, Petersen (GaTech), Ratsch HRL: Barvosa-Carter, Owen, Grosse, Gyure, Ross, Zinck Imperial: Vvedensky Support from DARPA and NSF under the Virtual Integrated Prototyping (VIP) Initiative and from ARO www.math.ucla.edu/~thinfilm

3. Outline Epitaxial Growth –molecular beam epitaxy (MBE) –layer-by-layer growth Kinetic Monte Carlo –atomistic description –Arrhenius rates Continuum model –island dynamics –level set method –boundary conditions Kinetic model for step edge –density of edge adatoms and kinks on boundary –obtain curvature diffusion Conclusions

4. ABES PEO Effusion Cells MBE Chamber RHEED REMS substrate temperature surface structure morphology monolayer thickness morphology monolayer thickness desorbed and scattered flux In, Ga, Al evaporators Valved As, Sb crackers STM Chamber Growth and Analysis Facility at HRL

5. STM Image of InAs 20nmx20nm 250nmx250nm 1.8 V, Filled States HRL whole-wafer STM surface quenched from 450°C, “low As” Barvosa-Carter, Owen, Zinck (HRL)

6. AlSb Growth by MBE Barvosa-Carter and Whitman, NRL

7. RHEED signatures RHEED = “reflective high energy electron diffraction” –intensity = a - b  (step edge density) –1 oscillation per crystal layer amplitude and decay rate for oscillations is indicator of surface quality Growth Recovery  min I=I 0 e (-t/  )  max Zinck, Owen, Barvosa-Carter (HRL)

8. Epitaxial Growth Growth of thin film as single crystal –crystal properties determined by substrate Layer-by-layer growth –layer (nearly) complete before initiation of next layer Surface features in layer-by-layer growth –adatoms –islands –step edges Data –STM: atomistic picture after growth –RHEED: diffraction intensity = c - (step edge density) Nanoscale morphology can significantly affect device performance

9. Basic Processes in Epitaxial Growth (a) deposition(f) edge diffusion (b) diffusion(g) diffusion down step (c) nucleation(h) nucleation on top of islands (d) attachment(i) dimer diffusion (e) detachment

10. Hierarchy of Models Large range of length and time scales –atomic scale: 1 Å = 10 -10 m –surface feature scale: 10 nm = 10 -8 m –device scale: 1 m = 10 -6 m –wafer scale: 1mm = 10 -3 m Hierarchy of models and simulation methods –ab initio (1 Å, 1 fs) –molecular dynamics (1 Å, 1 fs) –Kinetic Monte Carlo (KMC) (1 nm, 1 s) –continuum (10 nm, 1 ms) –bulk (1 m, 1 s)

11. Atomistic Description of Epitaxial Growth The Kinetic Monte Carlo Method

12. Solid-on-Solid Model Interacting particle system –Stack of particles above each lattice point Particles hop to neighboring points –random hopping times –hopping rate r depends on nearest neighbors r = r 0 e -E / kT E = energy barrier between state before and after hop Deposition of new particles –random position –arrival frequency from deposition rate Simulation using kinetic Monte Carlo method

14. SOS Simulation for coverage=.2 Gyure and Ross, HRL

15. SOS Simulation for coverage=10.2

16. SOS Simulation for coverage=30.2

17. Continuum Description of Epitaxial Growth The Island Dynamics/Level Set Method

18. Continuum Equations Island Dynamics/Level Set Model Description of epitaxial surface –  k = island boundaries of height k represented by “level set function”   k (t) = { x :  (x,t)=k} –Normal velocity v of step edge or island boundary is essential quantity –N= Number of islands –adatom density  (x,y,t) Level of description –continuum in lateral directions (x,y) –discrete (at atomic level) in growth direction z Valid for growth of very thin layers –application: quantum well devices, layer thickness 20 Å –coarse-grained eqtns (e.g KPZ or Villain) for thin film height h(x,y) not valid Diffusion dominant –inverse Peclet number R=D/F (for a=lattice constant=1) –R varies between 10 6 and 10 10, for MBE –F=Deposition flux, D=Diffusion coefficient

19. Island Dynamics/Level Set Equations Adatom diffusion equation  t  - D  2  = F - dN/dt Island nucleation rate dN/dt =  D  1  2 dx  1 = capture number for nucleation Level set equation (motion of  )  t  + v   = 0 v = normal velocity of boundary  To be determined –boundary conditions for  –boundary velocity v determined next –nucleation site

20. Boundary Conditions and Boundary Velocity Boundary condition at island boundaries (irreversible aggregation) (equilibrium: BCF) (mixed type) Normal velocity of boundary  : v = D [  n  ] (irreversible aggregation) v = D [  n  ] -v detach (attachment/detachment - Petersen) v = D [  n  ] + c  ss (edge diffusion)

21. Seeding of new islands Islands nucleate by random binary collisions between adatoms. Assuming that nucleation takes place continuously in time, the rate at which new islands are seeded is given by: where N(t) is the total number of islands nucleated up to time t and denotes a spatial average. Nucleation site chosen at random with spatial density  2 Every time N(t) increases by 1, it is time to seed a new island. –Initially  is set to -.5 at every gridpoint. A new island is seeded by raising the value of  by 1 at the nucleation site and at a few neighboring gridpoints. Atomistic fluctuations –in nucleation site are important –in nucleation time are not important

22. Computing the adatom density Finite difference equation for  –explicit method has severe timestep restriction  t < c  x 2 /D –implicit method required Resulting system has form A x n = b –A and b depend on surface geometry,i.e. on island boundaries –difficult to make A symmetric

23. Computing the adatom density Away from boundaries, use standard spatial discretization Near boundaries, use subcell discretization based on ghost fluid method (Fedkiw) Spatially first order accurate Resulting matrix system is symmetric. Solve using (cholesky) preconditioned conjugate gradient method

24. Update of the level set function

25. Evolution of and : time = 1.7

26. Level Contours after 40 layers In the multilayer regime, the level set method produces results that are qualitatively similar to KMC methods.

27. Simulation of Epitaxial Growth The Island Dynamics/Level Set Method S. Chen, M. Kang

29. Validity and Qualitative Features Conservation of Mass Dependence on nucleation site selection rule Comparison to KMC

30. Island Merger by Level Set Approach Efficient and accurate numerical method Merging of boundaries is automatically handled Method conserves mass time =.1 time =.9

31. Dependence on Nucleation Style Location distribution must be correctly represented random: 1; probabilistic:  2 ; deterministic: max 

32. Scaling of the Island Size Distribution (Stroscio et al PRB, 1994)

35. Island Dynamics vs. KMC Island dynamics is faster than KMC “in principle” –adatom hopping time KMC must resolve handled continuously by island dynamics No faster in practice (so far) –nucleation requires atomistic grid, small times –solution of diffusion equation is slow Some features easier to test –variation in statistic of fluctuations –capture zones of islands (Gibou) –stability Some physics easier to add, some harder –strain easier for island dynamics –reconstruction easier for KMC

36. Comparison of Level Set Method and Alternatives

37. Kinetic Theory for Step Edge Dynamics and Adatom Boundary Conditions with Weinan E

38. Step Edge Components adatom density  edge adatom density  kink density (left, right) k terraces (upper and lower) 

39. Adatom and Kink Dynamics on a Step Edge Attachment at kinks  kink velocity w Kink pair creation  kink creation rate g Kink pair collision  kink loss rate h Reverse processes do not occur in typical MBE growth  no detailed balance  nonequilibrium

40. Kinetic Theory for Step Edge Dynamics

42. Equilibrium Solution Solution for F=0 (no growth) Same as BCF theory D T, D E, D K are diffusion coefficients (hopping rates) on Terrace, Edge, Kink in SOS model Comparison of results from theory(-) and KMC/SOS (  )

43. Kinetic Steady State Solution for F>0 k >> k eq P edge =F edge /D edge “edge Peclet #” Comparison of scaled results from theory(-) and KMC/SOS (  ) for L=25,50,100

44. Macroscopic Boundary Conditions from Step Edge Model Assume slowly varying kinetic steady state along island boundaries Result is “Gibbs-Thomson” BC, but derived from atomistic theory rather than from thermodynamics Reference density  * from kinetic steady, not equilibrium  is curvature of island boundary

45. Constants in BCs

46. Conclusions Island dynamics model –appropriate for very thin films: continuum in x,y; discrete in z –level set simulation method –validated by comparison to SOS/KMC –derivation of boundary conditions Additional physics –attachment/detachment (Petersen) –strain –edge diffusion –multiple species –reconstruction effects