Inversion Domain Boundaries in GaN Nanowires ...

9 downloads 0 Views 4MB Size Report
the IDB* formation energy determined from first principles by Northrup et al.1 The Tersoff-‐. Brenner potential parameters of Nord et al. were used 3 and it was ...
Supporting Information for:

Inversion Domain Boundaries in GaN Nanowires Revealed by Coherent Bragg Imaging.  

Stéphane Labat1,*, Marie-Ingrid Richard1,2, Maxime Dupraz3,4, Marc Gailhanou1, Guillaume Beutier3,4, Marc Verdier3,4, Francesca Mastropietro1, Thomas W. Cornelius1, Tobias U. Schülli2, Joel Eymery5,6,*, Olivier Thomas1 *Corresponding  authors’  email:  stephane.labat@univ-­‐amu.fr,  [email protected]       1

Aix Marseille Univ., CNRS, Univ. Toulon, IM2NP UMR 7334, F-13397 Marseille, France 2

3

Univ. Grenoble Alpes, SIMAP, F-38000 Grenoble, France 4

5

6

ID01 ESRF, F-38043 Grenoble, France

CNRS, SIMAP, F-38000 Grenoble, France

Univ. Grenoble Alpes, F-38000 Grenoble, France

CEA, INAC-SP2M, "Nanophysique et semiconducteurs" group, F-38000 Grenoble, France

Table  of  Content     1:  Reconstruction  process   2:  Additional  measurements  on  other  nanowires   3:  Molecular  statics  simulations      

 

Supporting  Information  1:  Reconstruction  process   The  data  treatment  of  the  (004)  Bragg  peak  will  be  detailed  as  an  example  of  the  procedure  for   sample  reconstruction.  The  error  metric  that  quantifies  the  matching  between  the  retrieved  intensity   𝐴 𝑞 ²!"#!$"%"&  ,  and  the  measured  intensity    𝐼 𝑞

𝐸𝑟 =

𝐴 𝑞

!"#$%&"'  is  chosen  as:  

!"#!$"%"&

𝐼 𝑞

− 𝐼 𝑞 !"#$%&"'

!"#$%&"'

²

 

The  measured  intensity  consists  in  a  regular  array  of  512x256  values  or  pixels.  A  large  square  support   of  90x90  pixels  is  used  as  a  primary  support  in  the  Shrink-­‐Wrap  procedure.  After  5  cycles  of  50  ER   followed  by  100  HIO,  a  shrinking  procedure  is  used  to  decrease  the  size  of  the  support:  we  remove   pixels  from  the  edge  of  the  support  with  a  threshold  of  25  %  of  the  average  modulus.  A  new  iteration   is  done  with  the  updated  support  (5  cycles  of  50  ER  followed  by  100  HIO)  ended  with  the  same   shrinking  procedure.  This  shrinking  procedure  slows  down  around  the  140th  iteration  as  shown  by  the   drawing  of  the  support  size  as  a  function  of  the  iteration  number  (Fig.  S1).  This  corresponds  to  the   iteration  for  which  the  support  starts  including  non-­‐physical  solution  for  our  samples  in  terms  of   voids.  The  shrinking  procedure  is  stopped.  Then,  ten  supports  with  a  slight  variation  of  10-­‐pixels  size   from  the  one  retrieved  after  140  iterations  are  used  with  20  cycles  of  50  ER  +  100  HIO.  For  each   support,  one  hundred  initial  random  phase  sets  are  associated  to  the  measured  intensity  and  give   one  hundred  different  reconstructions.  The  noise  of  the  measured  intensity  precludes  discriminating   the  best  candidate  from  the  error  metric  criterion  alone  (Fig.  S2).  Indeed,  for  intensities  ranging   between  0  and  105,  a  Poisson  distribution  noise  involves  10-­‐3  error  bar  on  the  metric  values.  Thus  the   82  reconstructed  intensities  of  Fig.  S2  giving  an  error  metric  between  4.5  10-­‐3  and  5.5  10-­‐3  match  the   measured  intensity  in  the  same  way.  To  discriminate  these  solutions,  the  standard  deviation  of  the   modulus  maps  is  used.  Samples  with  the  best  homogeneous  modulus  map  and  similar  metric  errors   are  retained.  

 

  Figure  S1  ⏐ Support  size  determination.  The  Shrinking  procedure  used  a  threshold  of  25  %  of  the   average  value  of  the  modulus.  The  boxed  reconstructed  object  is  chosen  as  a  support  for  the   following  procedure  using  ER  and  HIO  only.    

0.016

Error metric

0.014 0.012 0.010 0.008 0.006 0.004

20

25

30

35

40

H L

45

Relative modulus standard deviation %

50  

Figure  S2  ⏐ Error  metric  and  modulus  map  homogeneity  for  hundred  samples  reconstructed  from   004  Bragg  peak.  More  than  80  %  of  the  results  give  the  same  metric  error,  but  have  significant   differences  in  reconstructed  modulus  homogeneity.         From  the  phase  maps,  one  evidences  the  phase  shift  between  domain  1  and  2  (see  Fig.  S3).  The  12   best  reconstructions  in  terms  of  modulus  homogeneity  are  used  to  calculate  the  average  phase  shift   and  the  error  bar  of  this  value  is  estimated  from  the  difference  between  the  maximum  and  minimum   value  divided  by  2.      

Average phase difference radian

H L

- 2.6 Φ1  -­‐ Φ2-­‐3=  -­‐2.80  ±  0.05  radians  

- 2.7 - 2.8 - 2.9 - 3.0

ΔΦexpected  for  IDB*=  -­‐3.07  radians    

- 3.1 - 3.2

0.10

0.15

0.20

0.25

H L

0.30

Standard deviation on the phase values of the largest domain radian   Figure  S3⏐Phase  shift  between  domain  1  and  2-­‐3  (Φ 1  -­‐ Φ 2-­‐3  )  and  phase  map  homogeneity  for  82   reconstructed  samples  from  004  Bragg  peak.  The  12  best  reconstructions  in  terms  of  modulus   homogeneity  are  reported  in  purple.    

Supporting  Information  2  :  additional  measurements  on   other  nanowires   The  004  reflection  has  been  measured  on  several  nanowires.  The  reconstructed  real-­‐space   images  show  a  variety  of  domain  configurations,  but  always  the  same  phase  shift  between   +c  and  -­‐c  domains,  equal  to  -­‐2.8  radians,  with  a  flat  phase  inside  each  domain.  This   observation  suggests  that  the  displacement  of  (c/2+8)  pm  across  the  IDB  is  an  intrinsic   property  of  the  material  and  independent  of  the  domain  configuration.  

 

Figure  S4⏐Reconstruction  of  two  nanowires  from  their  004  Bragg  reflections.  Reconstructed   modulus  a-­‐b  and  phase  c-­‐d  from  004  Bragg  peaks  measured  for  two  different  nanowires.  The   nanowires  present  different  domain  structures  but  the  same  phase  shift  between  the  +c  and  –c   domains.      

Supporting  Information  3  :Molecular  statics  simulations   Introduction Atomistic  simulations  were  carried  out  in  order  to  investigate  the  mutual  interactions   between  several  IDBs*.  Because  of  the  very  large  number  of  atoms  involved  in  a  realistic   configuration,  ab  initio  calculations  cannot  be  performed  and  we  decided  to  carry  out   molecular  statics  calculations  using  a  Tersoff-­‐Brenner  empirical  potential.     As  shown  by  W.H.  Moon  et  al.,2  the  Tersoff-­‐Brenner  potential  is  able  to  reproduce  quite  well   the  IDB*  formation  energy  determined  from  first  principles  by  Northrup  et  al.1  The  Tersoff-­‐ Brenner  potential  parameters  of  Nord  et  al.  were  used  3  and  it  was  verified  that  the  lattice   parameters  (a=  3.1809  Å,  c  =  5.1944  Å)  and  cohesion  energy  (  -­‐  4.528  eV/atom  )  were  within   the  precision  of  those  shown  in  table  3  of  their  publication.  The  c/a  ratio  is  found  very  close   to  the  ideal  wurtzite  axial  ratio  because  only  first  neighbour  interactions  are  considered.   However  experimental  values  of  c/a  are  also  close  to  this  value.4  

Single infinite planar IDB First,  a  single  ideal  planar  IDB*  configuration  was  studied.  For  that  purpose  a  GaN  box  with   periodic  boundary  conditions  (PBC)  in  the  directions  [001]  and  [210]  was  used.  In  the   direction  [010]  perpendicular  to  the  IDB*  surface  no  PBC  was  used  because  it  would  have   constrained  the  system.  As  this  introduces  two  surfaces  perpendicular  to  this  direction,  the   length  of  the  box  was  taken  large  enough  to  be  able  to  separate  surface  relaxation  effects   from  the  IDB*  relaxation.    

A  formation  energy  of  39  meV/Å2  for  IDB*  was  obtained,  in  the  same  range  as  the  value  of   25  meV/Å2  obtained  from  first  principles  calculations  by  Northrup  et  al.1    After  relaxation  the   displacements  induced  by  the  defect  in  the  direction  perpendicular  to  the  IDB  plane  was   found  equal  to  9.3  pm,  close  to  the  value  of  10  pm  deduced  from  the  bond  lengths  indicated   by  Northrup  et  al.1  Along  the  c-­‐axis,  the  Ga  terminated  crystal  on  one  side  of  the  defect  is,   after  relaxation,  translated  by  (c/2  -­‐  1.3  pm)  with  respect  to  the  N  terminated  crystal  on  the   other  side.  The  Simulations  show  also  that  the  strain  is  localized  in  the  2-­‐3  atomic  planes  on   both  sides  of  the  boundaries.  

Multiple IDBs in a nanowire This  simulation  was  carried  out  with  a  nanowire  size  and  geometry,  and  an  IDB  configuration   as  close  as  possible  to  the  experimental  ones.   Periodic  boundary  conditions  were  applied  along  the  nanowire  axis  corresponding  to  the  +c   or  -­‐c  axis  for  the  GaN  crystal.  The  simulation  box  contained  around  1.4  107atoms. In  the  case  of  a  complex  configuration  of  IDBs  such  as  the  one  encountered  in  the  GaN  wire,   a  complicated  displacement  field  is  expected,  as  each  of  the  two  domains  with  a  Ga-­‐ terminated  surface  (labelled  2  and  3,  see  Fig.  3b)  is  separated  from  the  main  N-­‐terminated   domain  (labelled  1)  by  two    IDB  planes  with  a  different  orientation.  Molecular  statics  was   used  to  relax  the  structure  and  determine  this  displacement  field.    The  uz  component  along   the  c-­‐axis  predicted  by  our  atomistic  simulations  (about  -­‐1  pm)  is  much  smaller  than  the   experimental  value  of  +8  pm  obtained  experimentally  from  the  004  reflection,  which  is   independent  from  the  in  plane  (ux,  uy)  displacement  field.    For  five  reflections,  the   displacement  induced  phase  shift  was  added  to  the  structure  factor  and  the  resulting  phase   maps  are  shown  in  Fig.  S5.  Because  of  the  displacement  field,  there  is  a  significant  difference  

-­‐  as  high  as  0.3  radian  -­‐  between  the  domains  2  and  3,  in  particular  for  the  hkl  reflections   with  h≠0,  which  is  also  not  observed  experimentally.     1. Northrup,  J.E.,  Neugebauer,  J.  &Romano,  L.T.    Inversion  Domain  and  Stacking  Mismatch   Boundaries  in  GaN.Phys  Rev.  Lett.77,  103-­‐106  (1996).   2. Moon,  W.H.  &Choi,  C.H.    Molecular-­‐dynamics  study  of  inversion  domain  boundary  in  w-­‐ GaN.Physics  Letters  A  352,  538–542,  (2006).   3. Nord,  J.,  Albe,  K.,  Erhart,  P.  &Nordlund,  K.  Modelling  of  compound  semiconductors  :   analytical  bond-­‐order  potential  for  gallium,  nitrogren  and  gallium  nitride.  J.  Phys.  :Condens.   Mater.  15,  5649–5662  (2003).   4. Paszkowicz,  W.,  Podsiadło,  S.,  Minikayev,  R.  Rietveld-­‐refinement  study  of  aluminium  and   gallium  nitrides,  Journal  of  Alloys  and  Compounds382,100–106  (2004).  

     

    FIGURE  S5⏐  Calculated  phase  maps  of  the  GaN  cross-­‐section  for  different  reflections.  The  maps   include  the  effect  of  the  structure  factor,  of  a  c/2  displacement  along  z  of  the  N  terminated  regions   and  of  the  additional  displacement  field.