Efficient Multi-scale Self-consistent simulation of planar Schottky-Barrier Carbon Nanotube FieldEffect Transistors and arrays

Tarek M. Abdolkader, Muhammad A. Alam

Abstract A numerical simulation tool for Schottky-Barrier Carbon Nanotube Field-Effect Transistors (SB-CNFETs), based on self-consistent solution of Poisson and continuity equations was developed. Method of moments is used for solving Poisson equation. Continuity equations are solved using current boundary conditions. Boundary currents are those currents tunneling through Schottky barriers at the contacts, which are calculated numerically using transfer matrix method. Regarding long devices, we have exploited the fact that, except near the contacts, the potential on most of the channel length is slowly varying to reduce the computational burden. During the solution of Poisson equation, mapping this part of the device into much smaller length has no perceptible effect on the accuracy of solution; however, it saves much of the simulation time and memory and allows the calculation of transport characteristics inaccessible to classical techniques. Keywords-carbon nanotube; device simulation

Carbon nanotube Top Gate Source Oxide Bottom Gate

Drain

I.

INTRODUCTION

According to the International Technology Roadmap for Semiconductors (ITRS), one needs to explore new technology options to replace silicon in order to sustain device performance beyond the 16 nm generation [1]. Among those technologies, Carbon nanotube Field-Effect Transistors (CNFETs) have attracted significant attention, owing to their excellent electrical, mechanical and optoelectronic properties. High mobility, direct bandgaps and symmetric bands are just a few of the characteristics that make CNFETs promising devices for future electronic applications [2], especially, in aligned array format suitable for scalable use in highperformance transistors [3]. There have been many theoretical simulation studies of ballistic transport in Schottky Barrier CNFETs using Nonequilibrium Green Function (NEGF) [2, 4-5] which is suitable for relatively short devices (< 100 nm). However, for devices longer than few Mean-free-paths (MFP), the transport is essentially diffusive and consequently simulated, more efficiently, within a semi-classical framework. On the other hand, it has been shown that quantum tunneling through Schottky barriers at the tube/metal contacts affect the performance of CNFETs significantly [6]. This demands,

simultaneously, to account for their effect through multi-scale simulation tools which combine both quantum and semiclassical approaches within a single framework. We adopt in this paper this simulation approach following previous work of [7-8]. This multi-scale approach becomes particularly important for aligned array transistors [3] (used for effective, thin film type transistors) with many parallel tubes of broad distribution of diameters: a computationally challenging problem for classical approaches. In this paper, we present a numerically efficient multi-scale simulation tool for planar Schottky-Barrier Carbon Nanotube Field-Effect Transistors (SB-CNFETs) shown schematically in Fig. 1. It is based on self-consistent solution of Poisson and continuity equations, including drift-diffusion transport. Continuity equations are solved in one dimension, with the currents tunneling through Schottky barriers at the contacts are used as boundary conditions. Carrier recombination is modeled assuming only band-to-band, radiative recombination. Fielddependent mobility including velocity saturation is assumed. Regarding Poisson equation, we have chosen the method of moments [9] due its efficiency for non-bulk transport structures. We have used a novel numerical mapping technique to reduce the computational burden while preserving high accuracy of solution. In section II, we will show the basic building blocks of the developed simulation tool, then, in section III, the new proposed technique (Mapping technique) is explained. Results showing the validity of Mapping technique along with

US 978-1-4244-4732-9/10/$25.00 2010 IEEE

dJ n = qR dx
dJ p dx = qR

(2)

(3)

where Jn and Jp are the drift-diffusion electron and hole currents respectively, given as,

J n = qn

d dn + qT dx dx

(4)

J p = qp

d dp qT dx dx

(5)

q is the electronic charge, n and p are the electron and hole concentrations, T is the thermal voltage, and is the electron or hole mobility. R is the recombination rate which is assumed here to be radiative band-to-band recombination,

R = ( np ni2 )

(6)

in which is the recombination coefficient and ni is the intrinsic carrier concentration. Details of solution of continuity equations will be given later in this section. After solving (4) and (5) yielding n and p, we perform variable transformation replacing (n-p) by a dummy quasiFermi level variable fdum(n-p, ) based on the physical band structure of carbon nanotubes. Poisson equation, then, takes the form,

Figure 2. Main building blocks of the developed simulation tool.

2 =

q(n p)

f dum (n p, )

(7)

verifying the operation of the simulation tool are discussed in section IV and conclusion is given in section V. II. MULTI-SCALE SIMULATION

The basic building blocks of the developed simulation tool are shown in Fig. 2. The simulation starts by finding an initial guess for the potential distribution (x) by solving Laplaces equation (assuming no charge in the device).

which becomes nonlinear after introducing fdum. Solution of (7) is accomplished using Newton-Raphson iterative method. The potential resulting from solution is compared to the old potential and the whole process is repeated until the difference between new and old potentials is within a certain tolerance (chosen here to be 0.001 V). The details of the solution of continuity equations are shown in Fig. 3. The solution starts by the calculation of the tunneling currents through Schottky barriers at both source and drain contacts. Tunneling currents are calculated numerically by solving Schrodinger equation using transfer matrix method [11-12], which has been proven to be superior to analytical WKB, or numerical Finite difference methods. Since the two continuity equations are coupled through the recombination term R, they should be solved iteratively. Beginning by electron continuity equation, we assume an initial guess for p, which is basically the intrinsic carrier concentration of the nanotube used. However, in case of the availability of solutions from previous bias points, the most updated value of

2 = 0

(1)

Method of moments [9] is adopted to solve Poisson/Laplace equations as it uses grid points only on the surfaces where charge exists, and not in the entire 3-D domain; Consequently, it will be well suited for simulating planar-gate CNFETs [10]. The resulting potential is used for the solution of electron and hole continuity equations in steady state,

1 1 0 0.5 -1 0 -0.5 -1 -1.5 -2 -2 0 0.05 0.1 0.15 0.2

Ec (eV)

10

x, distance [ m]
Figure 4. Illustration of mapping technique. In the inset, the potential distribution of a pilot small device is shown. The slowly varying potential part in the middle is mapped onto a larger length yielding the potential distribution of the real simulated device (blue curve).

resulting are re-mapped into the smaller length and then fed back to Poisson solver. If the total number of the mesh elements in the pilot and real devices are Np and N, respectively, and if the number of elements in the near-contacts regions is Nc, then, the Mapping factor M will be given by,
Figure 3. Steps for the solution of Continuity equations.

M =

p is used, instead. The solution yields n which is used to solve hole continuity equation resulting in new hole density distribution pnew. This new hole distribution is compared with the old one and the solution is repeated until the percentage difference between old and new solutions is less than 1%. III. MAPPING METHOD

N Nc N p Nc

(8)

For long devices, the key challenge of the multi-scale simulation is the asymmetry of the spatial resolution: the potential on most of the channel length is slowly varying except near the contacts. In this work, we symmetrize the problem by using a novel numerical mapping technique to reduce the computational burden dramatically. The mapping technique is illustrated in Fig. 4: We imagine a pilot device of much smaller length than the real device to be simulated. Solving Poisson equation on this pilot device will produce the potential distribution shown in the inset of Fig. 4. While preserving the portions of the distribution near the contacts, in which the potential is varying relatively rapidly, the slowly varying portion in the middle is mapped/rescaled onto larger length to produce the potential distribution of the real device (blue curve in Fig. 3). Mapping is accomplished by extending the length while preserving the potential gradient (electric field) of the mapped potential. After the solution of continuity equations on the real device length, the carrier concentrations

The overall scaling gain, N/Np will be less than M depending on the bias conditions which determines Nc. The length of the pilot device should be carefully chosen such that it is not so small that the slowly varying portion of the potential disappears, and consequently, mapping cannot be valid. On the other hand, if the pilot length is so large that it is comparable to the real length of the device (Np ~ N), the scaling gain will diminish. From (8), it is apparent that mapping technique is expected to be more effective for very long devices. In fact, above a certain length, the technique limits the computational burden independent of the channel length. IV. RESULTS AND DISCUSSION

The efficiency of the mapping technique is demonstrated through the comparison of simulation times with and without the use of mapping for devices of different lengths, as shown in Fig. 5. All simulations were accomplished using a mesh of 3.4 nm-length elements. For those using mapping, a pilot mapped length of 0.4 m was used. It is evident from the figure that, at least, an order of magnitude reduction in simulation time is gained by using mapping for device lengths > 1 m. Relative improvement in simulation time increases dramatically for longer devices.

10

simulation time [s]

d =1.3nm VG =0.5V 4 10 V =1V D

without mapping with mapping

0.6 0.4 0.2

VD =1V

d =1.3nm

Ec [eV]

-0.2 -0.4 -0.6 VG = 0.1V VG = 0.3V VG = 0.5V 0 2 4 6 8 10

10

10

0.5

1.5

L, Length [ m]
Figure 5. Comparison of the simulation time, with and without mapping technique, for different device lengths.

-0.8

x [ m], distance
Figure 7. The potential energy distribution along the channel of a 10.8 m length device at VD = 1V and VG = 0.1, 0.3, 0.5 V. The diameter of the device d = 1.3 nm.
x 10 15
4

10

-1

n,electron concentration [m-1]

Max error [V]

VD =1V
10
-2

VG = 0.5 V

2500

7.5

10

-3

0.5

1.5

L, Length [ m]
Figure 6. Comparison of the simulation time, with and without mapping technique, for different device lengths.

10

x [nm], distance
Figure 8. Electron (blue) and hole (black) concentrations at three different gate biases VG = 0.1V (solid), 0.3(dashed), and 0.5 V (dotted) along the channel of a 10.8 m CNFET with VD = 1V and diameter d = 1.3 nm.
VD = 0.4V 10
-2

Regarding the accuracy of solution, the potential distribution resulting from simulation using mapping is compared to full-length simulation results as a reference. The maximum error (difference) between the two is plotted versus device length and shown in Fig. 6. It is implied that the maximum error does not exceed 0.02 V. Although error is smaller for short devices, the reduction in simulation time in this case does not worth, as seen in Fig. 5. The operation of the simulator is examined through the verification of different physical phenomena observed in CNFETs. First, the effect of the bias voltages on the potential distribution is viewed in Fig. 7. It is evident from the figure that, except in the vicinity of source and drain contacts, the potential distribution along the device length is slowly varying. This is in agreement with the basic assumption on which the mapping method is based. On the other hand, electron and hole concentrations for the same device are plotted in Fig. 8 with VD = 1 V, VG = 0.1, 0.3, and 0.5 V. With increasing gate bias, the energy bands will be pushed down, allowing more electrons to be injected from the source and less holes injected from the drain.
The ID VG characteristics of the same device is plotted in

VD = 0.8V VD = 1.2V VD = 1.6V

10

-4

I D [ A]

10

-6

10

-8

L =10.8 m d =1.3nm
-0.5 0 0.5 1

-1

VG [V]
Figure 9. The potential energy distribution along the channel of a 10.8 m length device at VD = 1V and VG = 0.1, 0.3, 0.5 V. The diameter of the device d = 1.3 nm.

Fig. 9 for different drain voltages. The work functions of metal

contacts at source and drain are chosen such that the Fermi level on either contact is at the middle of the bandgap of the tube. It is seen from the figure that the drain current exhibits

p,hole concentration [m-1]

d =1.3nm VG =0.5V

5000

VD =1V

VG = 0.1 V VG = 0.3 V

R,recombination rate [m .s-1]

10

x 10

d =1.3nm V =1V
D

using the method of moments and also would like to thank Shuaib Salamat, department of ECE, Purdue University, for extensive technical discussion and supporting information. REFERENCES
[1] [2] ITRS, "EMERGING RESEARCH DEVICES," 2009. Z. Arefinia and A. Orouji, "Quantum Simulation Study of a New Carbon Nanotube Field-Effect Transistor With Electrically Induced Source/Drain Extension," Ieee Transactions on Device and Materials Reliability, vol. 9, p. 237, 2009. X. N. Ho, et al., "Scaling Properties in Transistors That Use Aligned Arrays of Single-Walled Carbon Nanotubes," Nano Letters, vol. 10, pp. 499-503, Feb 2010. N. Neophytou, et al., "Non-equilibrium Greens function (NEGF) simulation of metallic carbon nanotubes including vacancy defects," Journal of Computational Electronics, vol. 6, pp. 317-320, 2007. M. I. Ossaimee, et al., "Ballistic transport in Schottky barrier carbon nanotube FETs," Electronics Letters, vol. 44, pp. 336-337, Feb 28 2008. C. Chen, et al., "Investigation of Schottky-Barrier carbon nanotube field-effect transistor by an efficient semi-classical numerical modeling," Physics Letters A, vol. 374, pp. 309-312, 2009. Jing Guo and M. Alam, "Carrier transport and light-spot movement in carbon-nanotube infrared emitters," APPLIED PHYSICS LETTERS, vol. 86, 2005. D. L. McGuire and D. L. Pulfrey, "A multi-scale model for mobile and localized electroluminescence in carbon nanotube field-effect transistors," Nanotechnology, vol. 17, pp. 5805-5811, Dec 14 2006. S. Ramo, et al., "Fields and waves in communication electronics," book, 1994. N. Neophytou, et al., "Three-dimensional electrostatic effects of carbon nanotube transistors," Ieee Transactions on Nanotechnology, vol. 5, pp. 385-392, Jul 2006. T. M. Abdolkader, et al., "Solution of Schrodinger equation in doublegate MOSFETs using transfer matrix method," Electronics Letters, vol. 40, pp. 1307-1308, Sep 30 2004. T. M. Abdolkader, "A new approach for numerical simulation of quantum transport in double-gate SOI," International Journal of Numerical Modelling-Electronic Networks Devices and Fields, vol. 20, pp. 299-309, Nov-Dec 2007. J. Zaumseil, et al., "Electroluminescence from Electrolyte-Gated Carbon Nanotube Field-Effect Transistors," Acs Nano, vol. 3, pp. 2225-2234, 2009.

-1

6 V G = 0.1V V G = 0.3V V G = 0.5V V G = 0.7V V G = 0.9V 0 2 4 6 8 10

[3]

[4]

x [ m], distance
Figure 10. The recombination rate along the channel of a 10.8 m length device at VD = 1V and VG = 0.1, 0.3, 0.5, 0.7, and 0.9 V. The diameter of the device d = 1.3 nm. Recombination coefficient = 500 m/s.

[5] [6]

minimum at gate voltage VG which is nearly VD /2. This is attributed to the ambipolar behavior of the device due to the symmetry of potential barriers for both electrons and holes. One of the most interesting features of CNFETs is the presence of a light spot that moves along the channel with the change of bias [7]. This light spot arises from the enhanced recombination rate at the region of accumulation of both types of carriers. The recombination rate, R, along the channel is shown in Fig. 10. The value of recombination coefficient used is 500 m/s. R exhibits a peak located near the middle of the channel for ambipolar condition (VG = VD /2) as found in previous theoretical and experimental investigations [8, 13]. This peak is deviated toward the source (drain) with decreasing (increasing) the gate bias relative to the ambipolar bias condition. V. CONCLUSION

[7]

[8]

[9] [10]

[11]

[12]

[13]

A numerical simulation tool for Carbon Nanotube FieldEffect Transistors was developed. It is based on the selfconsistent solution of Poisson equation and continuity equations with drift-diffusion model. A novel numerical technique is used to reduce the computational burden of simulating relatively long devices. It is found that this technique reduces the simulation time by at least an order of magnitude for devices longer than 1 m. The larger is the length of the device, the more percentage saving in time we get. The error in potential distribution due to the use of this method is found to be within 0.02 V. Using the tool, we could explain broad range of physical phenomena of CNFETs including the effect of bias on the electric potential and carrier concentrations, ambipolar behavior, and movement of the light spot with changing bias. ACKNOWLEDGMENT Authors would like to thank Dr. Jing Guo, Assistant Professor at the department of ECE, University of Florida for supplying code needed for the solution of Poisson equation