As discussed in Section 2.1, defect states at a semiconductor–dielectric interface generally form a continuum of energy levels within the semiconductor bandgap. This continuum is made up of separate contributions from different species of states with different properties. The available evidence suggests that each of these sub-distributions may be approximately described by a Gaussian function centred around a certain energy [9], [134]. The energy levels of the states associated with such defects in principle depend on the local chemical environment of each defect. Consequently it is expected that the distribution of interface states at the Si–Al2O3 interface will, for example, be distinct from that of Si–SiO2, even if the major interface states arise from silicon dangling bonds in both cases.1
Measurements of the energetic distribution of the Si–Al2O3 interface states have been reported by a number of authors [39], [40], [43], [45], [46], [48]–[50], [52], [53], [91], [135]–[140]. However, relatively little consideration has been given to the significance of this distribution to a fundamental understanding of the interface properties, and little attempt has been made to compare and contrast it with the interface state distributions exhibited by other silicon–dielectric systems. In this section the Si–Al2O3 interface state distribution is presented, and its
1This remains so despite the likely presence of a thin SiO
xinterlayer at the Si–Al2O3interface (see Sections 1.2 and 9.1.3), as demonstrated in Chapter 8.
4.1. Interface state distribution 43 t Stn S– 'LW nr n S m p p p p S p – (nn(Lnrm psA psS p psS psA (a) Ql C−1 ' LW Ce ( C − (3(3 (3(( (3(− (3(1 (CC(LCe 32c 32− 3 32− 32c (b)
Figure 4.1: Interface state densityDit as a function of energy in the bandgap for various Al2O3 films on Si. The data have been selected to show the behaviour of Dit(E) over a wide range of concentrations and to illustrate some of the possible variations in the shape of the distribution.
characteristic features and modes of variation are described and discussed. Such an exposition will provide a foundation for later, more detailed discussion of the interface state properties.
Fig. 4.1 shows Dit(E) determined for p- and n-type surfaces passivated by various Al2O3 films. The films were deposited using APCVD with a variety of deposition conditions, reactants, and post-deposition treatments, and have been selected to show the behaviour of Dit(E) over a wide range of concentrations, and to illustrate some of the possible variations in the shape of the distribution. Data for Al2O3 deposited by thermal ALD is also shown for comparison.
From Fig. 4.1 it can be seen that the Si–Al2O3 interface state distribution is not symmetric in the silicon bandgap. Rather, in all cases it possesses a mini- mum in the lower half of the bandgap. This minimum is located nearEi−0.18 eV for surfaces with low values of Dit, and generally shifts to lower energies as Dit increases. On either side of the minimum, Dit increases monotonically towards the band edges. When viewed on a log scale, the increase of Dit through the middle part of the bandgap towards Ec generally appears to occur in three dis- tinct stages: first a relatively rapid increase above the minimum up to a point just below midgap, then a more gradual rise, approaching a plateau in some
cases, to about 0.2 eV above midgap, and finally another sharp climb towards the band edge, which is observed to continue with a more or less constant slope (i.e. an exponential behaviour) over the energy range in which the measurement is accurate. This complex energy dependence suggests the existence of multiple contributions to the distribution in this range. The density of states at all en- ergies above the minimum varies substantially between different samples, which suggests that these states are largely extrinsic. In contrast, the density of states in that part of the distribution nearest the valence band appears to be practically independent of sample processing, suggesting that these states may be intrinsic, possibly band-tail states.
While Dit is observed to increase towards both band edges within the energy range accessible to experiment, the extent of this increase is relatively slight. In- deed for samples with midgap Dit in the range of 1012eV−1cm−2, Dit at midgap may be higher than that near the valence band in the measurable part of the distribution. It would be a gross oversimplification simply to refer to the distri- bution as “U-shaped”, as is sometimes done. This would be to ignore the most interesting and pertinent features of the data (the distribution of states in the middle part of the gap) in favour of those elements that are most superficial, most uncertain, and least relevant to recombination.
Measurements by other authors using the most reliable methods (quasi-static C–V measurements) of Si–Al2O3 interfaces formed by a variety of other tech- niques (spray pyrolysis, thermal and plasma-enhanced ALD, and PECVD) [39], [40], [43], [46], [48], [49], [52], [53], [91], [136], [137], [140] agree well with the shape of the distributions shown here. This suggests that the observed char- acteristics are a general feature of the Si–Al2O3 interface, independent of the means used to form the film. Fig. 4.1b shows the extremely close match between the lowest Dit(E) distribution shown for APCVD Al2O3, and that measured for
thermal ALD Al2O3. In both cases the value of Dit at midgap is in the range of 3–4×1010eV−1cm−2, which is remarkably low, on par with the best values
reported elsewhere in the literature for Al2O3 [42], [44], [49]. This shows that, contrary to common preconceptions, APCVD is capable of producing Si–Al2O3 interfaces of a quality every bit as good as that achieved by ALD or other tech- niques.
Forp-type surfaces passivated by Al2O3, Dit could not in most cases be mea- sured accurately at energies of about 0.12 eV or more above midgap. This is due to a non-equilibrium behaviour of the inversion layer charge that occurs for these substrates. When sweeping the voltage from inversion to accumulation,
4.1. Interface state distribution 45
this manifests as an apparent shift in the onset of strong inversion to lower volt- ages, and in some cases the presence of a pronounced peak in the quasi-static capacitance in weak inversion. When sweeping from accumulation to inversion, it results in partial deep depletion of the silicon surface and a consequent dip in the apparent quasi-static capacitance. The result in either case is a significant error in the value of Dit extracted from the C–V curve in weak inversion, either an overestimation (for a sweep from inversion), or underestimation (for a sweep from accumulation). The onset of this non-equilibrium behaviour can be shifted to more positive voltages either by reducing the voltage sweep rate, or by increas- ing the measurement temperature. However, even at the lowest practical sweep rates (5 to 10 mV s−1), the effect remains significant at room temperature. The
same effect has been reported to occur for p- andn-type Si–SiO2 MIS structures at low temperature, with high lifetime substrates, or under illumination [141], and has been explained as being due to the inability of minority carrier gener- ation and recombination processes to maintain an equilibrium concentration of inversion layer charge. This explains why the effect is not observed for n-type surfaces passivated by Al2O3, since for these surfaces the inversion layer charge under the gate is coupled to that in the external inversion layer induced by the negative charge in the film beyond the gate contact area. It can therefore more easily maintain equilibrium with the voltage signal, in the manner discussed by [142].
It is notable that Dit measured for p-type substrates is consistently higher than that forn-type substrates with the same processing. It is not entirely clear whether this difference is real, or whether it is a measurement artifact. On the side of it being real the following evidence can be cited: 1) Dit measured by the conductance method at midgap was also found to be higher for p-type than for
n-type samples with the same Al2O3 film (compare Figs. 4.5 and 4.6), and 2)
J0 at undiffused n-type surfaces is found to be consistently lower than that of
undiffused p-type surfaces passivated by the same film (see Chapter 7), while J0
should be identical ifDit and Qtot are the same (for large Qtot).
On the other hand, it is shown in Section 8.2 that the apparent Dit depends, at least in some cases, on the direction of the voltage sweep used to measure the C–V curves. The use of the same sweep direction generally results in good agreement of Dit for p- and n-type samples in that case. Data measured using both sweep directions are not available in majority of cases, because such sweeps were not routinely recorded, which makes testing this hypothesis more difficult. However, in cases when such data is available, its use does not always result in a
significant difference in apparent Dit, or significantly better agreement between p and n-type data. The balance of the evidence therefore appears to be on the side of there being a real difference in Dit between p- and n-type surfaces. The reasons for this could perhaps relate to different surface carrier concentrations during the initial stages of deposition. However, we should remain cautious on this question.
In summary, it is concluded that the Si–Al2O3 interface states form a char- acteristically asymmetric distribution with a minimum in the lower part of the bandgap. The shape of this distribution varies relatively little between films pro- cessed under different conditions or deposited by different techniques, pointing to the fundamental similarity of the different interfaces. The apparent density of interface states is consistently found to be somewhat higher at p-type surfaces than at n-type, though there is reason for caution in interpreting this as a true difference in Dit.