2.2.3 Basic Characteristics of MOS Structure



next up previous contents
Next: 2.3 Experimental Results and Up: 2.2 Analytical Modeling of Previous: 2.2.2 Model for the

2.2.3 Basic Characteristics of MOS Structure

 

The impact of the field penetration into the gate on the flat-band potential, threshold voltage, inversion-layer charge and - characteristics of MOS structure is analyzed in the following.

For the sake of simplicity a uniformly doped bulk is assumed with concentration . For the relationship an expression analogous with A.14 may be written. Relationships 2.2, 2.4, 2.6, 2.16 and the relation form a closed implicit system of equations to determine the field and potential in the structure at applied bias . This system is solved by a sequential method. The Newton iterative procedure is employed to solve the field-potential equations with respect to potential. In the Newton algorithm, limiting and damping are applied on the potential increments, leading to a stable and fast convergency. This efficient technique has been developed by the author in [162].

For the reference device denoted as ``ideal'', is assumed. Moreover, is adopted in the calculations. Measurements have provided in polysilicon and in polysilicon [365]. Note that the exact position of the Fermi level in degenerately doped polysilicon is not well understood (see [275][200]). Regarding the data for the electrical band-gap narrowing in single-crystal silicon, they are usually extracted from the product in the quasi-neutral regions, assuming a rigid-parabolic-band model. Using this data we are able to reproduce the minority carrier concentration, but probably cannot accurately determine the Fermi level position in heavily doped silicon and polysilicon.

Flat-band potential

The flat-band potential loses its meaning for a nonhomogeneous structure like the one considered here, because the flat-band condition cannot exist in the whole structure in the general case. Hence we define the flat-band potential as the gate-bulk bias corresponding to the flat-band condition in the bulk, which we define by: ([95]). From equations 2.4 and 2.6 it follows

 

where is the solution of the Poisson equation in the bulk for vanishing surface field. Since we adopted a uniform bulk, holds. Two terms in expression 2.17 contribute to the -shift in implanted polysilicon-gate devices:

The charge at the gate/oxide interface affects through the term only. For a common value [520] it follows from equation 2.2. Assuming low-doped gate this field induces (accumulation in -type gate). Figure 2.4 shows the flat-band potential with respect to intrinsic level in ideal silicon band versus the activated impurity concentration , with as parameter. Showing instead of itself, the characteristics become independent of the bulk doping. The calculations based on both, Fermi-Dirac (FD) and Maxwell-Boltzmann (MB) statistics in the gate are presented.

 

For positive the variation of with is nearly logarithmic due to accumulation in the -type gate. has a weak influence on . For a negative (not experimentally detected [520]) the influence is stronger, because the gate becomes depleted. Note that positive charges essentially ``improve'' , since the shift occurs towards the degenerate value. According to the results shown in Figure 2.4, measurements of the flat-band potential cannot serve as a reliable proof that a sufficiently high ionized impurity concentration near the gate/oxide interface is achieved. For a -type gate positive may produce significant , because of depletion in the gate. Assuming and , as in the preceding example, we have whereas leads to and a very large shift in . In the latter case, although the bulk is holding at the flat-band, the onset of inversion takes place in the gate, being produced exclusively by the charge at the gate/oxide interface.

Measuring on MOS devices with different oxide thicknesses, the charge at the oxide/bulk interface (assuming vanishing ) may be extracted from the linearly extrapolated relationship [331][275][200][157]. This technique provides the slope proportional to , but only the sum for the intercept on the ordinate axis. Both factors, and depend on the unknown activated impurity concentration near the gate/oxide interface. Moreover, depends on too. Therefore, some measurements in addition to measurements of are necessary to separate (or ) and . A second open question is how defined by can be measured on MOS capacitors with nondegenerate gate by, for example, some of the well established - techniques ([331]).

In Figure 2.4 moderate doping has been considered. is weakly affected by and close to its degenerate value . If the traps in the polysilicon are taken into account, can vary significantly with decreasing doping levels due to trapping.

Our calculations have clearly shown small differences between the results obtained by Fermi-Dirac and Maxwell-Boltzmann statistics. The correction due to degeneracy in determining the Fermi barrier is given by the second term in A.24. Assuming a gate doped quite heavily this term increases by only and for -type and -type gate, respectively at room temperature (). Actually, the impact of FD statistics on our calculations (carried out at room temperature) is smaller than the uncertainty in the adopted model and the available experimental data for the band-gap narrowing. For dopant concentrations higher than the differences between FD and MB statistics become significant, but the polysilicon gates may then be assumed to be degenerate.

Threshold voltage

The threshold voltage is defined in the standard way as that which induces the minority surface concentration , for -type of bulk. From relationships 2.4 and 2.6, it follows

 

The upper index denotes the values at the threshold. Remember that holds for bulk doped uniformly. The last five terms at the right-hand-side in 2.18 are invariant with respect to the effects in the gate. Two factors produce the deviation of the threshold voltage of the nondegenerate-gate devices with respect to their degenerate-gate counterparts:

 

In thin-oxide devices the field

 

necessary to invert the oxide/bulk interface is very high due to the usually high doping . For 2.19, a uniformly doped bulk is assumed. Note that the bulk doping must be increased in thin oxide-devices, because the contribution of the body-factor to the threshold voltage becomes smaller in proportion with . As a consequence of very high the surface field in the gate is very high too and the surface potential can be large.

The results of the selfconsistent calculation at threshold are shown in Figure 2.5. and are given versus , with and being parameters. For different oxide thickness the bulk doping is adapted so that the ideal threshold voltage is the same for all devices (, shown as dashed line in the figure). The calculations show that both factors, and the deviations of from the degenerate value are of the same order. The influence of is considerably emphasized in thin-oxide devices (curve 1), because of the high corresponding . For thick oxides the term becomes quite small (curve 3). A positive charge at the gate/oxide interface reduces the field (equation 2.2) and therefore, it reduces . Assuming the values found in experiments [520] for , this ``screening'' effect is rather pronounced. In fact, a positive attenuates the increasing of due to insufficient dopant concentration in -gate/-channel devices. An eventual negative would have the opposite influence.

Inversion-layer charge and capacitances

The surface field is very high for strong inversion in thin-oxide devices, even at medium gate bias. For example: on result in . Since the field is very high too, the produced voltage drop in the gate can significantly lower the effective gate bias which is mostly responsible for the inversion-layer charge induced. As a consequence of a reduced the drain current decreases, resulting in a degradation of the driving capabilities of devices and the speed of the circuits. Figure 2.6 shows plotted against with and as parameters. is calculated by

 

for uniformly doped bulk. As is well known, formula 2.20 is very accurate (also in the subthreshold region) [164][41]. The calculations are carried out for a thin oxide of which is typical for sub- CMOS technology. A pronounced reduction in is obtained even at high . Note that for the differences of and from their degenerate counterparts are quite small, Figures 2.4 and 2.5. A doping higher than is necessary to obtain a negligible voltage drop in the gate at the highest operating gate bias. The attenuation of due to a screening is small for common values of . We may judge to be the basic parameter which determines the degradation of in strong inversion for a given .

 

The quasi-static (QS) - characteristics, corresponding to Figure 2.6, are given in Figure 2.7. The total QS capacitance of the polysilicon-gate/oxide/silicon structure may be defined by

 

where and are the total induced charge per unit area in bulk and gate, respectively. may include the charge trapped at interface and bulk traps in the polysilicon. Fixed charge in the oxide-volume is considered as constant. Charge trapped at the gate/oxide and oxide/bulk interfaces are included in and , respectively. Differentiating expression 2.6 with respect to one obtains

 

The total capacitances per unit area of the polysilicon gate and the bulk are defined by

 

Expressions employed for and are collected in Appendix A. Note that the interface traps are omitted in the calculations shown in Figure 2.7. The ideal capacitance follows from 2.22 for and the corresponding selfconsistent .

 

At a negative gate bias a small reduction in occurs due to accumulation in the gate. For positive effective gate voltages a significant lowering in is caused by depletion in the gate. The decrease in depends directly on and may be used as a figure of merit of the degradation in and the drain current. A sharp recovery to the ideal capacitance occurring for large positive voltages can be related to the inversion in the gate near the gate/oxide interface (points A and B in the figure). In Figure 2.6 the inversion in the gate is manifested as a change in the slope of the characteristics. Even for high the gate inversion occurs at a quite moderate gate bias in thin-oxide devices. For example: at the point B we have , , and . Remark that for lower the recovery of to the ideal value occurs at lower . This finding is in qualitative agreement with the experimental characteristics available in literature (Fig.1 in [281]).

The same characteristics presented in Figures 2.5 and 2.6 are calculated employing MB statistics in the gate, as well. The results, not shown here, differ only slightly from those obtained by FD statistics. With respect to quantities in the bulk, like surface field and inversion-layer charge, the influence of the degeneration due to Fermi-Dirac statistics in the gate has no practical relevance at room temperature. With regard to gate capacitance, small differences occur in the accumulation and at the onset of gate inversion. Note that, although their impact on the total capacitance is quite small, large differences between calculated by FD and MB statistics occur in gate accumulation and inversion.

Some specific conclusions may be drawn from the calculations in this section:

  1. Comparing with the conventional theory of MOS structures, there are two new parameters in our analysis: the activated impurity concentration near the gate/oxide interface and the charge at the gate/oxide interface . The impact of on the flat-band potential and the threshold voltage is found to be strong. Assuming a positive charge this is particularly true for -gate/-channel devices.
  2. In order to achieve the maximal performance of MOSFETs the lowering of the inversion-layer charge due to gate depletion must be negligible in the whole operating region. The activated concentration necessary to meet this goal is considerably higher than the concentration sufficient to obtain proper and , especially for thin-oxide devices. Finally, we believe the measurements of both, and are not proper indicators for the activated impurity concentration in the gate close to the oxide. It is indispensible to measure the complete quasi-static - characteristics in order to evaluate the performance of implanted-gate MOSFETs.
  3. It has been demonstrated that the assumption of Maxwell-Boltzmann statistics in the gate is reasonable for modeling the gate-depletion effect. The influence of degeneration due to Fermi-Dirac statistics is judged to be of no practical relevance at room temperature, with respect to bulk quantities.



next up previous contents
Next: 2.3 Experimental Results and Up: 2.2 Analytical Modeling of Previous: 2.2.2 Model for the



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