Solid-State Electronics Vol. 37, No. 3, pp.

411414, 1994

Pergamon

Copyright0 1994 ElsevierScienceLtd Printed in Great Britain.All rightsreserved 0038-l lOI/ 96.00 + 0.00

A SIMPLE MODEL FOR QUANTISATION EFFECTS IN HEAVILY-DOPED SILICON MOSFETs AT INVERSION CONDITIONS
M. J. VAN DORT, P. H. WOERLEE and A. J. WALKER Philips Research Laboratories, Prof. Holstlaan 4, 5656 AA Eindhoven, The Netherlands
(Received 13 July 1993; in revised form I8 August 1993)

Abstract-The transistor parameters of state-of-the-art MOSFETs are affected by quantisation effects of the carrier motion in the-inversion channel. To account for these effects in clas&l device stimulators, we show that a better modeling of the silicon bandnan at inversion conditions is E Y = E CoNV+ ?Ac in which ACis the position of the-first energy level with respect to the bottom of the :onductron ban& The improved modeling of the bandgap leads to a new model for the intrinsic carrier concentration ni . The model for n; has been tested against measurements and against self-consistent QM calculations. Excellent agreement is obtained. 1. INTRODUCTION

The quantum nature of the 2D electron gas in MOSFETs has been extensively investigated and is a well-established fact (for an excellent review article see e.g. Ref.[l] and references therein), but a full quantum-mechanical (QM) treatment is difficult and is for silicon MOSFETs hardly ever necessary. The device engineer therefore usually relies on the use of classical device simulators for the calculation of the Z-V characteristics of MOSFETs. This approach is adequate for conventional MOSFETs, but the operation of deep-submicron MOSFETs is now entering a regime where QM effects become noticable and a pure classical description of the physics is not sufficient for an accurate calculation of the inversion charge. In state-of-the-art MOSFETs high levels of substrate doping are needed to suppress leakage currents. In Ref.[2], it was experimentally shown that the use of high substrate doping levels affect the threshold voltage of MOSFETs in an anomalous way. This is directly caused by the quantisation of the motion of the minority carriers perpendicular to the interface. This leads to a splitting of the energy levels in the conduction band and to a displacement of the minority carrier distribution from Si-SiOz interface[3]. It is well known that the wavefunctions associated with the impurities overlap for sufficiently high doping of the silicon substrate. For donors, these wavefunctions form an impurity band just below the conduction band (for acceptors just above the valence band). This effect can be described as an apparent bandgap narrowing[4]. In MOS devices, quantum effects associated with the confinement of the minority carriers in the inversion layer can be treated as an effective bandgap widening. For MOSFETs, this effect is dominant.

The aim of this paper is to propose a simple model to account for these quantisation effects, which can be implemented in classical drift-diffusion device simulators. It will be shown that this can be done by the introduction of a new model for the intrinsic carrier concentration. The paper is organised as follows. In Section 2, we will describe the experiments. The new model for the intrinsic carrier concentration is presented in Section 3, and compared with measurements and with fullyselfconsistent QM calculations of the number of mobile carriers in the inversion layer. Some conclusions are finally presented in Section 4. 2. EXPERIMENTAL Large-area MOS capacitances were processed with a surface substrate doping level ranging from 3 x lOI to 3 x 108cm-3. Doping was achieved by means of a 30 and a 90 keV BF, implantation. The energies and the doses were chosen to obtain flat doping profiles near the interface. After the implants were done, a gate oxide of 140 A was grown. The polysilicon gates were doped by means of a phosphorus implantation, followed by a gate reoxidation and conventional S/D processing. The inversion-layer and depletion-layer capacitances were determined using the split CV method. The values of the capacitances were corrected for a stray capacitance of the wiring of -0.2pF. Quasistatic and high-frequency CV measurements were employed to determine the values for the flat-band voltage, the oxide charge and interface state densities (both negligible) and the acceptor concentration near the Si-SiOl interface. There is an excellent agreement between measured and simulated substrate doping levels[2]. For all samples, a constant deviation of -60 mV between the measured and theoretical flat411

412

M. J. VAN DORT et nl.

band voltage was observed. The measured flat-band voltage was used in the simulations. The measured number of mobile carriers versus the gate voltage is plotted in Fig. 1 for a doping level of the substrate of -2 x 10cm~3, and in Fig. 2 for a doping level of - 1 x IO* cmm3. The threshold voltages of all the devices were also determined. Figure 3 depicts the difference of the measured threshold voltage and the one calculated with a classical device simulator. The differences increase with the substrate doping level.

NA

Exp. data

E 6 2

New model n

I 2.5

I 3.5

The inclusion of the various subbands in the calculation of the electrical properties is hardly ever necessary, because the splitting of the energy levels is usually small compared to the thermal voltage kT/q. Many of these subbands are then occupied and a quantum description of the MOS device resembles the conventional one. Under these conditions, a classical approach based upon a simple parabolic dispersion relation is adequate. However, at high doping levels the normal electric field increases significantly, and QM effects become more pronounced[5]. It turns out that for a given population of electrons in the conduction band a surface potential Y$ is needed which is greater than the conventional value of YSoNV. We find: YgM = Y;oNv + A.c/q + E,Az. A6 is the energy bottom of the conduction band and energy level, En is the electric field the interface, and AZ the increase distance to the interface compared
In this equation,

Gate voltage

(V)

Fig. 2. As Fig. I, but now for a substrate doping level of I x IO* cm-. The solid line is the result of our new model for n,, whereas the dashed line has been calculated using the self-consistent QM theory of Baccarani and Wordeman. The number of mobile carriers has also been determined with the conventional model for n, (dashdotted line). There is good agreement between the data on the one hand and the results of the new model for n, and the self-consistent QM calculations on the other hand. There is a large discrepancy between the data and the conventional model for n,.

(1)

gap between the the first allowed perpendicular to of the average to the classical

solution AZ = (zQ) - (z ON) (cf. Fig. 4). This QM correction of the surface potential leads for instance to higher threshold voltages as was experimentally investigated in Ref.[2] (see also Fig. 3). From the physical point of view, it seems quite natural to incorporate the splitting of the energy levels into the modeling of the bandgap. The bandgap then becomes larger, viz. Eg = Ey + AC. The displacement of the electron distribution effectively increases the oxide thickness (r Et = r,, + c,, /cs, AZ), which modifies the gate capacitance. This is a minor effect for conventional MOSFETs, but is significant for advanced devices with thin gate oxides. The effect of the displaced electron distribution is incorporated in the model through the third term in eqn (I). This

0.6 0.5

0.4

New model i

0.3 0.2

r
-

- - - Analytical New model

model P

1x105 -0.5

I 0 0.5

I 1.0

I 1.5

I 2.0

0.1 1;1014 1x105 1x106 1x101 1x108 1x10

Gate voltage

(V)

Fig. I. The number of mobile carriers vs the gate voltage for

Substrate

doping

level (cmm3)

Fig. 3. Difference between the measured threshold voltage and the threshold voltage simulated with the conventional model for n, as a function of the channel dope (r,, = 140 A). The solid line has been calculated with our new model, the dashed line is the result of the analytical model of Ref.[2].

Quantisation effects in MOSFETs

413

(4

This is a good approximation for a triangular-shaped potential. The value of the proportionality factor p can be found from the theory, or can be experimentally determined from the shift of the long-channel threshold voltage. We find /I = 4.1 x lo- eV cm from the observed threshold voltage shift at high doping levels[2]. The new model for the bandgap results in a modification of the intrinsic carrier concentration in the inversion layer: n FM= n FNV exp[(E2M - Ey)/2kT],
(4)

Conventional

s
e

Fig. 4. (a) Schematic representation of the potential in strong inversion. In the state of weak inversion the triangular shape is adequate. In the quantum-mechanical picture the energy spectrum consists of a discrete set of energy levels. The first energy level does not coincide with the bottom of the conduction band. (b) Electron density vs distance. The average distance to the interface is larger in the quantum-mechanical framework. Hence, a larger band bending is needed for a given population in the conduction band.

in which n CoNVis the conventional model for the intrinsic carrier concentration. The new model for the bandgap should of course only be applied to the region where the electrons are confined to a layer near the interface. In the pinch-off region and in the bulk of the silicon substrate, the classical formulation of ni is more appropriate. In eqn (3) the sign of the normal electric field has been used to distinguish between the pinch-off region and the inversion layer. The bulk of the substrate can be excluded in the same way as is done for the modeling of the surface roughness scattering[l] and the surface impact ionisation[9] in MOSFETs. We have used the function F(Y) which is also used in the MINIMOS-4 mobility model[8]: F(Y)=2exp(-a*)/[1 +exp(-2a*)], (5)

term expresses the extra band bending needed to obtain a certain population of the conduction band due to the increase of the average distance to the interface. The calculation of AZ is not easy for arbitrary doping levels and gate voltages. A simple expression for AZ can however be found using a variational method to calculate the electron wave function taking only the first subband into account[6,7]. We find qE,Az x$Ac. Within this approximation, the increase of the average distance to the interface can thus be expressed as a function of the position of the first energy level. Combining the two effects, a better approximation of the bandgap in the inversion layer is:

with a = Y/Y,r. Here, Y is the coordinate perpendicular to the interface, Y = 0 denotes the position of the interface, and Y,, is a reference distance large enough to encompass the inversion layer (Y,r * 250 &9]). The intrinsic carrier concentration thus becomes: ni=ncoNv[l -F(y)]+F(y)n?M. (6)

All we have to do is find a simple expression for AC which can be evaluated rapidly during the numerical solution of the drift-diffusion equations. Several approximations are possible. We have used:
113

max(E,(0),0)2/3.

(3)

The transition region from the new value for ni to the conventional value occurs outside the inversion layer and no special requirements for the mesh are necessary. The new model for n, has been implemented in the classical device simulators CURRY[lO] and MEDICI[ 111. Although we have given a physical reasoning for our model for nit our equations have not been derived from first principles. The validity of this expression can only be tested by comparing the numerical results based upon eqn (6) with measurements and with self-consistent QM calculations. For these QM calculations, we rely on the formulation of Baccarani and Wordeman[ 121.These calculations take the first subband explicitly into account, and are valid for high doping levels where only a few subbands are occupied. For low values of NA many subbands are occupied and a simple classical description of the electron

414

M. J. VAN DORT et al.

motion is valid. We have determined the number of mobile carriers from our capacitance data and plotted the result versus the gate voltage in Fig. 1 for NA = 2 x 1OJ6 -. We see an excellent agreement cm between the theoretical curves and the experimental data. Notice that for these low values of NA the results of our new model for n, coincide with the simulations based on the conventional model for n,. The reason is that for small values of AE, the QM value of bandgap E 2 is approximately equal to the conventional value E y, and as a result n, x n FoNv. Figure 2 shows the number of mobile carriers as a function of the gate voltage for a doping level of - 1 x 10 cmm3. For these high values of N,, approx. 60% of the electron population in the inversion layer is in the lowest energy band, and the theoretical model of Baccarani and Wordeman[ 121 is applicable. We have solved these self-consistent equations for our case. The agreement between these QM calculations on the one hand and the numerical simulations based upon our new model for n, and our measurements on the other hand is excellent. The calculated curve with the Baccarani model coincides with the curve calculated with the new model for n,. For comparison, we have also included the solution of the drift-diffusion equations for the conventional model for n,. Already from Fig. 2, it is quite evident that there is a large difference between the threshold voltage calculated with the classical device simulator and the experimental data. This is more systematically investigated in Fig. 3, where this difference AL, = VyEAS - t FoNVis plotted as a function of the substrate doping level. If, on the other hand, the threshold voltage is calculated with our new model for ni, the calculated threshold voltage coincides with the experimental data for all doping levels of the substrates.
4.CONCLUSIONS

bandgap at inversion conditions, leading to a modification of the intrinsic carrier concentration, can describe the measurements in the high doping-level regime, where the conventional description fails to reproduce the experimental data. The new model has been tested on samples with a wide range of doping levels, and carefully verified against self-consistent QM calculations. We have chosen to compare our model with CV data (the number of mobile carriers) and not with the drain currents. For the calculation of the drain currents, we would have to use the models for the mobility as well, making a clean comparison with theory more difficult. It is therefore better to test the new model against capacitance data. Notice that our model gives the same results for the threshold voltage as the analytical model of Ref.[Z]. The advantage of the new approach is that the new model can be built in a device simulator, whereas the previous analytical model was entirely done in postprocessing and could not be used in a device simulator.
Acknowledgements-The authors thank J. W. Slotboom for valuable discussions and suggestions on bandgap modelling and A. Heringa and J. Peters for CURRY support. The authors further acknowledge the support of ESPRIT, project number 7236 (ADEQUAT).

REFERENCES I. T. Ando, A. B. Fowler and F. Stern, Rev. mod. Phys. 54, 437 (1982). 2. M. J. van Dort, P. H. Woerlee, A. J. Walker, C. A. H. Juffermans and H. Lifka, IEEE Trans. Electron Devices ED-39, 932 (1992). 3. J. R. Schrieffer, in Semiconductor Surface Physics (Edited by R. H. Kingston). np. 5569. Univ. of Pennsvlvania Press, Philadelphia, Pa (1957). 4. J. W. Slotboom and H. C. de Graaf, Solid-St. Electron. 19, 857 (1976). 5. Y. Ohkura, Solid-St. Elecfron. 33, 1581 (1990). 6. J. A. Pals. Phys. Lett. A 39. 101 (1972). 7. F. Stern, CRC Crit. Rev. Sohd Staie Sci:, p. 499 (1974). 8. S. Selberherr, IEEE Trans. Electron Devices ED&, 1464 (1989). 9. M. J. van Dort, J. W. Slotboom, G. Streutker and P. H. Woerlee. in Proc. SISDEP93. Vienna (1993). 10. S. J. Polak, C. den Heyer, W. H. A. Schilders and P. Markowich, Int. J. Num. Meth. Eng. 24, 763 (1987). 11. TMA MEDIC1 1.1, modified version (1992). 12. G. Baccarani and M. Wordeman, IEEE Trans. Electron Devices ED-XI, 1295 (1983).

In silicon devices of deep submicron dimensions, the physical description of the electron motion becomes more complicated. The drift-diffusion approximation can still be used provided some advanced physical models are built in the device simulators. In this paper, we have shown that quantisation effects can be implemented relatively simply without affecting the CPU time. A new model for the silicon