A Non-Charge-Sheet Analytic Theory for Undoped Symmetric Double-Gate MOSFETs from the Exact Solution of Poisson’s Equation using SPP Approach Jin He, Xuemei Xi, Chung-Hsun Lin, Mansun Chan*, Ali Niknejad, and Chenming Hu jinhe@eecs.berkeley.edu Department of EECS, University of California at Berkeley, CA, 94720 Department of EEE, HKUST, Clear Water Bay, Kowloon, Hong Kong * ABSTRACT A non-charge-sheet based analytic theory for undoped symmetric double-gate MOSFETs is presented in this paper. The formulation is based on the exact solution of the Poisson’s equation to solve for electron concentration directly rather than relying on the surface potential alone. Therefore, carrier distribution in the channel away from the surface is also taken care, giving a non-charge-sheet model compatible with the classical Pao-Sah model. The formulated model has an analytic form that does not need to solve for the transcendent equation as in the conventional surface potentials or Pao-Sah formulation. The validity of the model has also been demonstrated by extensive comparison with AMD double-gate MOSFET’s data Keywords: Double-gate, MOSFETs, Compact modeling. 1 INTRODUCTION As CMOS scaling continue to beyond the 50nm node, undoped (or lightly-doped) double-gate MOSFETs have become the most promise candidate device structure due to a number of unique features such as ideal 60 mV/decade subthreshold slope, volume inversion, free-dopantassociated fluctuation effects and so on . Compact modeling of this structure has been extensively studied. Most of these models, however, have to rely on the charge-sheet approximation. For thick film devices, carrier distribution with near a zero-thickness sheet can capture the essential part of double-gate MOSFET’s characteristics. For an ultra-thin double gate MOSFET, the charge-sheet approximation cannot be used because the inversion layer thickness is comparable with silicon film thickness and a number of results from the charge-sheet approximation become invalid. In a recent work, a one-dimensional (1-D) PoissonBoltzamnn equation was solved for the double-gate MOSFET to derive analytical solution for surface potential and inversion charge density by Prof.Taur’s group. This result can serve as a core foundation for developing noncharge-sheet-based model. Due to the numerical complexity and the need to solve a number of transcendental equations, the model is difficult to be implemented in circuit simulator. A simpler and yet equally accurate model is thus desired, not only for efficiency of simulation, but also to provide insight into the operation of the double-gate MOSFETs. In this work, we have derived a non-charge-sheet based analytic theory for undoped symmetric double-gate MOSFETs by solving the exact carrier concentration based on both the surface potential and the quasi-Fermi potential in the channel to account for the vertical potential profile and distribution of the carrier. We referred this approach as surface potential plus (SPP), which is compatible with the Pao-Sah formulation. The analytical model has also been verified by AMD double-gate MOSFET’s data. 2 ANALYTICAL SOLUTION Fig. 1 shows the coordinate and energy level distribution diagrams of a symmetric double-gate MOSFET. We assume that the same voltage is applied to the two gates having the same work function. The formulation starts with the 1-D Poisson's equation along to the vertical direction of the silicon channel considering only the mobile charge (electron) density for the undoped body. d 2φ qn (1) = dx ε 2 si where q is the electronic charge, εSi is the permittivity of silicon, and n is the intrinsic carrier density. According to Boltzmann statistics, the mobile electron concentration can be expressed in term of potential § q (φ − v ch · § q ( φ 0 − v ch · (2) n = n i exp ¨ © kT ¸ n 0 = n i exp ¨ ¹ © kT ¸ ¹ The spatial derivatives of the electron from (2) are dφ kT dn = dx qn dx d 2φ kT d 2 n kT § dn · = − ¨ ¸ dx 2 qn dx 2 qn 2 © dx ¹ (3) 2 (4) Substitution (4) into (1) gives an equation for electron concentration 2 d 2 n 1 § dn · q 2n 2 (5) = + ¨ ¸ dx 2 n © dx ¹ ε kT This normal differential equation has two mathematical solutions, one is a trigonometric function and another is a hyperbolic function. (6) n( x ) = c0 c0 n( x ) = 1/ 2 ª§ q 2 c ·1 / 2 º ª º § q 2c0 · 0 ¸¸ x» ¸¸ x » cos 2 «¨¨ cosh 2 «¨¨ «¬© 2εkT ¹ «¬© 2εkT ¹ »¼ »¼ Based on a physically reasonable consideration, the former 124 NSTI-Nanotech 2004, www.nsti.org, ISBN 0-9728422-8-4 Vol. 2, 2004 solution is taken. We choose the reference coordinate point as x = 0 and n( x ) = no , (6) is further simplified into (7) n (x n0 )= ª§ q 2n 0 « ¨¨ «¬ © 2 ε kT 2 cos · ¸¸ ¹ 1 / 2 º x» »¼ Substitution of (7) into (1) gives the corresponding electrical field and potential distributions in silicon film as 1/ 2 ª 2 º (8) kT −2 § q n0 · φ (x ) − φ 0 = ¸ « ¨¨ 2ε kT ¸¹ ¬« © ln cos q ª 2 n 0 kT º E (x ) − E ( x 0 ) = « » ¬ ε ¼ 1/ 2 x» ¼» ª§ q 2n 0 tan « ¨¨ «¬ © 2 ε kT · ¸¸ ¹ 1/ 2 º x» »¼ (9) The symmetry boundary condition of a symmetric double gate MOSFET set the electric field in the center of the silicon film to be zero. Thus, the reference coordinate takes the center of the film as the point of x = 0. The surface potential and the surface electric field are then given by φ s = v ch + ªn kT ln « 0 cos q « ni ¬ ª 2 n 0 kT º Es = « » ¬ ε ¼ 1/ 2 −2 ª§ q 2 n 0 tan « ¨¨ «¬ © 2 ε kT If we define Q = q in ª§ q 2 n 0 «¨ «¬ ¨© 2 ε kT · ¸¸ ¹ 1/ 2 · ¸ ¸ ¹ 1/ 2 T si º º »» 2 »» ¼¼ T si º » 2 » ¼ (10) (11) ³ n(x )dx as half of the total inversion charge, it can be expressed ]1 / 2 ª§ q 2 n 0 tan « ¨¨ «¬ © 2 ε kT · ¸ ¸ ¹ 1/ 2 T si º » 2 » ¼ (12) ε ox Substituting the surface potential and inversion charge into (13) results gives 1/ 2 1/ 2 ª§ q2n ·1/ 2 T º (14) kT ªn0 −2 ª§ q2n0 · Tsi ºº εsi ª2n0kTº 0 si ¸ ln« cos «¨¨ q «ni «¬© 2εkT¸¹ ¬ ¸ »» + tox tan«¨¨ 2 »» εox «¬ ε »¼ «¬© 2εkT¸¹ ¼¼ » 2» ¼ Eq.(14) is the exact closed form expression for electron concentration at the center of the channel silicon (n0) as a function of gate voltage, channel voltage, and silicon film. This expression is useful because of its exactness, but too complex to be implemented in compact model due to the transcendental equation. We will next simplify Eq. (14) to a more manageable form. The inversion charge expression given in Eq. (12) can be also be expressed as (15) ª§ q n · T º 2 ª§ q 2 n · 0 Q = 2ε si n0 kT tan «¨¨ ¸¸ «¬© 2εkT ¹ 1/ 2 2 in 2 2 1/ 2 0 sin 2 «¨¨ ¸¸ «¬© 2ε si kT ¹ 2 2 0 si in si ¸ si ≈ »≈¨ exp( f ) 2 » ¨© 2ε si kT ¸¹ 4 2ε si kT 2 ¼ This approximation works well for ultra-thin silicon case, leading to a negligible error. In fact, we will find in the following expression that this sine function term comes into effect in the sub-threshold region. In order to compensate the error leaded by this approximation, we use a dimensionless correction factor f . This factor should be a step function of the gate voltage, e.g. equal to unit below the threshold point and zero above the threshold point. An exact approximate solution of this factor is written as 2 (17b) ª tanh (v G − ∆ φ i ) º f = « ¬ v G − ∆ φ » ¼ i via comparison with the numerical result and 2-D simulation. In this case, we have 1/2 (17c) 2 ª · T º qQ T 2 § q n sin 0 « ¨¨ ¸¸ «¬ © 2 ε si kT ¹ si in » = 2 » 2 ε si kT ¼ si 2 exp (f ) Making use of this approximation, (16) is written as The surface potential, field and carrier concentration are controlled by the applied gate voltage. According to Gauss’s law, the gate voltage is Q (13) VG − ∆ψ i = φS + Eox tox = φ s + in tox VG −∆ψi −vch = On the other hand, since a tangent function of the inversion charge expression is mainly determined by cosine function, an popular approximation is used to substitute sine function that can be combined with Eq. (16). (17a) ª§ q n · T º § q n · T qQ T T si / 2 0 Q in = [2 ε n 0 kT region where the potential profile in the vertical direction is almost flat and resulting in volume inversion. In the subthreshold region case, the entire channel region contributes to conduction, and the inversion charge is almost proportional to the silicon film thickness. Therefore, we can express the inversion charge as (16) Q in = qn 0 T si / 2 1/ 2 0 si » sin 2 «¨¨ ¸¸ «¬© 2ε si kT ¹ 2 »¼ Tsi º » = 2ε si n0 kT 1/ 2 ª§ q 2 n · T º 2» ¼ 0 si ¸¸ » cos 2 «¨¨ «¬© 2ε si kT ¹ 2 »¼ We will use the inversion charge expression to eliminate the surface potential term in Eq.(13). This surface potential term comes into effect only in the sub-threshold Qin = (18) qn0Tsi / 2 exp( f ) ª§ q 2 n · 1 / 2 T º 0 si ¸ » cos 2 «¨¨ 2εkT ¸¹ 2» ¼ ¬«© Substitution of (18) into (13) and replacing the surface potential gives fkT kT Qin (19) VG − ∆ψ i − vch + q ln(qniTsi / 2) = q ln Qin + ε ox / tox This inversion charge equation is similar to that given by the bulk SPP model, except the sub-threshold slope n is replaced with by 60mV at room temperature. Thus, a similar approach as in the SPP can be used to obtain a consistent I-V and C-V characteristics To simplify subsequent derivation, all voltages are normalized by kT / q and inversion charge is normalized by Cox kT / q . (19) becomes after simplification § q 2 niTsitox · ¸¸ = ln qin + qin vG − ∆φi − vch + f ln¨¨ © 2kTε ox ¹ (20) The inversion charge in (20) is solved explicitly in terms of the Lambert W function to yield the following exact expression. ª § § q 2niTsitox · ·º (21) ¸¸ ¸» qin = W0 «exp¨¨ vG − ∆φi − vch + f ln¨¨ ¸ 2 kT ε ox ¹ ¹ ¼ » © ¬« © NSTI-Nanotech 2004, www.nsti.org, ISBN 0-9728422-8-4 Vol. 2, 2004 125 where W0 stands for the usual short-hand notation used for the principal branch of the ‘‘Lambert-W’’ function. The exact solution that we propose makes use of the principal branch of the Lambert W function which is defined as the solution to the equation Wew = x . The Lambert W function is a popular function in numerous physics applications. It has also been used to provide solutions to previously unsolved basic diode and bipolar transistor circuit analysis. Once the inversion charge is obtained, the I-V and C-V characteristics of undoped double gate MOSFET can be directly formulated. We next proceed to describe the modeling of the IV characteristics. According to Pao-Sah model, the current including the diffusion and drift component is written as 2 dv ch § kT · (22) ¸¸ q in I ds = µ wC ox ¨¨ q dy © ¹ For constant gate voltage, the differential of (20) can be written as dv ch = dq in / q in + dq in (23) here we neglect the derivative contribution of f factor since it is a step function. Substitution of (23) into (22) gives I ds § kT = µ wC ox ¨¨ © q 2 · ¸¸ ( q in dq in + dq in ) ¹ (24) As a result, an analytical current expression is obtained. 2 º § kT · ª q s2 − q d2 µw (25) ¸ + (q − q ) C ¨ I = ds ox L ¨ q ¸ « ¹ ¬ © 2 s d » ¼ As the charge formulation only accounts for half of the channel, the final current of a double-gate MOSFET should be doubled. Note that the final current expression is similar to that of the SPP derived by the bulk model. Thus, a unified framework can be used for bulk and the undoped symmetric double-gate MOSFETs. This results in simple methodology to obtain both the I-V and C-V characteristics with already formulated such as gate tunneling current, short channel effect, QME and non-uniform lateral body doping [19]. In addition, the unified approach to treat the different device structures significantly simplify the study of the relative advantages of different device structure to be used to extend the CMOS scaling roadmap to the nanometer regime. . 3 RESULTS AND DISCUSSION 3.1 Electron and potential in the silicon film Fig.2 illustrates the electron concentration and electrical potential distribution in the silicon film predicted by the analytical model for three different effective gate voltages. In low gate voltage or the subthreshold region, the electron distribution is almost constant along the vertical direction of the film as predicted by Qin = qn0Tsi / 2 . With the increase of the effective gate voltage, the device goes into the strong inversion region where the surface electron concentration shows several order higher than that of the silicon film center. In this case, device behaves like a 126 surface channel bulk MOSFET and the surface electron concentration dominates the inversion charge density. 3.2 Surface potential and inversion charge dependence on the bias Fig.3 shows the carrier charge per unit area in the channel versus the applied gate voltage, as calculated with (20), for three values of channel voltage. There are two distinct regions of operation in the undoped symmetric double-gate MOSFET, just like in a conventional MOSFET as shown in Fig.3(b). One significant result is the relation between the inversion charge density and the channel voltage. In contrast to the gate voltage, Fig.3 indicates that the channel voltage increases the inversion charge correspondingly decreases. One interesting observation is that the equivalent dielectric thickness has a slight effect on the surface potential in strong inversion region and has almost no effect on the sub-threshold region in the undoped bulk MOSFETs, as shown in Fig.4. This physical picture is in different from that of a conventional bulk MOSFETs where the sub-threshold region slope has a strong dependence on the oxide thickness. 3.3 Volume inversion Fig.5 demonstrates the effect of the silicon film thickness on the inversion charge versus gate voltage characteristics. The silicon film thickness affects the amount of sub-threshold inversion charge but has no effect on the strong inversion charge density. This verified the existence of the volume inversion and implies that decrease of the silicon thickness can effectively control the subthreshold region leakage current. It illustrates a potential weakness in the used of undoped symmetric double MOSFETs for nano-CMOS application. To optimize the device performance, the silicon film thickness should be reduced to suppress the off current. 3.4 I-V characteristics In order to verify the validity of the analytical model, we use this model to simulate a well-temped undoped double-gate MOSFET with the channel length of 10µm and gate oxide thickness of 3nm. A long channel device is used to avoid the complicated short channel effect to hinder the detail picture of 1-D analysis in the vertical direction of the channel. Short channel effects will be considered in future work. Fig.6 and Fig.7 show the DG MOSFET current voltage characteristics predicted by the derived analytical model with the constant mobility of 400 cm 2 / V .s . The model is continuous from the sub to the strong inversion region at all drain bias conditions, as shown in Fig.6. One observation is that the drain voltage has no effect on the sub-threshold current but increases the strong inversion current. The saturation characteristics given by the model is also observed from Fig.7. Fig. 8 shows this model fitting for AMD double-gate MOSFET’s data by using SPP modules on SCE, DIBL, QME, ect. In DG model. A good agreement is found between both. NSTI-Nanotech 2004, www.nsti.org, ISBN 0-9728422-8-4 Vol. 2, 2004 CONCLUSION 1 Tox=2nm -3 10 20 -5 10 10 -7 10 -9 10 0.0 0.3 0.6 0.9 Normalized inversion charge 30 -1 10 VGate=1V, Tox=2nm Normalized inversion charge Normalized inversion charge In this paper, we have developed a non-chargesheet based analytical theory for undoped symmetric double-gate MOSFET. We have derived the exact analytical solution of the electron concentration and density as an explicit function of the gate voltage and silicon film, thickness valid for all device operation regions. The result agrees well with the Pao-Sah current formulation and physically accurate. The presented model works with a more generic SPP framework that can be used to model a wide range of devices to be used in nano-CMOS technology and been verified by AMD double-gate data. 15 -2 10 Tsi=10nm Tsi=25nm Tsi=35nm -5 10 10 5 -8 10 0 -11 0 1.5 1.2 20 10 40 Tsi=6nm Tsi=25nm Tsi=35nm 1 10 10 0.0 0.2 Gate voltage [ V ] 0.4 0.6 0.8 1.0 1.2 channel voltage [ V ] Fig.5 The mobile carrier density as a function of gate voltage and channel voltage with the different silicon thickness for undoped DG MOSFETs. 10 -4 -5 8.0x10 Vds=0.8V Vds=0.8V Vds=0.5V Drain 10 -5 10 -6 10 -7 Vds=0.1V -5 6.0x10 Drain current [ mA ] Drain current (mA) Gate Oxide Source Normalized inversion charge 4 Oxide -5 4.0x10 Vds=0.5V -5 2.0x10 Gate -0.2 0.0 0.2 0.4 0.6 0.8 Vds=0.1V 1.0 0.0 -0.2 Fig.1. Diagrams of the structure and energy levels of undoped double-gate MOSFETs. 10 18 10 17 VG − ∆ψ V V 10 G G − ∆ ψ i − ∆ψ i -6 -4 -2 0 10 19 10 18 10 17 10 16 = 0 . 5 7 4V i = 0 . 4 8 2 1V = 0 .4 0 9 4V 16 -8 10 2 4 6 8 0.5 0.4 0.3 -8 -6 -4 -2 0 2 4 6 1 .2 x 1 0 -4 1 .0 x 1 0 -4 8 .0 x 1 0 -5 6 .0 x 1 0 -5 4 .0 x 1 0 -5 2 .0 x 1 0 -5 V g = 1 .2 V I V g = 0 .9 V V g = 0 .6 V 0 .0 0 .0 0 .5 1 .0 ds 2 .0 2 .5 [ V ] T ox =2nm, T si=30nm 1 25 Vch=0.5V Fig.7. The drain current versus the drain voltage with the different gate voltage for a 10um DG undoped MOSFETs. V ch=0.5V Vch=0.3V Qin/CoxVt Qin/CoxVt 1 .5 V 30 0 1.0 8 Position in silicon film , x [um ] Fig.2. The mobile electron and potential distribution as a function of the silicon film thickness for undoped MOSFETs with symmetric double-gate structure. 10 0.8 V g = 1 .5 V VG −∆ψi = 0.4094V Position in silicon film, x [ um ] 10 0.6 VG −∆ψi =0.574V VG −∆ψi =0.4821V (mA) 19 0.4 Fig.6. The drain current versus the gate voltage with the different drain voltage for a 10um undoped MOSFETs with symmetric double-gate structure. T si=16nm , T ox=2nm 0.6 0.2 Vg [ V ] ds 10 20 Electric potential [ V ] 10 -3 Electron concentration [ cm ] T si=16nm, T OX=2nm 20 0.0 Vg [ V ] Vch=0.1V 20 V ch=0.3V 15 V ch=0.1V Tox=2nm, Tsi=30nm 10 0.6 0.9 1.2 1.5 0.3 Vg [ V ] 0.6 0.9 Vg [ V ] 1.2 10 -5 10 -6 10 -7 10 -8 0.6 Surface potential [ V ] Vch=1.5 V 2.0 Vch=1.0 V 1.5 Vch=0.5 V 1.0 Vch= 0.0 V 0.5 10 -4 L g=55nm Tsi= 26nm Lg=80nm Tsi= 26nm Tox=1nm Tox=2nm Tox=5nm 10 -9 10 -5 10 -6 10 -7 0.0 0.5 1.0 0.4 -5 10 -6 10 -7 10 V D=0.1V Data V D=1.2V Data MDE simulation results -8 Gate Voltage (V) 0.5 10 V D=0.1V Data V D=1.2V Data MDE simulation results V D=0.1V Data V D=1.2V Data MDE simulation results 0.7 2.5 -4 1.5 Fig.3. The mobile carrier density as a function of gate voltage with the different channel voltage for DG undoped MOSFETs. Surface potential [ V ] 10 Drain Current (A) 0.3 -4 Lg=105nm Tsi= 26nm 0 0.0 Drain Current (A) 0.0 10 5 Drain Current (A) -1 10 0.0 0.5 1.0 Gate Voltage (V) 10 -8 0.0 0.5 Gate Voltage (V) Fig.8 Comparison between the analytical model and AMD double-gate data. 0.3 0.2 0.1 0.0 0.0 0.5 1.0 1.5 Vg [ V ] 2.0 2.5 3.0 0.0 0.0 0.3 0.6 0.9 Vg [V] 1.2 1.5 Fig.4 Surface potential versus gate voltage with the different oxide thickness and channel voltage for DG undoped MOSFETs. Acknowledgement: This work was supported by SRC under the contract number 2002-NJ-1001 and partially supported by a grant from RGC of Hong Kong under the number HKUST6111/03E. We also thank AMD Corporation for providing experimental data and Prof. Taur from UC of San Diego for useful discussion. NSTI-Nanotech 2004, www.nsti.org, ISBN 0-9728422-8-4 Vol. 2, 2004 127 1.0