In the previous sections, we have discussed metal-impurity accumulation at dislocations by segregation in the strain field and in the dislocation core and by precipitation at dislocations. Each impurity state will be associated with a different recombination activity since the relevant defect states will be fundamentally different for the different cases. Related to impurities at dislocations, the main purpose of external gettering is to reduce local carrier recombination at dislocations, which is difficult to follow by spatially averaging techniques and requires local measurements by e. g. LBIC or EBIC. Recent developments or spatially resolved luminescence [190-192] (photoluminescence, PL, and electroluminescence, EL) probe the local excess carrier lifetime by measuring the intensity of the band-band luminescence (BB)) [193-195] or dislocation luminescence, which provides additional information.
In a proof-of-concept experiment the effect of gettering and passivation on the dislocation luminescence has been studied following the fact that its intensity is closely related to competing nonradiative recombination channels due to impurities or intrinsic defects. Drozdov et al.  for the first time reported dislocation-related luminescence in silicon. Basically, it consists of four nonuniformly broadened bands labeled D1, D2, D3, D4. Typical PL spectra measured at T = 5 K in samples with different dislocation densities Nd are shown in Figure 4.14(a). The D-band luminescence always appears in the presence of dislocations and remains stable even after annealing at 1000°C, which provides evidence that it cannot be attributed to any thermally unstable point defects.
Polarization measurements reported showed that the elctric field vector of the D3 and D4 luminescence light is in the glide plane of dislocations and is strongly correlated with the direction of the dislocation Burgers vector [198-200]. This implies that D3 and D4 are related to dislocations themselves. For the D1 and D2 luminescence, the situation is more complicated, but the electric field vector is found to be perpendicular to the dislocation glide plane. The conclusion is that D1 and D2 are associated with the dislocation network, but not with point defects or impurity atoms randomly distributed away from dislocations. The ratio of intensities of D1, D2 to intensity of D4 luminescence depend strongly on the local dislocation density and morphology of the dislocation arrangement, which may strongly vary for different samples. At low dislocation density, D3 and D4 luminescence usually prevails, while at high dislocation density the D1 luminescence becomes dominant.
The energy positions of D3 and D4 luminescence were found to depend on the width of stacking-fault ribbon between 30° and 90° partials of dissociated 60°-dislocations [201, 202], which provides evidence that the D3 and D4 lines originate from recombination processes at regular parts of dissociated 60° dislocations. It is widely accepted now that the D4 luminescence corresponds to no-phonon recombination of electrons in the 1D band at EDe with holes in the 1D band at EDh [153, 203, 204] at dissociated 60°-dislocations (channel ‘1’ in Figure 4.12(a)), while the D3 is the TO phonon-assisted replica of D4.
The origin of D1 and D2 luminescence is still not fully understood. According to the literature, some impurity atoms in the dislocation core , dislocation jogs , segments of dislocations of special types (like Lomer dislocations) appearing due to dislocation reactions, multi-vacancies in the dislocation core  are typical examples
of the possible origin of the D1 luminescence. When selecting a model it should be taken into account that the D1 band usually dominates the luminescence spectra in samples with a high local dislocation density (see, e. g., Figure 4.13(a)) where dislocations could interact with each other. This supports the idea about jogs or Lomer dislocation segments, but clearly does not rule other models such as multivacancies in the dislocation core or oxygen-related defects at dislocation jogs. It should further be noted that usually the D1 luminescence consists of several overlapping luminescence lines [208, 209]. Therefore, it can not be excluded that several different defects at dislocations are responsible for D1 luminescence. As an example, annealing of dislocations in conditions when they can accumulate a significant number of oxygen atoms considerably influences the intensity distribution between different lines of the ‘D1-family’ [210-212].
Among all dislocation luminescence bands, the D1 is most interesting for practical applications because it survives up to higher temperature than the others, with possible applications in LED devices (see Chapter 1). For illustration, Figure 4.14(b) shows typical luminescence spectra in a sample with relatively high dislocation density ND & (3 – 5) x 108 cm-2 . All luminescence bands D1-D4 are present at low temperature, while at high temperatures the D1 band dominates. The shift of luminescence to lower energy with temperature nicely correlates with the temperature dependence of the Si bandgap .
Figure 4.14 PL spectra of dislocations in crystalline silicon, (a) PL spectra recorded at T = 5K for dislocation densities of ND & 105 cm-2 (bottom), ND & 107 cm-2 (center) and
Nd & 5 x 108 cm-2 (top). Please note the strong increase of D1 and D2 intensities with increasing dislocation density. (b) PL spectra recorded at different temperatures as indicated for a dislocation density ofND & 5 x 108 cm-2. Only the D1 line is obtained at around room temperature. The spectra are normalized to their integral intensity. The shift of luminescence to lower energy with temperature increase corresponds to a decrease ofSi bandgap. Note the absence of the 3D band-to-band luminescence at 1.1 eV .
Figure 4.15 Deep levels at dislocations and D1 luminescence duringgettering and passivation (float-zone Si, Nd = (3 — 5) x 108 cm-2). Labels refer to ‘p1’: after plastic deformation at 750PC, ‘p2’ after subsequent AlG, ‘p3’ after additional PDG, and ‘p4’ after a final hydrogen passivation, (a) DLTS spectra showing the drastic reduction of deep-level concentration due to gettering and passivation, (b) Temperature dependences ofDI-luminescence intensity that shows a drastic increase especially at high temperatures due to the reduction of competing non-radiative recombination (after ).
The D1 luminescence is a suitable probe to study the effect of gettering and passivation on recombination-active defects at dislocations [196, 213]. The reason is the competition of radiative-i. e. D-band luminescence – and non-radiation recombination-via deep states-at dislocations. Along this route, a series of experiments has been performed that simultaneously follow the deep-level spectrum at dislocations via DLTS and the D-band luminescence via PL and EL after gettering and passiviation treatments. Figure 4.15 shows typical results for p-type float-zone silicon that was plastically deformed to 2.2% resulting in a dislocation density of ND = (3-5) x 108 cm-2. The DLTS spectra in (a) demonstrate a strong reduction of the deep-level density in the lower half of the bandgap as a result of AlG (‘p2’), additional PDG (‘p3’), and a final hydrogen passivation (‘p4’). It should be noted that all gettering treatments have been terminated by a slow cooling to room temperature in order to promote precipitation of residual metal impurities. As a summary, these experiments show at least qualitatively, that nonradiative recombination at dislocations can be reduced drastically by gettering and passivation.