2.3.2 Additional Physical Phenomena in Gate



next up previous contents
Next: Conclusion Up: 2.3 Experimental Results and Previous: 2.3.1 Comparison with Experimental

2.3.2 Additional Physical Phenomena in Gate

 

We propose several possible physical causes for the above phenomena as follows:

1)
Quantum effects in the strong accumulated layer increase the surface potential and lower the semiconductor capacitance in comparison with classical results. For example, in -type low-doped silicon () the effect becomes important at surface fields higher than at room temperature [136]. These effects have been investigated in detail in [453] for high field strengths and doping levels, employing the rigorous model from [452]. Differences between the surface potential calculated classically and quantum mechanically (QM) become quite large at very high oxide fields. For , the surface band-bending are [453], and in -type gate with at . At lower doping , , and at the same field and temperature. Note that the results denoted in [453] as classical solution, have been obtained by MB statistics. When applying FD statistics for a three-dimensional electron gas instead of MB statistics, the quantum-mechanical correction is smaller, but still remains significant.
In general, the changes in the semiconductor capacitance have a larger influence over the total capacitance when becomes comparable to the semiconductor capacitance - that means, at thin oxides. In the case of -gate/-channel and -gate/-channel devices, the lowering of semiconductor capacitance in the accumulation due to quantum-mechanical effects occurs at the same time in both, gate and bulk leading to an enhanced deviation from the classical theory. This could be the explanation for the very slow rise of from depletion to heavy accumulation that is measured on the -gate/-channel device with oxide.

2)
Interface trapped charge at the gate/oxide interface of both, donor and/or acceptor nature could reproduce a low slope of curve, as demonstrated in Figures 2.9, 2.11 and 2.12. With increasing it is, however, necessary to assume a higher amount of traps to explain the experiments. In addition, we did not observe any significant change in the curve of -gate/-channel devices when varying the signal from kHz to MHz! Therefore, if there are any traps at the gate/oxide interface, their repopulation must be very fast. An additional conclusion is that the minority carriers can invert very quickly in the gate in our -gate/-channel devices. We observed that an opposite effect has been measured in a lightly doped -type gate in [367]gif. An evident finite time response of minority carriers in the gate is measured in -gate/-channel devices in [226] and [269]. Whether the differences between gate-types regularly appear and why is not clear at present. A smooth inversion in the gate may come out as an artifact in measurements when the AC signal is high, due to finite time response of minorities [269]. For the characteristics in Figure 2.9 we, however, did not observe an -dependence.

3)
Traps at grain boundaries in polysilicon: Traps discussed in 2), by which one is able to explain the experimental observations at the inversion side of the - curve may be related to traps at the grain boundaries residing at the polysilicon/oxide interface. The typical density of grain interface sites is of the order , which is comparable with that assumed to fit the experiment. In addition, if the first layer in the polysilicon close to the interface is built of very small grains, the equivalent volume trap density can be very large close to the interface and can no longer be neglected compared to . A significant trapping of majority carriers takes place just at the interface leading to formation of a shallow resistive layer. Our knowledge of the polysilicon-gate structure in the very first from the gate/oxide interface is quite indigent at present, although some studies may be found in the literature [6].

4)
Nonuniform gate-doping distribution in direction perpendicular to the interface: Increasing from the gate/oxide interface towards the gate-body can explain a low slope in the versus curve. However, the region of interest is very shallow. The depletion region in the gate becomes only at for device in Figure 2.9. Note that the distance between impurity atoms is on the average at this doping level (). To explain the experiment in Figure 2.9 should be assumed to increase within the first from the interface in this device. After becoming fully depleted at a moderate , this shallow low-doped region acts as a constant at a higher . A further increase in is proportional to , but not to as it is for constant doping. Contrary to our erroneous conclusion in [167] the analytical modeling confirmed with numerical simulation has shown that a nonuniform dopant distribution can explain well the smooth inversion in the gate, when the transition region for is very shallow and the doping level in the gate-body is high.

5)
Gate doping fluctuations in the plane parallel to the interface: Motivated by the previous finding that the width of the depletion layer in the gate is comparable with the average distance between impurity atoms, we propose that short-distance fluctuations in the local dopant concentration may have an impact on the - characteristics. To confirm this hypothesis we assume that the large-area device consists of many non-interacting small devices in parallel, each of them with a constant dopant concentration in the gate. This approach is known as the parallel array model, already applied in several similar analyses [331]. The impurity concentration is different in each particular small device, with an average value of across the large device. Considering a physically small volume , the number of dopant atoms contained in becomes . In the constant volume , obeys the Poisson distribution, as for example interface point charges do [331], with mathematical expectation of and standard deviation of . It follows the standard deviation of the dopant concentration . For a sufficiently large that holds, may be well approximated by a gaussian distribution as is done in our calculations. In each single device the gate charge and the surface potential are calculated by the selfconsistent one-dimensional analytical model introduced in Section 2.2, for chosen randomly. The result for the large-area device follows after building the statistical averages. The choice of is discussed briefly. We assume , where is the depletion region width at the doping level . For the parallel array model to be reasonable must be larger than : , with parameter . If is fixed as interesting for the onset of gate inversion, it follows . In depletion, can be considered as constant and we obtain . In both cases the relative standard deviation increases with increasing doping level . For example

is assumed in the calculation. The second example corresponds to

Figure 2.9. A large-area device contains small devices in the latter case.

 

Let us consider the total depletion in the gate. Moving carriers are neglected. The surface electric field governed by the oxide field is negligibly influenced by variations in , because of . We omit from discussion. The gate capacitance in depletion is related to by . Therefore, becomes approximately gaussian distributed as well. For the bulk capacitance in 2.23, holds on the inversion side of the - curve. Thereby, the gate capacitance becomes which can be further approximated by , valid if . Simple analytical considerations lead to , where the line over the quantity denotes the mathematical expectation. Similarly, for the surface potential in depletion one obtains . We have exploited the inequality , valid for a random variable which is gaussian distributed. Therefore, the statistical variations in the gate doping not only induce fluctuations in across the large device, but also increase the mean value of and lower the mean value of measured on a large device. A similar result has been obtained by numerical simulation of the effect of bulk doping fluctuations on the threshold voltage of small MOSFETs in [339]. The slight changes of and in depletion region following from analytical considerations are completely in coherence with results from statistical simulations, Figure 2.13.

More important than small changes occurring in the gate depletion is the impact of the point fluctuations of dopants on inversion at the gate/oxide interface. Let us assume . It follows for the Debye length at . A parallelopiped with base and height (comparable with the depth of depleted region) contains only dopant atoms on average, namely there are atoms in the first from the gate/oxide interface. Any macroscopic approach to and space charge density is no longer valid with respect to the carrier concentrations. Model-calculations accounting for the dopant fluctuations and experiments have shown that the stretch-out of the gate-channel capacitance of MOS-system already takes place at a bulk doping of at room temperature [504] (see also [506]). The same effect occurs in the gate-channel capacitance associated with the gate (minority carrier component in on the inversion side), producing a stretch-out in as well. This effect has been at least qualitatively reproduced in our statistical simulations shown in Figure 2.13, in comparison with experimental data for similar devices. Moreover, by increasing the dopant concentration , the gate-channel capacitance becomes smoother. This fact can qualitatively explain Fig.1 in [281], which shows a fast inversion in the low-doped gate and a smooth recovery for the heavy-doped one. Finally, we think the random distribution of dopants in a shallow depleted region close to the gate/oxide interface is an obvious explanation for the smooth inversion in heavily doped gates [167].

6)
Deep band tails occur in heavy doped space-charge regions. This fact is established both, experimentally [61] and theoretically by calculations [280]. The theory of band tails is dealing with the phenomena in 5), with respect to the quantum states available for moving carriers, accounting for carrier-impurity, impurity-impurity and carrier-carrier interactions, on rigorous first-principle footing. This complex theory is covered rather extensively in the literature e.g. [429][319][239][238][180][37]. Specifics of the gate depletion are the absence of screening when the gate enters into depletion and an enlarged statistical fluctuation in very shallow regions of interest.
In [61], the HF - characteristics of MOS capacitors with bulk doped heavily is measured and modeled. A very large total band-gap narrowing was necessary to fit the minimum of high-frequency on the inversion side when the rigid-parabolic-band model with constant band-gap narrowing is assumed. For example: the extracted values and are much higher than the values of about and for the electrical band-gap narrowing in the quasi-neutral regions at and , respectively. These large values are in accord with the theoretical predictions in [280].
When the gate enters into weak depletion the screening of impurity ions becomes weak due to the increased carrier concentration-dependent screening length and the lattice-disorder increases close to the interface, as calculated in [14]. Many majority carriers remain localized in deep band states. Note that these states do not exist at the flat-band condition in the gate. In fact, by the rigid-band approach and constant (carrier concentration independent) band-gap narrowing we have practically eliminated the moving carriers in discussion in 5). Due to majority carriers in deep states, depletion is only partial. An average voltage drop in the gate and the width of depletion region increase and the gate capacitance becomes lower comparing with the rigid-parabolic-band model. Thereby, we expect lowering of in the weak gate depletion. In stronger depletion, the discussion in 5) holds. At high the minority carriers begin to fill the deep states in the corresponding opposite tail. As a consequence, an overall slope of the curve is reduced on the inversion side of the characteristics. In addition, filling of deep band states by minorities leads to a smooth inversion in the gate. The latter effect has been heuristically reproduced by interface traps in Figures 2.9, 2.11 and 2.12.





next up previous contents
Next: Conclusion Up: 2.3 Experimental Results and Previous: 2.3.1 Comparison with Experimental



Martin Stiftinger
Sat Oct 15 22:05:10 MET 1994