A substitution (at least here) consists of rules how to enlarge a tile and replace the enlarged tile with other tiles. If the union of the latter ones is similar to the original tile, then the substitution is called self-similar substitution. For example, the substitution for the Penrose Rhombs is not self-similar, but the substitution for the Robinson Triangles is.
In other words, a substitution is a self-similar substitution, if $\sigma(T)=T$. A substitution tiling is called self-similar, if it can be generated by a self-similar substitution. It is known that any - sufficiently nice, i.e., repetitive and flc wrt translations - tile substitution in the plane can be made self-similar, by using fractal boundaries.
A weaker version is described by the term ‘self-affine’ tiling [LW96] , [BG94] . The definition of this reads exactly as above if one replaces ‘similar to’ with ‘affine image of’.
  [BG94]
  Bandt, C and Gelbrich, G
  Classification of self-affine lattice tilings
  
    J. London Math. Soc.
  
  1994,
   50,
   pp. 581-593,
  
  [LW96]
  Lagarias, J C and Wang, Y
  Self-affine tiles in $\mathbb{R}^n$
  
    Adv. Math.
  
  1996,
   121,
   pp. 21-49,
  
    
      
        MR1399601