14 3 However even in the very dilute regime of ~10 cm the

random distribution of the impurities will place some of the neutral donors close enough together to create a significant interaction. This effect can be examined in terms of the molecular model of Bajaj et al <1975). This model is sometimes referred to as the charge transfer model (see eg Thomas et al 1981). In the molecular model the energy levels of the free space neutral hydrogen molecule H^ or the hydrogen molecular ion H^ are scaled, in a similar manner to that for neutral hydrogen when studying neutral donors, using the effective mass and dielectric constant for the semiconductor. It is found that the energy of some of the transitions is relatively independent of the inter-donor separation. When account is taken of the probability distribution for the impurity separations a moderately sharp line can be observed in spite of the random nature of the inter-nuclear separations. Lines attributable to transitions in the hydrogenic

molecules and have been observed at energies of

*

~0.6 R in GaAs. Similar states have been observed in InP (Reeder et al 1986). However as the interaction between the molecules depends strongly on the overlap of the wavefunctions, the intensities and energies of any transitions will be strongly dependent on the magnitude and orientation of an applied magnetic field. The transitions observed by Bajaj et al and attributed to these molecular states decreased in amplitude rapidly as a magnetic field was applied and had vanished by 1 Tesla.

Peak Inversion or Notch Effects in Shallow Donor Spectra.

undoped and doped GaAs which have been made since the early 1970’s, there is still considerable uncertainty over the assignments of specific donor species to the individual central cell components. Even the dominant central cell component X^, seen in almost all undoped samples irrespective of growth technique, has had its assignment changed from silicon to sulphur as recently as 1982 (Low et al 1982c).

There are numerous reasons which contribute to the uncertainties in assigning a particular impurity species to a central cell component. Firstly, although sharp central cell components are seen in very high purity undoped material, samples back doped with a particular impurity species invariably have a higher impurity concentration and significantly broader central cell components. The doping must be sufficiently strong so that the desired impurity species can be positively identified from other background contaminants (whose concentration may also be inadvertantly increased), but not so strong that the central cell components broaden so far that no structure can be resolved at all. Unfortunately the range of suitable concentrations is very small and few back doped samples show unambiguous central cell structure.

A further problem with back doped samples is that the application of band gap radiation does not have as great an effect on the linewidths as in high purity undoped samples, since the higher doping level leads to faster donor-acceptor recombination rates and thus relatively small changes in the ionized impurity concentration.

However in addition to these general problems of working with back doped samples there is a further effect which can lead to confusion over the origin of peaks in the central cell structure. This is the ’peak inversion* or ’notch effect’, where a single central cell component is distorted so that it has two peaks separated by a minimum

at approximately the position where the maximum should have occurred. Although the effect has been seen in a number of samples studied by groups at Oxford and St. Andrews it has only recently been acknowledged in the literature (Low et al 1983b). The major problem associated with identifying its origin has been the apparently random way in which it affects samples. Some samples may be completely unaffected, whereas others may show ’notching* on only one of the central cell components, and some may be distorted beyond recognition.

Various mechanisms have been proposed to account for these effects, and usually involve some quirk of the photoconductivity mechanism. Bad contacts can give noisy spectra, and generate unwanted photovoltaic or bolometric signals. Alternatively, negative photoconductivity could create a notch, and has been shown to arise from the presence of compensating impurities, where the sign of the photosignal was found to depend on sample thickness (Kaplan et al 1968). As the apparent sample thickness could be influenced by the magnetic field or the application of band gap illumination it might be possible to observe size effect resonances in the extrinsic absorption at various combinations of laser wavelength, magnetic field and light intensity etc.

However these explanations were rather hypothetical. The problem of peak inversion has been put on a much more rigorous basis by the work of Stillman, Low and Lee (1985). Normally it is assumed that as the impurity concentration is very low the absorption of extrinsic radiation is very small, and consequently the behaviour of the material is not changed in the presence of FIR radiation. Stillman et al have shown that in fact this is not true: even in very high purity material the absorptance of a sample can be very high, almost unity, and consequently the dielectric response of a sample close to an

impurity transition is very frequency dependent. The effect can be so great that the complex dielectric constant can cross over from positive to negative values, at which point a sharp notch can appear

in the absorption spectrum. Stillman et al modelled the dielectric response function and were able to calculate the absorption spectrum

in the vicinity of an impurity transition, and generate notching effects with various degrees of severity.

Their model predicts structure on the central cell peaks which is in good agreement with that observed experimentally in certain samples. In particular their model predicts that notching becomes more severe as the sample thickness is increased, as the impurity concentration is increased, and as the imhomogeneous linewidth decreases. The latter effect can lead to an increase in the severity of notching if a magnetic field is applied or if band gap illumination is applied or if the sample temperature is reduced since all these effects can result in reductions in the inhomogeneous linewidth.

Stillman et al were able to demonstrate experimentally the increase in notching as the magnetic field or sample thickness is increased. One of the principle implications of this work is that studies of the central cell structure must be made on relatively thin samples, of the order of lOiun in order for meaningful results' to be obtained. Much of the work described in the following sections of this chapter provides detailed confirmation of their work.

Experimental Results.

Five samples of very high purity epitaxial n-type GaAs have been studied. The electrical characteristics of these samples are given in Table 4.1. Mobilities and ionized impurity concentrations have generally been obtained from Hall analyses performed by the crystal growers. A detailed description of the Hall analysis for the sample RR98B has been published (Colter, Look and Reynolds 1983; Look and Colter 1983).

The samples can be divided into three groups according to the growth technique. These are

i Molecular Beam Epitaxy (MBE). MBV20 and MBV380 are fairly typical samples produced by MBE, with mobilities of the order of 50000

2 ^ 2

cm V s . Both samples were cloverleaf structures and were