PHYSICAL REVIEW B                                      VOLUME 54, NUMBER 11                                          15 SEPTEMBER 1996-I

           Diffuse scattering from interface roughness in grazing-incidence x-ray diffraction

                    S. A. Stepanov,* E. A. Kondrashkina,* M. Schmidbauer, R. Ko¨hler, and J.-U. Pfeiffer
                             MPG-AG ``Ro¨ntgenbeugung,'' Hausvogteiplatz 5-7, Berlin D-10117, Germany

                                                                  T. Jach
                             National Institute of Standards and Technology, Gaithersburg, Maryland 20899

                                                             A. Yu. Souvorov
                    European Synchrotron Radiation Facility, Boi te Postale 220, Grenoble Cedex F-38043, France
                                                         Received 12 March 1996 
                A theory of x-ray diffuse scattering from interface roughness in grazing-incidence diffraction  GID  is
              presented. The theory assumes dynamical diffraction of x rays from perfect multilayers with the diffuse
              scattering from roughness calculated in the distorted-wave Born approximation. This permits the calculation of
              scattering due to roughness at all points on the diffraction curves, including the vicinity of the Bragg peaks. It
              is shown that the measurements of diffuse scattering in GID can provide information on atomic ordering at
              crystal interfaces which is not accessible by usual x-ray specular reflection and nonspecular x-ray scattering.
              The theory is found to be in good agreement to the two GID experiments carried out with an etched Ge surface
              and an AlAs/GaAs superlattice at the Cornell High-Energy Synchrotron Source and European Synchrotron
              Radiation Facility, respectively. In the case of the etched Ge surface, an anti-Yoneda dip in the diffuse
              scattering pattern at the Bragg peak and two symmetrical shoulders on the Bragg curve wings have been found
              and explained. In the case of the AlAs/GaAs superlattice, the diffuse scattering has been separated from GID
              by means of high-resolution measurements. A comparison between diffuse scattering in GID and diffuse
              scattering in grazing incidence far from the diffraction conditions has shown that the atomic ordering was
              preserved in the interface roughness, while it was partially destroyed in the surface roughness.
               S0163-1829 96 05035-7 



                       I. INTRODUCTION                                   observed, while in Ref. 21 an attenuation of the tails with an
                                                                         exponent similar to that found by Ne´vot and Croce24,25 for
   The combination of Bragg diffraction and total external               the tails of x-ray specular reflection curves was predicted.
reflection  TER  effects in grazing-incidence x-ray diffrac-                 We have assumed that discrepancies between Refs. 15
tion  GID  opens up a wealth of possibilities in the study of            and 21 were due to the diffuse scattering of x rays from
thin surface layers of crystals.1­4 GID has been applied with            surface roughness that was measured in Ref. 15 in combina-
success to investigations of the surface treatment and surface           tion with the diffracted intensity of GID. Therefore, the ef-
oxidation of semiconductor wafers,5,6 to studies of structure            fect of roughness on GID has become the subject of our
transformations during ion implantation,7­9 and to analysis              study.
of strain relaxation in epitaxial layers1,10,11 and multilayers.12           In Sec. II a theory of diffuse scattering arising from sur-
   The measurements of diffuse scattering  DS  in GID can                face roughness as well as correlated and uncorrelated inter-
provide additional information on crystal lattice defects in             face roughness in multilayers  ML's  is presented for GID.
surface layers.13,14 However, along with diffuse scattering              Our model is based on the dynamical theory of GID in
caused by lattice defects, strong scattering due to surface and          multilayers26­28 and the distorted-wave Born approxima-
interface roughness can be expected in GID by analogy with               tion.29­32 The results are general for GID and extremely
x-ray TER studies, since the angle of incidence of the x rays            asymmetric x-ray diffraction, and are applicable in the case
is equally small in both these cases.15­19 Roughness can also            of normal lattice strains in multilayers.
give rise to a change in the diffracted intensities.20­22 Thus               In Sec. III we derive expressions for the diffuse scattering
one has to distinguish between the effects of roughness and              observed in GID with different experimental setups. In Sec.
crystal lattice defects on GID.                                          IV the theory is applied to the explanation of Ref. 15, where
   The effect of roughness on x-ray Bragg diffraction has not            GID curves were taken from a sample with strong surface
been adequately explored. As shown in Ref. 23, in normal-                roughness produced by etching.
incidence diffraction geometries the effect is relatively weak.              In Sec. V some numerical examples demonstrating the
In GID, there are several theoretical predictions21,22 of a              role of diffuse scattering in GID studies of semiconductor
strong effect of roughness on the coherent reflection. How-              multilayers are given. The effect of the correlation between
ever, only a few experimental results20,15 are known, and                roughness of different interfaces is discussed. Calculations
they are not in good agreement with the theory. In particular,           are carried out for different experimental geometries.
in Ref. 15 an enhancement of the intensity of the tails of the               In Sec. VI we describe a high-resolution GID experiment
diffracted beam of GID measured from a rough surface was                 carried out with an AlAs/GaAs superlattice at the optical

0163-1829/96/54 11 /8150 13 /$10.00                              54      8150                        © 1996 The American Physical Society



54                                   DIFFUSE SCATTERING FROM INTERFACE ROUGHNESS . . .                                                  8151

                                                                           and 35  or dynamical3 diffraction theory. The kinematical
                                                                           theory is applicable  i  far from the Bragg peaks from perfect
                                                                           crystals,  ii  for imperfect  mosaic  crystal structures, or  iii 
                                                                           for GID from very thin layers  1-10 monolayers . The dif-
                                                                           fracted intensities in these cases are weak and given by the
                                                                           DWBA. Then the scattering from interface roughness can be
                                                                           calculated in the kinematical theory either simultaneously
                                                                           with the diffraction from layers17­19 or in the second-order
                                                                           DWBA. The dynamical theory is applicable at all points on
                                                                           GID curves corresponding to bulk reflections from perfect
                                                                           structures. In this case, a strong diffraction intensity at the
                                                                           Bragg peak and the effect of diffraction on the incident and
                                                                           specular waves are taken into account. That is just the case
                                                                           for the both experiments discussed in Secs. IV and VI, where
                                                                           the measurements are taken near the Bragg peaks of bulk
                                                                           reflections. Therefore, we use the dynamical diffraction
                                                                           theory in order to explain the peculiarities of diffuse scatter-
                                                                           ing near the Bragg peaks of GID. For the reader's conve-
                                                                           nience, a short outline of the dynamical theory is given be-
                                                                           low.For the sake of simplicity, we neglect the changes in x-ray
      FIG. 1. Schematic layouts of grazing-incidence x-ray diffraction     polarization by diffuse scattering in GID. These changes can
for  a  a real incident wave Ein0 and  b  for an ``imaginary'' incident    be simply added to the model, but they are too weak  of the
wave Eout
          0 which is inverted with respect to one of diffuse scattered     order of noncoplanarity of GID, i.e.,  10 3) to be of an
waves shown in  a  by a dashed line. The cones in  a  schematically        interest for present-day experiments. Our derivations below
illustrate DS along the diffracted and reflected waves of GID. For         are carried out for   polarization. The equations are simply
the explanation of other vectors, see the text.                            extended for   polarization by incorporating cos(2 B) in the
                                                                           x-ray polarizabilities  
beamline of the European Synchrotron Radiation Facility                                                h and   h¯ .
 ESRF . These direct measurements of diffuse scattering in                           A. Some results from the dynamical theory
GID are compared to calculations based on our model. We                                         of GID in multilayers
conclude with some possible uses of diffuse scattering in
GID and their distinction from similar scattering in total ex-                In the framework of the dynamical diffraction theory, the
ternal reflection.                                                         wave fields Ein(r) and Eout(r) can be found as solutions to
                                                                           the system of the dynamical diffraction equations in
                             II. THEORY                                    multilayers.26­28 Let us consider x-ray Bragg diffraction in a
                                                                           multilayer  see Fig. 1 . It is assumed to be a stack of N
      In GID, the diffuse scattering can be observed along the             perfect crystalline layers with laterally matched lattice spac-
directions of both diffracted and reflected beams  see Fig.                ing, and every layer can possess its own lattice spacing in the
1 a  . We restrict our consideration to the scattering along               direction normal to the surface: an                   n
                                                                                                                       z  az  az , where
the diffracted beam, since the majority of measurements are                  anz  az , and index n denotes the number of the layer in
made in this manner.1,5­14,20,22 The analysis of the scattering            the stack counted from the surface. This model corresponds
along the reflected beam of GID is analogous.                              to a so-called unrelaxed multilayer containing no misfit dis-
      The x-ray diffuse scattering is due to the deviations                locations.
  (r) of the polarizability of the scatterer from an ``ideal''                At the first step, we assume flat interfaces between the
distribution  id(r). The most effective way for its calcula-               layers. For Bragg diffraction from a reciprocal-lattice vector
tion, in the lowest-order perturbation   (r), is to apply the              h approximately parallel to the surface, the polarizability can
reciprocity theorem.32 The amplitude of DS can be repre-                   be presented as the following sum over the polarizabilities of
sented as                                                                  the layers:

                                                                                                N
               f    2/4     Eout r    r Ein r d3r,                  1                 flat r      n r H z zn H zn 1 z ,                   2 
                                                                                               n 1
where Ein(r) and Eout(r) are the wavefields in the ideal ob-
ject  with flat interfaces  produced by the incident x-ray                                  n r   n           n        n
                                                                                                        0  h ehn*r   e hn*r.              3 
                                                                                                                       h¯
wave and the wave backprojected to the object from the ob-                                                     n            n
servation point, respectively  see Fig. 1 . The parameter   is             Here H(z) is the steplike Heaviside function, and zn are the
the magnitude of the wave vector of these waves in vacuum.                 coordinates of interfaces. Reciprocal-lattice vectors hn in the
The approximation  1  is commonly referred to as the                       nth layer differ slightly from the mean vector h because of
distorted-wave Born approximation  DWBA .                                  the variations in the normal lattice spacing between layers:
      In case of GID, the wave fields Ein(r) and Eout(r) can be            hn h  hznZ, where   hzn  h, and Z is a unit vector
found with the help of either kinematical33  see also Refs. 34             along the internal surface normal.



8152                                                                                                                S. A. STEPANOV et al.                                                                                                54

      In dynamical Bragg diffraction, the x-ray wave field in                                                                                 The substitution of Eq.  10  into Eq.  9  adds the phase
each layer can be expanded over the sum of the transmitted                                                                                 factor to Dhnj corresponding to the expansion of the x-ray
and diffracted Bloch waves with wave vectors k0n and                                                                                       wave field over h instead of hn in  4 . Thus we can transfer
khn k0n hn ,                and                  amplitudes                            D0n                  and               Dhn ,        this phase factor from  9  to  4 , and proceed to the expan-
respectively:36­38                                                                                                                         sion over h for the whole multilayer. At that, the dispersion
                                                                                                                                           equations  8  remain formally unchanged because the expo-
                     En r  D0neik0n*r Dhneikhn*r.                                                                                    4     nents on the right side cancel each other.
                                                                                                                                              The amplitudes D
      The amplitudes D                                                                                                                                                  0n j can be found with the help of the
                                      0n and Dhn can be treated as constants                                                               boundary conditions for the x-ray waves and their deriva-
satisfying the dynamical diffraction equations in each layer:                                                                              tives at each interface.2­4 It is convenient to present the
                    k2                2                                                                                                    boundary conditions in a (4 4) matrix form,9,26­28
                     0n  0                                   nD                  n D
                           k2              D0n  0 0n  h¯ hn ,                                                                                                                                  U 
                            0n                                                        n                                                                                 SvEv S1F1 D1 ,
                                                                                                                                     5 
                    k2                2                                                                                                                               S  L D                          U D
                      hn  h                                  n                                                                                                         1F1          1 S2F2                   2 ,
                                                                       D              nD
                           k2              Dhn  h 0n  0 hn .                                                                                                                                                                             11 
                            hn                                    n                                                                                                                 . . .
      Making use of the fact that the lateral components of all                                                                                                          L                                 U 
vectors k                                                                                                                                                     SN 1FN 1DN 1 SNFN DN .
             0n and khn coincide because they remain unchanged
at refraction and specular reflection, and substituting  see                                                                               Here Ev (E0 , 0, Es , Eh) and Dn (D0n1, D0n2, D0n3,
Fig. 1 a  :  0z   sin 0 ,  hz   sin h , k0zn  un , and                                                                                     D0n4) are the four-component vectors composed by un-
h                                          n
     zn   n   (1  az/a), we can rewrite equations  5  as                                                                                   known x-ray amplitudes  E0 , Es , and Eh are the amplitudes
                                                                                                                                           of the incident, specularly reflected and diffracted waves in
                     u2                                      n                        n                                                    vacuum; see Fig. 1 a  . Parameters S
                      n sin2 0  0 D0n   D
                                                                                      h¯         hn ,                                                                                                     v , Sn , and Fn denote
                                                                                            n                                              (4 4) characteristic matrices of layers:
                                                                                                                                     6 
               u                                                            n                    n
                    n  n 2 sin2 h  0 Dhn  h D
                                                                                                  n         0n .
      The condition  2                     2                                                                                                                   1 0 1 0
                                                                                                                                                                 0             1                     0                1        
                                 h  0 presuming the elastic scattering of                                                                             Sv                                                                           ,     12 
x rays, gives                                                                                                                                                 sin 0            0          sin 0                       0
                                                                                                                                                                 0         sin 
                           sin2                                                                                                                                                     h                0               sin h
                                           h   sin 0    2  ,                                                                         7 
where   (2 0h h2)/ 2 is the standard parameter for the
angular deviation of the incident wave from the Bragg con-                                                                                                                      1Vnj  
dition in the dynamical theory of diffraction.                                                                                                                        Snj                                     ,                          13 
                                                                                                                                                                                         u
      The values of u                                                                                                                                                                         n j
                                 n are determined by the dispersion equa-
tion which is the condition for the existence of a solution of                                                                                                                 Vnj unj   
Eq.  6 :                                                                                                                                   and F(U,L)
                                                                                                                                                 ni j      ijexp(iunj z(U,L)), where indices (L) and (U)
                                                                                                                                           indicate that the exponents are evaluated at the lower  upper 
      u 2                        n                                                                     n            n         n
        n  sin2 0  0    un  n 2 sin2 h  0      .
                                                                                                                    h¯        h            boundary of layer, (L)
                                                                                                                         n    n                                            n (U)n 1.
                                                                                                                                     8        The solution to Eq.  11  is straightforward:
Equation  8  is a fourth-order polynomial equation for un ,                                                                                      E         1S            1S                                1 S  U D
and has four solutions. As shown in Ref. 39, there are always                                                                                         v Sv       1Ft1S1        2Ft2 * * * SN 1 NFN                                 N ,  14 
two solutions corresponding to x-ray waves damping with                                                                                    where Ftnij  ijexp( iunj tn) and tn z(L) z(U). After cal-
z  Im(un) 0 , and two other solutions corresponding to the                                                                                 culating the matrix product in the right-hand side of Eq.  14 ,
waves growing with z  Im(un) 0 . The latter waves are                                                                                      we obtain four linear equations for four amplitudes: Es ,
usually treated as being specularly reflected from the lower                                                                               Eh , D0N1, and D0N2. The other amplitudes are given by
interfaces of the layers. For each of the solutions, Eq.  6                                                                                Eq. 11 . For the details of the numerical procedure, the
gives ( j 1, . . . ,4):                                                                                                                    reader is referred to Refs. 26­28.

                           u2  sin2                                    n                                                                   B. Amplitude of diffuse scattering from individual fluctuations
             D              n j                             0  0
                  hn j                      n                               D0nj VnjD0nj .                                           9                                 in interface positions
                                                h¯n                                                                                           Following the approach developed for x-ray reflection by
      The polarizabilities  n                                                                                                              rough surfaces,29­32 we describe rough interfaces between
                                                 h correspond to vectors h
                                                       n                                                             n which               layers as local fluctuations  z
vary from layer to layer. Therefore, it is more convenient to                                                                                                                            n( ) in the interface positions
                                                                                                                                           (  is the coordinate vector along interfaces . This approach
use the expansion over h  see, e.g., Ref. 40 :                                                                                             is obviously valid at length scales much greater than the
                                                                                                                                           interatomic distances. Then instead of  2  the polarizability
                            n                    n                               n
                                  h                         exp   h z .                                                             10 
                                      n           h                              z                                                         of the layers is given by



54                                                  DIFFUSE SCATTERING FROM INTERFACE ROUGHNESS . . .                                                                                        8153

                               N
                 rough r      n r H z zn  zn    
                              n 1

                               H zn 1  zn 1    z ,                                                15 
      Respectively, the fluctuations of polarizability can be
written as a sum over the interfaces,
                                                          N
             r   rough r   flat r     Pn r   n r ,  16 
                                                         n 1

where P(r) is the steplike function introduced by analogy
with Refs. 30 and 31:
                    1, z  zn,zn  zn    ,  zn    0
       Pn r    1, z  zn  zn   ,zn ,  zn    0
                       0                    elsewhere
                                                                                                  17 
and
                  n r    n                  n 1           n           n 1
                               0  0                     h  h  eih*r
                               n   n 1 e ih*r.                                                    18 
                                     h¯       h¯                                                            FIG. 2. Four different diagrams of x-ray scattering from a fluc-
                                                                                                         tuation of interface position  shaded area  in GID as given by the
      Thus the perturbation of the crystal polarizability due to                                         four terms in Eq.  25 . Coherent and diffuse scattered waves are
roughness possesses the form                                                                             shown by thick and thin vectors, respectively. Horizontal lines
                                                                                                         present the Bragg planes in crystal.
                       N
          n r     P                          n           n                    n
                             n  r     0   heih­r    e ih­r .                                                                              4          4
                   n 1                                                        h¯                                                   2                                  exp iQnijzn 
                                                                                                  19                     f n 4     Enij
                                                                                                                                        i 1 j 1                               iQnij
      Substituting Eq.  4  and Eq.  19  into Eq.  1  we also
assume29­32 that the wave fields do not change considerably                                                                          d2  eiQnij zn    1 eq* ,                                 24 
on the scale of the roughness. Then the wave fields of one of
two layers forming an interface  e.g., of the lower one  can                                             where it is denoted
be used at both sides of the interface, and we obtain the
scattering amplitude                                                                                                     E          in          out              n       in      out    n
                                                                                                                              ni j D0niD0n j  h DhniDhn j   h¯
          N             2 N 4 4                                 z                                                                              in         out          in       out     n
                                                                     n  zn                                                          D D  D D     ,                                            25 
      f     f                                                                                                                                  hni 0n j                0ni hn j         0
                 n                      d2                                    dz
         n 1           4   
                            n 1 i 1 j 1                         zn                                                                Q                        in          out
                                                                                                                                       ni j    uni un j     .                                 26 
          ei  in out                  in     out
                 0   0    i  uni un j  z Din                          in
                                                         0ni Dhnieih*r                                      The four terms in Eq.  25  can be treated as shown in
                                                                      out           out                  Fig. 2.
                  0   heih*r    h¯e ih*r   D0n j Dhn jeih*r .                                                i  Term 1 corresponds to the Bragg diffraction of x rays
                                                                                                  20     by the fluctuations of interfaces  Fig. 2 a  .
      We are interested in diffuse scattering about the direction                                            ii  Term 2 corresponds to the triple Bragg diffraction:
of the grazing diffracted beam. Therefore,  out                                            in            the incident wave is diffracted by ML, then diffracted at
                                                                             0   ( 0  h),
and it is convenient to write                                                                             h in the fluctuations, and diffracted by the multilayer again
                                                                                                          Fig. 2 b  .
                                     in       in                                                             iii  Term 3 corresponds to the Bragg diffraction of the
                                     0   0Br qin,                                                 21     incident wave by the multilayer, and the small-angle scatter-
                             out             in                                                          ing of the diffracted wave by the fluctuations  Fig. 2 c  .
                             0    0Br h qout,                                                     22         iv  Term 4 corresponds to the small-angle scattering of
                                     q qin qout,                                                  23     the incident wave by the fluctuations and the Bragg diffrac-
                                                                                                         tion of the scattered wave by the multilayer  Fig. 2 d  .
where  in0Br is the vector in the incidence direction, exactly                                              One can see that, in all four cases, x rays are scattered
satisfying the Bragg condition. The terms in Eq.  20  con-                                               through the large angle 2 B due to the diffraction from the
taining exp( ih*r) oscillate with   at an atomic scale and                                               Bragg planes. That is, the momentum transfer due to scatter-
can be neglected. Then, after carrying out the integration                                               ing from roughness is small and the intensity of DS must be
over z, Eq.  20  is transformed to                                                                       much greater than that in Refs. 41 and 42. In those cases,



8154                                                                                S. A. STEPANOV et al.                                                              54

x-ray scattering through large angles corresponding to large
momentum transfers was measured from amorphous multi-
layers with rough interfaces in a geometry formally similar
to GID.

                 C. Cross section of diffuse scattering
                        from statistical roughness
   Proceeding from Eq.  24  to the cross section of diffuse
scattering, we obtain

        d                      4         N               4          4
                                                                               *
        d     f 2  16 2                                     EnijEn i j 
                                    n,n  1 i,i  1 j,j  1
                 ei Q          *
                         ni j Qn i j  zn Q                    *
                                                        ni jQn i j   1

                    d2   d2  eiq      
                  eiQ                         *
                         ni j zn    iQ                   
                                              n i j  zn     1 ,                             27 

where   . . .   denotes averaging over random functions
 zn( ) assumed to be Gaussian.
   The application of the general formula43  exp( j jxj) 
 exp(                                                                                                 FIG. 3. Different degrees of collimation in the scattering plane
           jk j k xjxk /2), where   j are constants and x j are
Gaussian random variables, to Eq.  27  gives                                                       of grazing-incidence x-ray diffraction:  a ,  b , and  c  show single-,
                                                                                                   double-, and triple-crystal experimental schemes, respectively. Co-
                                                                                                   herent and diffuse scattered waves are shown by thick and thin
   d              4      N          4              4
                                                                          *         *              vectors, respectively. The incident and scattered beams are colli-
   d   S 16 2                           CnijEnijCn i j En i j                                      mated and analyzed in a direction normal to the plane of the figure.
                       n,n  1 i,i  1 j,j  1

                  d2  eQ *                                                                         ( j 1 and 2 . This procedure can be called ``the specular-
                              ni jQn i j Knn     1 eq* ,                                    28     reflection approximation.'' Below, we show that it has more
                                                                                                   applications.
where C                                  2 2
              ni j exp(iQnijzn  nQnij/2)/Qni j ,  n is the rms
height of roughness,  2                                       2
                                              n   zn(0) ,                 and Knn ( )
   z                                                                                                        III. INTEGRATED DIFFUSE SCATTERING
        n(0) zn ( )  is a correlation function. The same cor-
relation functions proposed for diffuse scattering in x-ray                                               IN DIFFERENT TYPES OF GID EXPERIMENT
TER should be valid here.29,31,32,44,45                                                               In the majority of cases the incidence and the exit angles
   Thus expression  28  for diffuse scattering in GID is for-                                      of diffracted beam are controlled in GID experiments be-
mally similar to that in TER  compare Refs. 30­32 and 44­                                          cause these angles determine the penetration depth of GID
47 . However, there are two essential differences. The physi-                                      inside the samples. However, the angles in the Bragg plane
cal difference is that diffuse scattering in TER originates                                        are not always controlled  see Fig. 3 , and that provides
from   0  the fluctuations of the mean target density , while                                      some averaging of the pattern and simplification of Eq.  28 .
in GID it is mainly due to   h  the crystal structure of the                                          The general formula  28  is applicable in the case of
fluctuations . That means that diffuse scattering in GID pro-                                      triple-crystal measurements only, when the incident beam is
vides information about the degree of crystal structure per-                                       collimated and the scattered beam is analyzed in two planes
fection at interfaces. The mathematical difference consists of                                      Fig. 3 c  . That would be the most informative case, but the
the dependence of Enij and En i j  on q, which is not the case                                     low intensity of GID might impose serious experimental
in TER. Hence the DS pattern is far more complicated in                                            limitations.
GID than in TER, and depends on four diffraction angles: the
angles of the incident and the scattered x-ray waves with
respect to the surface and their deviations from the Bragg                                                 A. Single-crystal scheme: No angular resolution
angle in the surface plane. An analytical integration of Eq.                                                               in the Bragg plane
 28  over one of the components of q, like that used to re-                                           Some authors48,27,12 have performed measurements where
duce the calculations in the case of diffuse scattering in                                         the incident beam is collimated in  0 , but not in  , and the
specular reflection, is thus not possible.                                                         diffracted waves are separated over their exit angles by a slit
   One simplification occurs when DS is studied far from the                                       or a position-sensitive detector  PSD . This single-crystal
direction of propagation of the coherent diffracted beam                                           scheme49 is based on Eq.  7 , where    2sin(2 B)(   B)
where the Bragg diffraction of waves scattered by roughness                                         10 5 is proportional to the deviation of the in-plane angle
can be neglected. Then Dout                                        out         out2       n
                                     hn j 0 and un j   ( 0                               0)1/2,      from the kinematic Bragg angle  B . The large width of
giving Dout
              0n j as solutions to the specular reflection problem                                 Bragg peaks for  0 and  h    10 2­10 3 makes these



54                                                                   DIFFUSE SCATTERING FROM INTERFACE ROUGHNESS . . .                                                                                                                  8155

measurements very convenient. However, the coherent re-                                                                                   in the Bragg plane, but the acceptance of the scattered beam
flection and diffuse scattered radiation are all counted to-                                                                              is limited by an analyzer crystal.
gether  see Fig. 3 a  .
      If neither the in-plane angle of the incident or diffracted                                                                                    IV. ANALYSIS OF A GID EXPERIMENT
beam is collimated, then waves with large Bragg deviations
dominate in the incident and scattered fans, and the solutions                                                                                             WITH AN ETCHED Ge SURFACE
to the specular reflection problem can be used for both                                                                                      The theory given in Secs. II and III has been applied to
Ein(r) and Eout(r). Thus the dynamical diffraction problem                                                                                interpreting the results of the experiment in Ref. 15. In this
of GID need not be considered for the calculations of DS at                                                                               study, carried out at the Cornell High-Energy Synchrotron
all. The experiment integrates over both the in-plane compo-                                                                              Source  CHESS , the grazing-incidence diffraction was mea-
nents of q, providing  (x) (y)/ 2 in integral  28 . Thus one                                                                              sured from  220  Ge planes at   1.55 Å. A double-crystal
finds                                                                                                                                     scheme of measurements was used. The incident beam was
                                                                                                                                          collimated in the incidence angle  
         d                              N                2             2                                                                                                                                   0 within 0.25 mrad, and
               1            2                                                                      n in                    out            in the diffraction angle within 0.0014 mrad. The sample was
         d   S16 2                                          Cnij  hD0niD0nj
                                  n,n  1 i,i  1 j,j  1                                                                                    rocked through the Bragg angle  B at fixed  0   4 mrad,
                                                                                                                                          and the entire scattered intensity was collected over the take-
                            C                          n  in                out
                                 n i j   h D0n i D0n j  *                                                                                 off angle  h and over the in-plane exit angle.
                                                                                                                                             The sample surface consisted of two different parts: a
                            eQ          *
                                  ni jQn i j Knn  0  1 .                                                                           29     high-quality polished part and an etched quadrant provided a
As follows from Eq.  29 , the DS measured in the single-                                                                                  spectrum of surface roughness. The measured GID curves
crystal scheme is completely determined by K                                                                                              for these two parts are presented in Figs. 4 a  and 4 b . The
                                                                                                         nn (0). In the
case where the roughness of different interfaces is com-                                                                                  curve taken from a smooth surface coincides well with the
                                                                                                                                          theoretical calculations for the perfect case. A peculiarity of
pletely uncorrelated, K                                                2
                                             nn (0)  n nn , the measurements of                                                           both the experimental and theoretical curves in the perfect
integral DS provide the rms roughness height.                                                                                             case is the zero reflection coefficient to the left of the Bragg
      In the case of small completely correlated roughness,                                                                               peak, where the diffracted wave becomes surface trapped
where Knn (0)  n n  and the exponent in  29  can be ex-                                                                                    the angle  
panded (                                                                                                                                                   h becomes imaginary due to Eq. 7  . Contrary to
                    nQni j 1), formula  29  for the diffuse scattering                                                                    the smooth surface case, the experimental curve for a rough
becomes very similar to that for the intensity of coherent                                                                                surface exhibits two nearly symmetrical shoulders at both
GID calculated in the DWBA  see, e.g., Ref. 22 :                                                                                          sides of the Bragg peak. These shoulders are obviously due
           d                                 N          2                                                                                 to diffuse scattering at surface roughness, because the effect
                     1       S                                                        n in              out                               of roughness on the coherent beam would appear to be a
               d   16 2      n   hD0niD0nj
                                        n 1 i, j 1                                                                                        Debye-Waller attenuation of the intensity on the wings.21
                                                                                                                                             The diffuse scattering has been calculated with the help of
                            ei  uin out                                     2              in    out
                                        ni un j    zn  n 2 uni un j    2 2.  30  Eq.  31 . In the case of only one interface and   0, it is
                                                                                                                                          greatly simplified:
      The expression for GID differs from  30  by the substitu-                                                                                                                           2                    2 Q*2 /2
tion of                                                         n                                                                             d 2                3                                  e  2 Qi         j
                    h instead of  n                                                                                                                                           out                                            in in
                                                                h . The consequences of this                                                                                     2                                               D *
analogy are discussed in Sec. V, where we give some nu-                                                                                        d   S 16 2  D0                                                                      j
                                                                                                                                                                                     i, j 1                QiQj*           Di
merical examples.                                                                                                                                                 
                                                                                                                                                          dx eQiQ*jK x  1 eqxx,                                                          32 
         B. Double-crystal scheme: Partial angular resolution                                                                                                     
                                        in the Bragg plane                                                                                where Dout 2sin out/(sin out uout), Q                                                  in uout),
      In some more advanced  double-crystal  experiments, the                                                                                         0                  0                     0                         i  (ui
                                                                                                                                          uout (sin2 out                                       in are the two solutions to the
incident beam is collimated in the Bragg plane, while the                                                                                                  0          0)1/2, and ui                  in
entire in-plane spread of the scattered beam is accepted  Fig.                                                                            dispersion equation  8  with Im(ui ) 0.
3 b  . Then the ``specular reflection'' approximation can be                                                                                 Equation  32  was integrated over  h , and renormalized
applied to Eout(r) and, additionally, one can average Eq.  28                                                                             to the reflectivity  see Eq.  2.13  in Ref. 29 :
over the components of q normal to  h . This procedure re-                                                                                                                           1               d 
duces the integral in  28  to a one-dimensional one,                                                                                                              R 2                                 2
                                                                                                                                                                      DS Ssin 0 0 d  d h .                                               33 
           d                                      N           4                  2
                    2             3                                                                      in         out
            d   S 16 2                                            CnijDniD0nj                                                             The correlation function was chosen in the Gaussian form:
                                        n,n  1 i,i  1 j,j  1                                                                              K( )  2exp(  2/ 2), where   is the rms height and   is
                             C                         in            out                                                                  the lateral correlation length of roughness. The integration in
                                   n i j Dn i D0n j  *                                                                                     33  was carried out numerically.
                                                                                                                                             The normalized calculated flux of DS for different   is
                               dx eQ *
                                                         ni jQn i j Knn  x  1 eqxx,                                                31     presented in Fig. 4 c . The shape of DS curves strongly de-
                                                                                                                                          pends on the lateral correlation length of roughness: at
where q                                                                               n                        n
               x q* h / , and Dni D0ni  h Dhni  0 The same                                                                                greater correlation lengths DS is concentrated closer to the
situation takes place when the incident beam is uncollimated                                                                              coherent diffracted beam of GID. The dependence of calcu-



8156                                                         S. A. STEPANOV et al.                                                                                    54

                                                                            lated DS on   is presented on Fig. 4 d . The intensity of DS
                                                                            quickly grows with  , while the shape of the curves is prac-
                                                                            tically independent of   in a wide range up to   10 Å. At
                                                                              10 Å, the DWBA starts to diverge at    B 0, where
                                                                            the diffracted wave of GID is surface trapped. This diver-
                                                                            gence is due to a small penetration depth of surface-trapped
                                                                            x-rays  see Fig. 4 e  . The DWBA breaks down when the
                                                                            x-ray wave fields undergo large changes on the scale of the
                                                                            height of the roughness.
                                                                               The roughness height for the case in Fig. 4 b  measured
                                                                            with a 3- m profilometer tip was   200 Å. This value was
                                                                            obviously beyond the applicability of the DWBA. In particu-
                                                                            lar, the DWBA fails to explain the attenuation of maximum
                                                                            reflectivity at the Bragg peak in Fig. 4 b . However, taking
                                                                            into account the weak dependence of the shape of DS curves
                                                                            on  , we can fit the shape of the curves at small  , and then
                                                                            extrapolate the data to higher   where the DWBA diverges.
                                                                               The fitting procedure was carried out in two steps: first,
                                                                            the shape of the DS curve was fitted at    B 0 where the
                                                                            coherent reflection is zero  see Fig. 4 a  . Then the calculated
                                                                            DS was added to the calculated coherent reflectivity attenu-
                                                                            ated by some empirical factor ch in order to fit the maximum
                                                                            of the reflection coefficient:  R 2                                       2          2
                                                                                                                                  total ch R coherent  R DS .
                                                                            The factor ch was introduced to account for the relative con-
                                                                            tributions of coherent diffracted and scattered radiation when
                                                                            the roughness was great.
                                                                               The parameters of the fit presented in Fig. 4 b  are
                                                                              1600 Å and ch 0.4. The roughness rms height fitted at
                                                                               B 0, where the DWBA does not diverge is   43 Å.
                                                                            This value is consistent with the profilometer data, since the
                                                                            rms roughness at   1600 Å need not be as great as the rms
                                                                            roughness at   30 000 Å corresponding to the horizontal
                                                                            resolution limit of the profilometer. The long-wavelength
                                                                            roughness measured with the profilometer could not cause
                                                                            the DS on the tails of the curve Fig. 4 b  because the DS
                                                                            corresponding to the long-wavelength roughness is strongly
                                                                            concentrated near the Bragg peak. This roughness was prob-
                                                                            ably responsible for the broadening of the experimental
                                                                            Bragg peak in Fig. 4 b . If, on the other hand, we consider
                                                                            the possibility that etching causes the crystal structure disor-
                                                                            dering, then, the parameter   h in Eq.  32  would be re-
                                                                            duced by a static Debye-Waller factor, and the same intensity
                                                                            of DS would be achieved with a greater  .
                                                                               Both the experiment and theory show a dip in the DS
                                                                            pattern at the Bragg peak. In Ref. 15 this dip was supposed
                                                                            to be due to a cutoff of the maximum roughness wavelength
                                                                              to which the experiment should be sensitive because of the
                                                                            limited coherence length of the diffracted beam ( 3  m .
                                                                            The theory presented here contains no assumptions on the
                                                                            coherence length of the source, and the interpretation of this
   FIG. 4. Fitting of GID data measured in Ref. 15  Ge,  220                effect is different. As follows from Eqs.  31  and  32 , the
Bragg planes,   1.55 Å, and  0   4 mrad . Data are represented              intensity of DS in the double-crystal scheme is approxi-
by dots, and fits by solid lines.  a  Smooth surface  fit assumes GID       mately proportional to the total intensity of the x-ray wave
with no diffuse scattering .  b  Rough surface  fit is GID plus DS          field illuminating the interface:
calculated with   1600 Å and   43 Å .  c  Calculated shape of
DS curves vs correlation length of roughness  the dashed line indi-
cates the total x-ray intensity at crystal surface .  d  Intensity of DS           d                 2                              2
                                                                                         2                  in in
vs height of roughness  the DWBA diverges at                                                                         *                         in          in
                                                              B 0 for               d      Di Di      0   D0i Dhi  2
                                                                                              i,i
  20, 30, and 40 Å .  e  Penetration depth inside the crystal cal-                                     1                           i 1
culated for two wave fields Din1,2 of GID at  0   4 mrad  the
decrease in penetration depth at    B 0 causes the divergence of                                                                         in     in         in
                                                                                                                                  0  E0  Es  Eh   2.
the DWBA in  d  .                                                                                                                                                 34 



54                                  DIFFUSE SCATTERING FROM INTERFACE ROUGHNESS . . .                                             8157

      The right side of Eq.  34  exhibits a dip at    220 , pro-
viding the minimum in DS  see the dashed line in Fig. 4 c  .
Thus, the observed effect has the same origin as the Yoneda
peaks50,29 in x-ray DS and the secondary emission yield ap-
pearing near the critical angle for TER. Yoneda peaks are
due to the enhancement of x-ray intensity at the surface near
the critical angle for total external reflection. Here we have a
dip in intensity at the atomic planes near the Bragg angle,
which is matched in this instance with the threshold angle for
total internal reflection  with the angle  , where  h becomes
an imaginary quantity due to Eq.  7  and the diffracted wave
becomes surface trapped . This dip provides a minimum in
DS in Fig. 4 b  as well as a minimum in the fluorescence
yield from lattice sited atoms which was observed in GID
standing wave experiment.51,52 Therefore, our case can be
referred to as an ``anti-Yoneda effect.''
      At   0, the critical angle for total internal reflection
given by Eq.  7  may not coincide with the Bragg angle.
Then two dips in DS are predicted by our theory, the dip at
the critical angle for total internal reflection being stronger
than that at the Bragg angle.                                          FIG. 5. Calculated x-ray reflectivity curves vs the takeoff angle
                                                                    for the single-crystal scheme of GID. The curves are for a 20-period
                                                                    AlAs/GaAs superlattice (t
         V. DIFFUSE SCATTERING IN MULTILAYERS                                                    AlAs 125 Å and tGaAs 95 Å  on a  001 
                                                                    GaAs substrate. The parameters of the calculations are as follows:
      In the case of multiple and periodic rough interfaces, the     220  reflection,   1.5 Å;  0 0.3°,   5 Å, and   2000 Å.
effects of roughness become much more prominent. We have            Curves 1 and 2 show DS for uncorrelated ( z 0) and completely
carried out calculations for a sample similar to that discussed     correlated ( z  ) roughnesses of interfaces, respectively. Curves 3
in Refs. 27 and 31: an AlAs/GaAs superlattice consisting of         and 4 present the coherent GID reflection for flat and rough inter-
20 periods of 125 Å AlAs and 95 Å GaAs on a  001 -                  faces, respectively.
oriented GaAs substrate. The calculations assumed a  220 
Bragg reflection of   polarized incident x rays with                GID measurements is due to fluctuations in the crystal struc-
  1.5 Å and                                                         ture   h , while the parameters of roughness given by x-ray
                      0 0.3°. The correlation function was cho-
sen in the form suggested by Ming et al.,44                         reflectivity refer to the fluctuations of material density
                                                                      0 . In the case where the crystal structure is destroyed by
              K                                                     roughness, the data of TER and GID may disagree: the latter
                   nn      n n e   /  2e  zn zn  / z,       35      may exhibit a reduced intensity corresponding to a smaller
with a rms roughness height  n 5 Å and a lateral correla-            . Then, to isolate the diffuse component of GID, one has to
tion length   2000 Å.                                               select scattered radiation in double- or triple-crystal measure-
      The calculated intensity of DS for the single-crystal case    ments.
of GID renormalized to a reflectivity as in Eq.  33  is pre-           A comparison between the diffuse scattering in TER and
sented in Fig. 5 as a function of exit angle  h . Curves 1 and      GID can be used for investigating the crystal structure of
2 correspond to the uncorrelated ( z 0) and completely cor-         rough interfaces. In our model, one can add a Debye-Waller
related ( z  ) roughness of interfaces, respectively. One           factor for   h and   h¯ in Eq.  25  describing the attenuation
can see that in both the cases the curves exhibit multilayer        of polarizabilities due to disordering of the crystal structure
Bragg peaks at the same angular positions as the peaks of           at rough interfaces. The value of this Debye-Waller factor
coherent GID  curves 3 and 4 . Therefore, it is difficult to        can be found from the difference in   given by TER and
separate the diffracted and diffuse intensities. In the case of     GID. Thus the diffuse scattering in GID can deliver a mea-
correlated roughness, the DS about the multilayer Bragg             sure of crystal structure ordering at rough interfaces. This
peaks becomes much more prominent, and the intensity at             information is not accessible by conventional x-ray scatter-
large  h is comparable to the intensity of coherent GID. The        ing techniques.
shape of the DS curve is very similar to that of coherent              The calculations for the double-crystal scheme of GID are
GID, as given by  30  and well seen in Fig. 5, but the de-          presented in Fig. 6 for noncorrelated  a  and completely cor-
crease in DS intensity with  h is slower. These results may         related  b  interface roughness in multilayers. Due to the
explain the discrepancies between the theory and experiment         in-plane angular collimation of the incident x rays, for each
at high angles observed in the single-crystal measurements             B the diffracted wave exits the crystal at the certain
of GID in multilayers in Ref. 27.                                   angle  h , as given by Eq. 7 . The exit angle of the diffracted
      To distinguish the effect of DS on single-crystal GID         wave at different    B is traced on the maps by thick solid
curves, one might measure the parameters of roughness by            stripes. The intensity at all other points on the maps can be
x-ray reflectivity and diffuse scattering in TER, substitute the    attributed to DS. Thus a separation of coherent and diffuse
roughness value into Eq.  29 , and subtract the calculated DS       scattering is possible.
from the measured GID curves. However, as follows from                 The maps in Fig. 6 clearly show the bunching of the DS
Eq.  29  and Fig. 2, the DS contributing to the single-crystal      into resonance diffraction sheets  RDS's  for correlated in-



8158                                                     S. A. STEPANOV et al.                                                      54

                                                                         h can exit the crystal at each    B . Diffuse and coherent
                                                                        scattering could be separated out in this case by means of the
                                                                        triple-crystal scheme. The calculations of DS for relaxed su-
                                                                        perlattices could be performed using Eq.  1  and the wave
                                                                        fields of GID found in Ref. 53.

                                                                            VI. MEASUREMENTS OF DIFFUSE SCATTERING
                                                                                     FROM AlAs/GaAs MULTILAYER
                                                                           In order to provide a comprehensive test of our theory, we
                                                                        have undertaken high-resolution measurements of GID from
                                                                        a 20-period AlAs/GaAs superlattice  SL . The GID experi-
                                                                        ment has been carried out at the optics beamline BL10 of
                                                                        ESRF.
                                                                           The superlattice grown by molecular-beam epitaxy on a
                                                                         001  GaAs substrate was characterized in the laboratory by
                                                                        x-ray Bragg diffraction, x-ray specular reflection, and non-
                                                                        specular x-ray scattering. A  -2  scan near the  004  Bragg
                                                                        peak did not reveal any strain relaxation, indicating that the
                                                                        superlattice possessed a laterally matched crystal structure.
                                                                        The thickness of the layers obtained from the fitting of this
                                                                        scan was tAlAs (154 1) Å, and tGaAs (73 1) Å. The
                                                                        x-ray-diffraction data also indicated a sample surface miscut
                                                                        of ( 0.38 0.02)° along  110 . The x-ray specular reflec-
                                                                        tion  -2  scans confirmed the thickness of AlAs and GaAs
                                                                        layers and revealed a surface transition layer with a thickness
   FIG. 6. Calculated maps of diffuse scattering for the double-        of (18 2) Å, probably due to natural oxidation. A fit to the
crystal scheme of GID experiments. The parameters of the calcula-       specular reflection curve gave the rms height of interface
tions are the same as in Fig. 5. The equal-intensity map  a  is for     roughness  i (4 0.5) Å. Finally, the longitudinal and
uncorrelated interface roughness, and  b  is the same for completely    transverse scans of nonspecular x-ray scattering at grazing
correlated roughness. Thick stripes marked by arrows show the po-       incidence ( -2  scans with   offsets of the sample and  
sition of the coherent wave of GID.                                     scans at fixed 2  positions of the detector, respectively  were
                                                                        carried out in order to determine the vertical and lateral cor-
terface roughness. This effect is completely analogous to the           relation lengths of roughness. The longitudinal scans exhib-
formation of RDS's  ``Holy bananas''  in DS during x-ray                ited very sharp RDS maxima corresponding to a complete
specular reflection.31 The vertical black fringe at   0 cor-            correlation between the roughness of different interfaces.
responds to the anti-Yoneda minimum in DS which we have                 Therefore, the data were fitted with the simple correlation
discussed above for scattering from a rough surface.                    function  35  assuming  z  . The transverse scan along
   Another interesting peculiarity displayed by Fig. 6 b  is            RDS-8 (2  3.172°) and the corresponding fit are shown in
the appearance of RDS's at negative    B where the dif-                 Fig. 7. The fit consists of two independent contributions of
fracted wave of GID is surface trapped and cannot exit the              interface and surface roughness, which explain the central
crystal. This effect is surprising because the surface-trapped          part of the curve and the Yoneda peaks at the wings, respec-
wave has a small penetration depth inside the crystal  see              tively. The fitted parameters are   4 Å,   3000 Å for the
Fig. 4 e  , and one could expect only a few interfaces con-             interface roughness, and   9 Å,   500 Å for the surface
tributing to DS at these angles. However, two types of x-ray            roughness. The height of the surface roughness corresponds
wave fields are generally excited in the crystal under GID:             to half of the thickness of the surface transition layer found
one wave can be roughly connected to the diffracted wave in             by reflectometry. The lower-order RDS indicated a more
vacuum, and the other one to the incident wave. The angle of            complicated spectrum of roughness. For example, the trans-
the latter wave to the surface is not small in our example,             verse scan at RDS-5 (2  2.033°) was better explained by a
thus providing a greater penetration depth. This effect can be          combination of the same surface roughness and a sum of
used for the experimental measurements of DS because the                interface roughness with   4 Å,   4500 Å,   3 Å,
separation of the diffracted beam is unnecessary. We note a               3000 Å and   2Å,   1500 Å. However, in the first
small difference in the positions of RDS's at positive and              approximation the parameters   4 Å,   3000 Å found at
negative    B .                                                         RDS-8 were applied to modeling the results of the GID ex-
   As long as the coherent and diffuse scattering can be dis-           periment.
criminated with the double-crystal scheme of the GID ex-                   The experimental configuration of the GID experiment is
periment, the triple-crystal scheme  Fig. 3 c   is not of par-          shown in Fig. 8. The grazing-incidence diffraction was mea-
ticular interest. The situation might change in the case of             sured from the  220  AlAs/GaAs planes in the double-crystal
relaxed superlattices containing a distribution of lattice spac-        scheme corresponding to Fig. 3 b , and similar to the experi-
ings along the lateral direction. In this case, Eq.  7  becomes         ment at CHESS.15 However, in contrast to Ref. 15, the an-
inapplicable,53 and a fan of diffracted waves with different            gular spectrum of the diffracted beam was analyzed with a



54                                  DIFFUSE SCATTERING FROM INTERFACE ROUGHNESS . . .                                                   8159

















      FIG. 7. Transverse scan through the eighth-order resonance
sheet (2  3.172°) of x-ray diffuse scattering taken from an AlAs/
GaAs superlattice far from diffraction conditions. Experimental             FIG. 9. Double-crystal rocking curve of GID taken from the
data are represented by circles, and the theoretical fit by the solid    AlAs/GaAs superlattice at  0 0.5°. Dots represent experimental
line. The fit is the sum of contributions of completely correlated       data, and the solid line is a theoretical simulation convoluted with
interface roughness (  4 Å,   3000 Å  and surface roughness              the rocking curve of the five-reflection monochromator. The top
(  9Å,   500 Å .                                                         line is the magnified reflection coefficient for the specular x-ray
                                                                         beam. Vertical marks indicate the sample positions where PSD
                                                                         spectra of diffracted intensity were taken  see Fig. 11 .
linear PSD, and the experimental setup was optimized in
order to provide a good separation of GID and DS.                        from a Ge wafer: the width of the Bragg peak corresponded
      A Si 111  double-crystal monochromator was tuned to                to the calculated value. The angular resolution of the PSD
  1.40 Å. The crystals in the monochromator were slightly                over the takeoff angle was 15 sec of arc.
offset in angle to suppress the third and higher harmonics.54               The front of the beam impinging on the sample was re-
The divergence of the x-ray beam in the vertical plane over              stricted by the output slits S3 with vertical and horizontal
 0 was mainly determined by the monochromator since the                  sizes of 0.04 and 0.2 mm, respectively. This provided an
vertical divergence of synchrotron radiation at ESRF is  1               illuminated area smaller than the sample size, and eliminated
sec of arc. The collimation of the beam in the horizontal                edge reflections. The primary and the secondary slits S1 and
plane over   was provided by five  220  reflections in a                 S2 restricted the beam front to 1 2 mm2 and 0.1 1
channel-cut Si crystal. The five-reflection collimator was               mm2, respectively, and reduced the background in the ex-
used to suppress the tails of the x-ray beam and provide a               perimental hutch.
better discrimination of GID and DS at the position-sensitive               In the first step of the experiment, the PSD was replaced
detector.                                                                by a scintillation counter. The diffracted beam was recorded
      The estimated parameters of the incident beam at the               while scanning   as in Ref. 15 with no separation of the
sample were:   /  1.2 10 4,   0 6 sec of arc, and                        coherent and diffuse components. A second counter  Si pho-
   4 sec of arc. The last parameter is given with account                todiode  simultaneously recorded the rocking curves of the
of the dispersion effect caused by the difference in the Bragg           specularly reflected x-ray beam. Figure 9 presents the mea-
angles of Si and GaAs  220  reflections. The horizontal di-              sured and calculated rocking curves for the incidence angle
vergence was examined by recording a  220  rocking curve                  0 0.5°. The parameters for the calculations are taken from
                                                                         the laboratory data presented at the beginning of this section
                                                                         and the algorithm is described in Sec. II and elsewhere.21
                                                                         The experiment clearly demonstrates the high quality of the
                                                                         superlattice:  i  the half-widths of the Bragg peak and the
                                                                         superlattice peaks correspond to the calculated parameters;
                                                                         and  ii  the rocking curve of the specular beam exhibits a
                                                                         maximum at the first superlattice Bragg peak, thus proving
                                                                         that the diffraction is dynamical.
                                                                            At the same time, the reflectivity at the wings of measured
                                                                         GID curve is noticeably higher than expected and is probably
                                                                         due to DS. Figure 10 shows the map of DS calculated for the
      FIG. 8. The scheme of the high-resolution GID experiment car-
ried out on the optical beamline of ESRF. The linear position-           experimental conditions according to  31 . In contrast to the
sensitive detector  PSD  provided an angular analysis of radiation       example given in Fig. 6 b , the DS is concentrated along the
scattered along the diffracted beam of GID. The intensity of the         diffracted beam of GID because of a greater lateral correla-
specular x-ray beam was recorded integrally using a Si photodiode.       tion length. However, a characteristic pattern of DS with
For the integral GID measurements presented in Fig. 9, the PSD           periodic superlattice peaks in two directions is well devel-
was replaced by a scintillation counter.                                 oped. The SL peaks parallel to the  -axis are especially



8160                                                         S. A. STEPANOV et al.                                                         54
















   FIG. 10. Calculated map of diffuse scattering in GID
( 0 0.5°) for the AlAs/GaAs superlattice studied at ESRF. The
thick stripe marked by the arrow shows the position of the coherent
wave of GID. The theoretical curves in Fig. 11 are the sections of
this map drawn along the vertical axis at different    B .

strong, and should be clearly seen in PSD observations
which cut the map along the  h axis.
   The PSD spectra have been taken at different
 -positions of the sample as marked in Fig. 9. The results are
shown in Fig. 11. The curves in the figure are vertically
shifted with respect to each other as explained in the figure
caption. The experiment is compared to the theoretical
curves, which consist of four parts

           I                                    IF
                Total   h ,     IGID   h ,     IDS   h ,    

                                ISurf
                                  DS    h ,     IBackgr .         36 

Here IGID( h ,  ) RGID  ( h) R5  ( h)   (  )] is
the product of x-ray Bragg reflections from the sample and
the five-reflection collimator, respectively;  ( h) is the
x-ray Bragg deviation for the sample given by Eq.  7 , and
  (  )  2 sin(2 B)   is the difference in   for collima-
tor and sample due their angular misalignment    corre-
sponding to a given PSD spectrum. The terms IIF                   Surf
                                                        DS and IDS
are the DS intensities for the interface and surface roughness,
respectively, calculated according to Eq.  31  with the pa-
rameters found from the laboratory measurements. Finally,
IBackgr is a constant term equal to the experimental back-
ground.
   Different terms in Eq.  36  explain different peculiarities
of the experimental spectra in Fig. 11. The coherent reflec-
tion provides the floating peaks marked by arrows. The po-
sition of these peaks is given by Eq.  7  with     (  ).
The strongest effect of coherent reflection is observed at
   3 arc sec, where two coherent peaks are found. At
   0, the contribution of coherent reflection is small and
invisible on the spectra.
   DS due to interface roughness forms regular superlattice                  FIG. 11. PSD spectra of GID intensity vs exit angle at
peaks, with a maximum intensity around the position of the                 0 0.5° and different    B for the AlAs/GaAs superlattice. Dots
coherent reflection, as expected from the map presented in                present experimental data, and solid lines are calculations with pa-
Fig. 10. At    0 and    65 arc sec, the integrated inten-                 rameters taken from the laboratory studies of the sample. Arrows
sity of DS peaks exceeds the integrated intensity of the dif-             mark coherent peaks. The curves are successively shifted by 103 for
fraction peak itself, thus explaining the high-intensity wings            clarity. The central curve is then reduced by a factor of 20 because
of the experimental curve in Fig. 9. It is worth noting that su-          of its high intensity.



54                                        DIFFUSE SCATTERING FROM INTERFACE ROUGHNESS . . .                                              8161

perlattice peaks are observed for    0. This proves that the            ness in crystals can give rise to x-ray diffuse scattering about
x-ray wave field in GID possesses greater penetration inside            the diffraction beams of GID. The theoretical model is based
the superlattice, even though the diffracted wave is surface            on the distorted-wave Born approximation29 and the dynami-
trapped.                                                                cal theory of GID by multilayers.26­28 The model can be
      Finally, the surface roughness provides a maximum in DS           applied to asymmetric x-ray diffraction and to the diffraction
near the critical angle of TER  at                                      of neutrons. Expressions have been obtained for GID experi-
                                               h 0.3°) for the data
corresponding to    169 and 313 arc sec, and    0.                      ments in different geometries. The effect of interface-
Here we have found that the intensity of DS calculated due              interface roughness correlations on x-ray diffuse scattering in
to the surface roughness with   9 Å was a factor of 4                   GID has been taken into account. It has been demonstrated
greater than observed in the experiment. The theoretical                that in the case of periodic interfaces, interface-interface cor-
curves presented in Fig. 11 correspond to   4.5 Å, half of              relations give rise to the formation of resonance sheets in DS
the value determined from the laboratory measurements. We               similar to ``Holy bananas''31 in small-angle x-ray scattering.
suggest that the difference in the observed   values is due to             When applied to the analysis of GID data taken from a Ge
the distinction between diffuse scattering in TER and GID:              crystal etched to provide a roughened surface, the theory
the diffuse scattering in TER is proportional to   2                    obtains the observed diffracted intensity and diffuse scatter-
                                                           0 , while    ing. The intensity shoulder on the experimental curve at
that in GID is mainly given by   2h , a measure of the atomic              
ordering in roughness. The crystal structure at the surface                    B 0, where the GID intensity must otherwise be zero,
                                                                        has been explained by diffuse scattering from the surface
could be partially destroyed by the oxidation, giving rise to           roughness. A dip in the diffuse scattering near the Bragg
weaker DS in GID. Thus our experiment confirms the con-                 peak can be attributed to an ``anti-Yoneda'' effect.
clusion derived in Sec. V that the measurements of DS in                   High-resolution measurements of GID have been re-
GID may provide information on atomic ordering at inter-                ported. This approach allowed the diffracted flux from an
faces which is not accessible by DS in TER.                             AlAs/GaAs superlattice to be resolved into GID and diffuse
      While the agreement between our theory and experiment             scattering due to correlated interface roughness. The experi-
is reasonable, some discrepancies can be attributed to the              mental results are in good agreement with the theory.
simplified model of the correlation function used in the cal-           Measurements of diffuse scattering in GID are sensitive to
culations. It is worth noting that no diffuse scattering due to         atomic ordering in roughness, thereby providing information
crystal structure defects in the superlattice was observed. As          which is not accessible by conventional small-angle x-ray
shown in Refs. 55 and 56, a peak of DS due to pointlike                 scattering.
defects could be expected at  h   0 0.24°. Some discrep-
ancies between the theory and experiment at small  h near                                    ACKNOWLEDGMENTS
the critical angle of TER might be due to this kind of scat-
tering, but the scattering from interface roughness was the                One of us  S.A.S.  is pleased to thank S. Sinha  Argonne
major contribution.                                                     National Laboratory  and V. Kaganer  Institute of Crystal-
                                                                        lography, Moscow  for stimulating discussions. We also
                          VII. CONCLUSIONS                              wish to acknowledge the valuable advice from T. Baumbach
                                                                         Institut Laue-Langevin, Grenoble  and J. Ha¨rtwing and U.
      We have presented a theory and experimental results on            Lienert  ESRF  concerning the experimental scheme, and we
the effects of roughness in grazing-incidence diffraction by            are especially grateful to R. Hey  Paul-Drude Institute, Ber-
crystals. It has been shown that surface and interface rough-           lin  for the preparation of the AlAs/GaAs superlattice.


*Permanent address: Institute for Nuclear Problems, 11 Bo-              11 V. H. Etgens et al., Phys. Rev. B 47, 10 607  1993 .
      bruiskaya St., Minsk 220050, Republic of Belarus.                 12 U. Pietsch et al., J. Appl. Phys. 74, 2381  1993 .
1 W. C. Marra, P. Eisenberger, and A. Y. Cho, J. Appl. Phys. 50,        13 S. Grotehans et al., Phys. Rev. B 39, 8540  1989 .
      6927  1979 .                                                      14 D. Bahr, W. Press, R. Jebasinski, and S. Mantl, Phys. Rev. B 51,
2 V. G. Baryshevskii, Pis'ma Zh. Tekh. Fiz. 2, 112  1976   Sov.            12 223  1995 .
      Tech. Phys. Lett. 2, 43  1976  .                                  15 T. Jach, D. B. Novotny, M. J. Bedzyk, and Q. Shen, Phys. Rev. B
3 A. M. Afanas'ev and M. K. Melkonyan, Acta Crystallogr. A 39,             40, 5557  1989 .
      207  1983 .                                                       16 V. Holy´, T. Baumbach, and M. Bessiere, J. Phys. D 28, A220
4 T. Jach, P. L. Cowan, Q. Shen, and M. J. Bedzyk, Phys. Rev. B             1995 .
      39, 5739  1989 .                                                  17 H. A. van der Vegt et al., Phys. Rev. B 51, 14 806  1995 .
5 A. L. Golovin and R. M. Imamov, Phys. Status Solidi A 80, K63         18 H. A. van der Vegt, W. J. Huisman, P. B. Howes, and E. Vlieg,
       1983 .                                                              Surf. Sci. 330, 101  1995 .
6 P. A. Aleksandrov et al., J. Appl. Crystallogr. 18, 27  1985 .        19 M. Sauvage-Simkin et al., Phys. Rev. Lett. 75, 3485  1995 .
7 A. L. Golovin, R. M. Imamov, and E. A. Kondrashkina, Phys.            20 R. M. Imamov, E. A. Kondrashkina, D. V. Novikov, and S. A.
      Status Solidi A 88, 505  1985 .                                      Stepanov, Kristallografiya 32, 852  1987   Sov. Phys. Crystal-
8 R. M. Imamov et al., Poverkhn. Fiz. Khim. Mekh.  USSR  3, 41             logr. 32, 501  1987  .
       1987 .                                                           21 S. A. Stepanov and R. Ko¨hler, J. Appl. Phys. 76, 7809  1994 .
9 S. Rugel, G. Wallner, H. Metzger, and J. Peisl, J. Appl. Crystal-     22 G. T. Baumbach, S. Tixier, U. Pietsch, and V. Holy´, Phys. Rev. B
      logr. 26, 34  1993 .                                                 51, 16 848  1995 .
10 A. A. Williams et al., Phys. Rev. B 43, 5001  1991 .                 23 P. A. Aleksandrov, A. M. Afanas'ev, and M. K. Melkonyan, Kri-



8162                                                       S. A. STEPANOV et al.                                                                 54

       stallografiya 26, 1275  1981   Sov. Phys. Crystallogr. 26, 725     41 T. Salditt, T. H. Metzger, and J. Peisl, Phys. Rev. Lett. 73, 2228
        1981  .                                                                   1994 .
24 L. Ne´vot and P. Croce, Rev. Phys. Appl. 15, 761  1980 .               42 M. Rauscher, T. Salditt, and H. Spohn, Phys. Rev. B 52, 16 885
25 A. Caticha, Phys. Rev. B 52, 9214  1995 .                                      1995 .
26 S. A. Stepanov, Kristallografiya 39, 221  1994   Sov. Phys. Crys-      43 L. D. Landau and E. M. Lifshitz, Statistical Physics  Pergamon,
       tallogr. 39, 182  1994  .                                                 Oxford, 1980 , Pt. 1.
27 S. A. Stepanov, U. Pietsch, and G. T. Baumbach, Z. Phys. B 96,         44 Z. H. Ming et al., Phys. Rev. B 47, 16 373  1993 .
       341  1995 .                                                        45 Y. H. Phang, D. E. Savage, R. Kariotis, and M. G. Lagally, J.
28 S. A. Stepanov and R. Ko¨hler, J. Phys. D 27, 1923  1994 .                    Appl. Phys. 74, 3181  1993 .
29 S. K. Sinha, E. B. Sirota, S. Garoff, and H. B. Stanley, Phys. Rev.    46 J. Daillant and O. Be´lorgey, J. Chem. Phys. 97, 5824  1992 .
       B 38, 2297  1988 .                                                 47
30                                                                              S. K. Sinha, J. Phys.  France  III 4, 1543  1994 .
      V. Holy´ et al., Phys. Rev. B 47, 15 896  1993 .                    48
31                                                                              A. L. Golovin, R. M. Imamov, and S. A. Stepanov, Acta Crystal-
      V. Holy´ and T. Baumbach, Phys. Rev. B 49, 10 669  1994 .
32 V. M. Kaganer, S. A. Stepanov, and R. Ko¨hler, Phys. Rev. B 52,               logr. A 40, 225  1984 .
                                                                          49
       16 369  1995 .                                                           We refer to this scheme as a single-crystal one according to the
33 G. H. Vineyard, Phys. Rev. B 26, 4146  1982 .                                 classification given in Fig. 3, although different realizations may
34 H. Dosch, Critical Phenomena at Surfaces and Interfaces (Eva-                 contain more crystals  e.g., monochromators .
                                                                          50
       nescent X-Ray and Neutron Scattering), Springer Tracts in Mod-           Y. Yoneda, Phys. Rev. 131, 2010  1963 .
                                                                          51
       ern Physics Vol. 126  Springer-Verlag, Berlin, 1992 .                    T. Jach and M. J. Bedzyk, Phys. Rev. B 42, 5399  1990 .
35 S. Dietrich and A. Haase, Phys. Rep. 260, 1  1995 .                    52 T. Jach and M. J. Bedzyk, Acta Crystallogr. A 49, 346  1993 .
36 R. W. James, in Solid State Physics, edited by H. Ehrenreich and       53 A. P. Ulyanenkov, S. A. Stepanov, U. Pietsch, and R. Ko¨hler, J.
       D. Turnbull  Academic, New York, 1963 , Vol. 15, p. 53.                   Phys. D 28, 2522  1995 .
37 B. W. Batterman and H. Cole, Rev. Mod. Phys. 36, 681  1964 .           54 The  222  reflection is forbidden, and the higher harmonics are
38 Z. G. Pinsker, Dynamical Scattering of X-Rays in Crystals                     additionally suppressed due to their lower critical angle of
        Springer-Verlag, Berlin, 1981 .                                          specular reflection from the sample.
39 P. A. Aleksandrov, A. M. Afanas'ev, and S. A. Stepanov, Kristal-       55 A. M. Afanas'ev et al., Fiz. Tverd. Tela  Leningrad  27, 2274
       lografiya 29, 197  1984   Sov. Phys. Crystallogr. 29, 119                  1985   Sov. Phys. Solid State 27, 1365  1985  .
        1984  ,                                                           56 A. M. Afanas'ev et al., Kristallografiya 31, 1066  1986   Sov.
40 S. Takagi, J. Phys. Soc. Jpn. 26, 1239  1969 .                                Phys. Crystallogr. 31, 630  1986  .