Nanotube for Global Interconnects Manoj Kumar Majumder #1 , B. K. Kaushik *2 , S. K. Manhas #3
# Department of Electronics and Computer Engineering, Indian Institute of Technology-Roorkee, Roorkee, INDIA 1 manojbesu@gmail.com, 2 samanfec@iitr.ernet.in, 3 bkk23fec@iitr.ernet.in
Abstract- Multi-wall carbon nanotubes (MWNTs) have potentially provided an attractive solution over single-wall carbon nanotube (SWNT) bundles at deep sub-micron level very large scale integration (VLSI) technologies. This paper presents a comprehensive analysis of propagation delay for both MWNT and SWNT bundles at different interconnect lengths (global) and shows a comparison of area for equivalent number of SWNTs in bundle and shells in MWNTs. It has been observed that irrespective of the type of CNTs, propagation delay increases with interconnect lengths. For same propagation delay performance, the area occupied by SWNT bundle is more than the MWNTs for a specified interconnect length.
Keywords Carbon nanotube, propagation delay, area, VLSI, NOC, nanotechnology. I. INTRODUCTION Continuous advances in VLSI technologies leads to development of more complex chips which contains millions of interconnections that integrate the components on the IC chip. At present VLSI technology, interconnection plays most significant role in determining the size, power consumption and clock frequency of digital system. Advancement of technological development leads for the solution of long path interconnects, which is known as global interconnects. For this global level, conventional interconnect technologies used by copper or aluminium fails because of their increasing resistively with length which may cause some serious problems like electro migration and hillocks or voids formation in the successive levels of interconnect paths [1]. For this reason, researchers like to introduce some new materials as carbon nanotubes (CNTs) or graphene nanoribbons (GNRs) which may be a possible solution for future VLSI technologies of global interconnects. CNTs are known as allotropes of carbon [2] and made by rolling up a sheet of graphene into a cylinder. Graphene is a monolayer sheet of graphite with sp 2 bonding of carbon atoms arranged in a honeycomb lattice structure. The sp 2 bonding in graphene is stronger than sp 3 bonds in diamond [3] which makes graphene the strongest material. CNTs have not only unique atomic arrangement but also interesting physical properties [4]; including current carrying capability [5], long ballistic transport length [4], high thermal conductivity [6], and mechanical strength [7]. These remarkable properties make CNTs one of the most promising research materials for future VLSI technology. The extraordinary physical properties of CNTs make them exciting prospects for a variety of applications in microelectronics/ Nan electronics [8], spintronics [9], optics [10], material science [11], mechanical [12] and biological fields [13]. Particularly, in the area of nanoelectronics, CNTs and GNRs show their prospects as energy storage devices (such as super capacitors [14]), energy conversion devices (including thermoelectric [15] and photovoltaic [16] devices), field emission displays and radiation sources [17], nanometer semiconductor transistors [18], nanoelectromechanical systems (NEMS) [19], electrostatic discharge (ESD) protection [20], as well as interconnects [21, 22] and passives [23]. Structure of CNTs depends on chiral indices i.e., the rolling up direction of a graphene sheet. Depending on the chiral indices (n, m), CNTs can get their unique armchair or zigzag structures. For armchair CNTs, the chiral indices is defined by n = m [3] and for zigzag CNTs, it is n or m = 0 [3]. For other values of n and m, CNTs are known as chiral. Depending upon their different structures, CNTs can exhibit both metallic and semiconducting properties. By satisfying the condition n m =3i (where i is an integer), the armchair CNTs are always metallic and zigzag CNTs are either metallic or semiconducting in nature depending on their chiral indices [3, 4]. Statistically, a natural mix of CNTs will have 1/3 rd of metallic and 2/3 rd of their semiconducting chirality [24]. These unique structures of CNTs basically depend on the rolling up directions of graphene sheet. CNTs are also classified into single-walled nanotubes (SWNTs) and multi-walled nanotubes (MWNTs). The main difference between SWNT and MWNT arises from number of concentrically rolled up graphene sheets. For SWNTs, only one rolled up graphene sheet is there whereas for two or more concentrically rolled up graphene sheets, CNTs are known as MWNT. SWNT bundle is formed by packing together a large number of SWNTs in a bundle form. The parasitic elements of SWNT bundle and MWNTs depend on number of metallic SWNTs in the bundle and number of shells in the MWNTs. It is very difficult to achieve ballistic transport for MWNTs for a long length whereas 978-1-4577-0240-2/11/$26.00 2011 IEEE 104 SWNTs have mean free path of the order of a micron [25] due to its diameter in nano range. The paper is organized as follows. In section II, different interconnect models of SWNT bundle and MWNT are reviewed. Section III introduces simulation setup for calculating propagation delay on the basis of described models in section II. Graphical plots of propagation delay for MWNT shells with equivalent numbers of SWNTs in bundles at different global interconnects lengths are presented in section IV. Finally, section V draws a brief summary. II. INTERCONNECT MODELS OF SWNT BUNDLE AND MWNTS Depending on different interconnect parasitics such as resistance, capacitance and inductance, RLC circuit models (Fig. 1 and 5) for both MWNT and SWNT bundle are presented [26, 27] on the basis of semi-classical one- dimensional electron fluid model [28, 29]. This 1-D electron fluid model is derived from a classical two-dimensional electron fluid theory [4, 30] taking into account electron-electron repulsive force. The MWNT interconnect model presented in [26, 27] is defined in terms of number of shells in MWNTs which is formulated as [27], - 1 2 D D I N M ( = + (
(1) Where DI and DN are the outermost and innermost shell diameters of MWNT, respectively. is the spacing between shells in a MWNT and the value is defined as = 0.34 nm [26]. The square bracket term of equation (1) is a floor function and the factor is the ratio of metallic shells to those shells in a MWNT. When DI 10 nm [27], then one-third of the shells are metallic and the rest of the shells are semiconducting. In the model [27] of Fig.1, magnetic inductance is neglected compared with kinetic inductance as the value of magnetic inductance is much smaller than kinetic inductance for a CNT [25]. The kinetic inductance per unit length is formulated by approximating the kinetic energy per unit length of a 1-D wire [3, 4, 29] and expressed by the equation [25],
2 2 h L K e v F = (2) where h is the planks constant, e is electron charge and v F
is the Fermi velocity of graphene (= 810 5 m/s). Therefore, value of kinetic inductance of a CNT is given by 16 nH/ m [25]. Due to band structure, SWNT has two propagating channels and for each channel, electrons have the properties of spin up and spin down [4]. For this fourfold band degeneracy, kinetic inductance per unit length is arrived as 4 nH/ m [31]. In the given model of Fig.1 electrostatic capacitance (C E ) does not exist in inner shells of MWNTs as outermost shells of MWNTs shields inner shells from the ground plane. But, one additional electrostatic capacitance (C S ) between the neighboring metallic shells and can be formulated as [26, 27],
2 0 l n C S D i D j | | | | \ .
= (3) where 0 is the permittivity of the vacuum, D i and D j are the diameters of the i th and j th metallic shells, respectively and i < j. The circuit model presented in Fig.1 is simplified in Fig. 2 by considering identical inner shells of MWNT. In the simplified model presented in Fig. 2, it is assume that there are no variation in distributed parameters, R and L k as they are same for each shell of MWNT. Now, by assumption of equal potential across components of each shell in a MWNT, the circuit model of Fig. 2 is again simplified to an equivalent circuit model which is presented in Fig. 3. For all the circuit models presented in Figs [1-3], R C is the contact resistance and its ideal value is defined as 3.2 K per shell [26]. The values of C E and C Q are of the same order. C Q appears in parallel for all metallic sheets with a series combination of C E . Therefore, for M number of metallic sheets C Q appears as (M C Q ) and the value of C E appears as in the previous case.
Figure 1. Equivalent circuit of a metallic MWNT interconnects [27].
Figure 2. Simplified equivalent circuit of a metallic MWNT interconnects [27]. The RLC model of SWNT bundle [27, 32] considers all individual SWNTs at parallel directions in the bundle. The spacing between SWNTs in the bundle is determined by vander waals forces between the atoms in adjacent nanotubes. The metallic nanotubes are distributed randomly with a probability = 1/3 due to the lack of chirality [27]. The 105 proportions of metallic nanotubes are therefore potentially be increased using techniques reported in [33] and [34]. Therefore, total number of metallic SWNTs is formulated as shown in Fig. 4 [27].
R C /M R/M L k /M M X C Q C E
Figure 3. Simple equivalent circuit model of a metallic MWNT interconnect
2 N y N N N x y ( ( ( ( ( ( ( (
= (4) where N x and N y are the number of SWNTs in the bundle at horizontal and vertical directions, respectively. The distance between adjacent SWNTs in bundle can be derived from the cross-section of a bundle (Fig. 4) as [27],
b d d =+ (5) where = 0.34 nm is the spacing between the SWNT in the bundle and d is the diameter of SWNT. The equivalent circuit of SWNT bundles interconnect is shown in Fig. 5 where Na and Nb denote the number of upper level and bottom level SWNTs (Fig. 4) [27].
Figure 4. Cross-section of a SWNT bundle interconnects [27]. For the RLC model of SWNT bundle presented in Fig. 5, the assumption is made for all SWNTs in the bundle are identical and each SWNT has same potential across it [27]. The equivalent circuit of Fig. 5 is further simplified in Fig. 6, where electrostatic capacitance between neighbouring metallic SWNTs (Cb) has no effect on the circuit behaviour and (Nx CE) can be regarded as an electrostatic capacitance between SWNT bundle and ground plane. In the upper level of SWNTs, CE does not exist as the bottom level shields the upper levels SWNTs from the ground plane, as shown in Fig. 5. Quantum capacitance (CQ) arises as 4CQ due to fourfold band degeneracy structure of SWNT [4]. The values of CE and CQ are nearly same in magnitude. CQ of all metallic SWNTs are oriented in parallel and then is serial with CE. Therefore, CQ can be neglected for large value of N. The resistance and inductance of all metallic SWNTs are in parallel in the bundle and N times smaller than that of a SWNT. R C /N a R/N a R C /N b R/N b L k / N a L k / N b C Q (N b ) C E (N b ) C b C Q (N a ) R C /N b R C /N a 2 ~ N Y 1
Figure 5. Equivalent circuit of a SWNT bundle interconnects [27]. R C /N R/N L k /N NX C Q N X X C E
Figure 6. Simplified equivalent circuit of a SWNT bundle interconnects III. SIMULATION SETUP FOR CALCULATING PROPAGATION DELAY In this paper, propagation delay for SWNT bundle and MWNT is measured at different interconnect lengths of 100, 200, 500, 1000 and 2000 m by using the geometries suggested in [26, 27]. Simulation setup uses CMOS inverter at 32 nm technology node for which the technology parameters (length and width) of NMOS is taken as 32 nm and 640 nm and for PMOS, these parameters are taken as 32 nm and 1280 nm, respectively. As shown in Fig. 7 and Fig. 8, the input rise time is triggered at 90% whereas the output fall time is target at 10%. The delay is analysed for different number of shells in
R C /N R/N L k /N Propagation Delay O/P I/P N X C Q N X X C E
Figure 7. Simulation setup for SWNT Bundle interconnects. 106 MWNTs and for different number of SWNTs in the bundle. For each type of MWNT shells and SWNT bundles,the propagation delay is measured for 20 distributed elements at different interconnect lengths. R C /M R/M L k /M Propagation Delay O/P I/P MX C Q C E
Figure 8. Simulation setup for MWNT interconnects. IV. RESULTS AND DISCUSSIONS In this paper, different simulation results for propagation delay are presented for MWNTs and SWNT bundles at different interconnect lengths (global). A comparative analysis for SWNT bundles and MWNTs are also presented here in terms of propagation delay at different interconnects lengths. Fig. 9 presents the propagation delay of different numbers of MWNT shells for different interconnect lengths whereas Fig. 10 shows the same analysis for different number of SWNTs in bundle. In both cases, it is shown that propagation delay for both MWNT and SWNT bundles are increased with interconnect lengths but decreased with the increasing number of shells and number of SWNTs in bundle, respectively. By analysing the simulation results, a comparative study is done between the number of shells in MWNTs and number of SWNTs in the bundle. This comparative study will show the equivalent number of SWNTs in the bundle for a given number of shells in MWNTs at specified interconnect length (global).
0 500 1000 1500 2000 -50 0 50 100 150 200 250 300 350 400 P r o p a g a t i o n
D e l a y
( n s ) Interconnect Length (m) No of shells=4 No of shells=10 No of shells=20
Figure 9. Propagation delay of MWNTs for different numbers of shells with varying interconnects lengths (global). For a certain propagation delay, the equivalent numbers of SWNTs in bundle and shells in MWNTs for different interconnect lengths of 100, 200, 500, 1000 and 2000 m are graphically presented in Figs. 11, 12, 13, 14 and 15, respectively. Finally, these equivalent numbers of MWNTs and SWNT bundles at different interconnect lengths are graphically presented in Fig. 16.
The area is analysed for equivalent number of shells in MWNTs and number of SWNTs in the bundle (Fig. 17) in terms of diameters of SWNT bundle and MWNT. The areas of MWNTs depend on the number of shells and innermost and outermost diameters which is obtained by using equation (1). From this graphical analysis, it is clear that the area is minimized for equivalent shells of MWNTs than the equivalent number of SWNTs in bundle.
0 500 1000 1500 2000 0 100 200 300 400 500 600 P r o p a g a t i o n
D e l a y
( n s ) Interconnect Length(m) SWNTs=4 SWNTs=9 SWNTs=13 SWNTs=17 SWNTs=25 SWNTs=58 SWNTs=99 SWNTs=105 SWNTs=111 SWNTs=163 SWNTs=660
Figure 10. Propagation delay of SWNT bundle with varying interconnects lengths (global).\
107 Figure 11. . Equivalent number of shells of MWNTs for different number of SWNTs in bundle at specified interconnect length of 100m for same performance of propagation delay.
Figure 12. Equivalent number of shells of MWNTs for different number of SWNTs in bundle at specified interconnect length of 200m for same performance of propagation delay. 2 4 6 8 10 12 14 16 18 2 4 6 8 10 12 14 16 18 20 22 MWNT SWNT Bundle Propagation Delay (ns) N o .
o f
S h e l l s
( M W N T ) 0 100 200 300 400 500 600 700 N o .
o f
S W N T s
i n
B u n d l e
Figure 13. Equivalent number of shells of MWNTs for different number of SWNTs in bundle at specified interconnect length of 500m for same performance of propagation delay. 10 20 30 40 50 60 70 80 2 4 6 8 10 12 14 16 18 20 22 MWNT SWNT Bundle Propagation Delay (ns) N o .
o f
S h e l l s
( M W N T ) 0 100 200 300 400 500 600 700 N o .
o f
S W N T s
i n
B u n d l e
Figure 14. Equivalent number of shells of MWNTs for different number of SWNTs in bundle at specified interconnect length of 1000m for same performance of propagation delay. 0 50 100 150 200 250 300 350 400 2 4 6 8 10 12 14 16 18 20 22 MWNT SWNT Bundle Propagation Delay (ns) N o .
o f
S h e l l s
( M W N T ) 0 100 200 300 400 500 600 700 N o .
o f
S W N T s
i n
B u n d l e
Figure 15. Equivalent number of shells of MWNTs for different number of SWNTs in bundle at specified interconnect length of 2000m for same performance of propagation delay.
2 4 6 8 10 12 14 16 18 20 22 0 20 40 60 80 100 120 E q u i v a l e n t
S W N T s
i n
b u n d l e No. of Shells (MWNT) Interconnect length = 100 m Interconnect length = 200 m Interconnect length = 500 m Interconnect length = 1000 m Interconnect length = 2000 m
Figure 16. Equivalent number of SWNTs in the bundle for different numbers of shells in MWNTs at different interconnects lengths (global) for same performance of propagation delay. 0 200 400 600 800 1000 1200 1400 1600 1800 0 200 400 600 800 1000 1200 1400 1600 1800 2000 E q u i v a l e n t
A r e a
o f
S W N T
B u n d l e
( n m 2 ) Area of MWNTs (nm 2 ) Interconnect length = 100m Interconnect length = 200m Interconnect length = 500m Interconnect length = 1000m Interconnect length = 2000m
108 Figure 17. Analysis of area for different number of shells in MWNTs with equivalent number of SWNTs in bundle at different interconnect lengths (global). V. CONCLUSION In this paper, models for MWNT and SWNT bundle interconnects are reviewed on the basis of one-dimensional fluid theory. These models are compared at 32 nm technology nodes in terms of propagation delay and area. Simulation result shows that MWNT requires lesser area than SWNT bundle at different interconnect lengths for same performance of propagation delay. Therefore, for future VLSI technology, MWNTs may be proven as more promising candidate than copper. REFERENCES [1] A. K. Goel, High-speed VLSI Interconnections, Willey-IEEE Press, Sep. 2007. [2] R. Satio, G. Dresselhaus, and S. Desselhaus, Physical Properties of Carbon Nanotubes. London, U. K.: Imperial College Press, 1998. [3] H. Li, C. Xu, N. Srivastava, and K. Banerjee, Carbon Nanomaterials for Next-Generation Interconnects and Passives: Physics, Status and Prospects, IEEE Trans. Electron Devices, vol. 56, no. 9, pp. 1799-1821, Sep. 2009. [4] A. Javey and J. Kong, Carbon Nanotube Electronics, Springer, 2009. [5] B. Q. Wei, R. Vajtai, and P. M. Ajayan, Reliability and current carrying capacity of carbon nanotubes, Appl. Phys. Lett., vol. 79, no. 8, pp. 1172-1174, 2001. [6] P. G. Collins, M. Hersam, M. Arnold, R. Martel, and P. Avouris, Current saturation and electrical breakdown in multiwalled carbon nanotubes, Phys. Rev. Lett., vol. 86, no. 14, pp. 31283131, 2001. [7] S. Berber, Y.-K. Kwon, and D. Tomanek, Unusually high thermal conductivity of carbon nanotubes, Phys. Rev. Lett., vol. 84, no. 20, pp. 46134616, 2000. [8] P. Avorious, Z. Chen, and V. Perebeions, Carbon-based electronics, Nat. Nanotechnol., vol. 2, no. 10, pp. 605-613, Oct. 2007. [9] K. Tsukagoshi, B. W. Alphenaar, and H. Ago, Coherent transport of electron spin in a ferromagnetically contacted carbon nanotube, Nature, vol. 401, no. 6753, pp. 572-574, Oct. 1999. [10] J. A. Misewich, R. Martel, P. Avouris, J. C. Tsang, S. Heinze, and J. Tersoff, Electrically induced optical emission from a carbon nanotube FET, Science, vol. 300, no. 5620, pp. 783786, May 2003. [11] N. Wang, Z. K. Tang, G. D. Li, and J. S. Chen, Materials science: Single-walled 4 carbon nanotube arrays, Nature, vol. 408, no. 6808, pp. 5051, Nov. 2000. [12] M. F. Yu, O. Lourie, M. J. Dyer, K. Moloni, T. F. Kelly, and R. S. Ruoff, Strength and breaking mechanism of multiwalled carbon nanotubes under tensile load, Science, vol. 287, no. 5453, pp. 637640, Jan. 2000. [13] F. Lu, L. Gu, M. J. Meziani, X. Wang, P. G. Luo, L. M. Veca, L. Cao, and Y. P. Sun, Advances in bioapplications of carbon nanotubes, Adv. Mater., vol. 21, no. 2, pp. 139 152, 2009. [14] C. Du, J. Yeh, and N. Pan, High power density supercapacitors using locally aligned carbon nanotube electrodes, Nanotechnology, vol. 16, no. 4, pp. 350353, Feb. 2005. [15] P. Wei, W. Bao, Y. Pu, C. N. Lau, and J.Shi, Anomalous thermoelectric transport of Dirac particles in graphene, Phys. Rev. Lett., vol. 102, no. 16, p. 166 808, Apr. 2009. [16] H. Ago, K. Petritsch, M. S. P. Shaffer, A. H. Windle, and R. H. Friend, Composites of carbon nanotubes and conjugated polymers for photovoltaic devices, Adv. Matter., vol. 11, no. 15, pp.12811285, 1999. [17] W. B. Choi, D. S. Chung, J. H. Kang, H. Y. Kim, Y. W. Jin, I. T. Han, Y. H. Lee, J. E. Jung, N. S. Lee, G. S. Park, and J. M. Kim, Fully sealed, high-brightness carbon-nanotube field-emission display, Appl. Phys. Lett., vol. 75, no. 20, pp. 31293131, Nov. 1999. [18] P. G. Collins, M. S. Arnold, and P. Avouris, Engineering carbon nanotubes and nanotube circuits using electrical breakdown, Science, vol. 292, no. 5517, pp. 706709, Apr. 2001. [19] H. Dadgour, A. M. Cassell, and K. Banerjee, Scaling and variability analysis of CNT-based NEMS devices and circuits with implications for process design, in IEDM Tech. Dig., 2008, pp. 529532. [20] Hyperion Catalysis. [Online]. Available: http://www.fibrils.com. [21] F. Kreupl, A. P. Graham, G. S. Duesberg, W. Steinhogl, M. Liebau, E. Unger, and W. Honlein, Carbon nanotubes in interconnect applications, Microelectronics Eng., vol. 64, no. 14, pp. 399408, Oct. 2002. [22] J. Li, Q. Ye, A. Cassell, H. T. Ng, R. Stevens, J. Han, and M. Meyyappan,Bottom-up approach for carbon nanotube interconnects, Appl. Phys.Lett., vol. 82, no. 15, pp. 24912493, Apr. 2003. [23] H. Li and K. Banerjee, High-frequency effects in carbon nanotube interconnects and implications for on-chip inductor design, in IEDM Tech. Dig., pp. 525528, 2008. [24] P. L. McEuen, M. S. Fuhrer, and H. Park, Single-Walled Carbon Nanotube Electronics, IEEE Trans. Nanotechnology, vol. 1, no. 1, pp. 78-85, Mar. 2002. [25] N. Srivastava and K. Banerjee, Performance Analysis of Carbon Nanotube Interconnects for VLSI Applications, in Proc. IEEE/ACM International conference on Computer-aided design, 2005. [26] Y. Xu and A. Srivastava, A Model of Multi-Walled Carbon Nanotube Interconnects, in Proc. 52nd IEEE International Midwest Symposium on Circuits and Systems, 2009. [27] A. Srivastava, Y. Xu, and A. K. Sharma, Carbon nanotubes for next generation very large scale integration interconnects, Journal of Nanophotonics, vol. 4, no. 041690, pp. 1-26, May 2010. [28] Y. Xu and A. Srivastava, A model for carbon nanotube interconnects, Int. J. Circ. Theor. Appl., vol. 38, pp. 559-575, Mar. 2009. [29] P. J. Burke, Lttinger Liquid Theory as a Model of the Gigahertz Electrical Properties of Carbon Nanotubes, IEEE Trans. Nanotechnology, vol. 1, no. 3, pp. 129-144, Sep. 2002. [30] G. Miano and F. Villone, An integral formulation for the electrodynamics of metallic carbon nanotubes based on a uid model, IEEE Trans Antennas and Prop., vol. 54, no. 10, pp. 2713-2724, 2006. [31] A. Nieuwoudt and Y. Massoud, Understanding the Impact of Inductance in Carbon Nanotube Bundles for VLSI Interconnect Using Scalable Modeling Techniques. IEEE Trans. Nanotechnology, vol. 5, no. 6, Nov. 2006. [32] Y. Xu, A. Srivastava, and A. K. Sharma, Emerging Carbon Nanotube Electronic Circuits, VLSI Design, vol. 2010, pp. 1-8, 2010. [33] N. Peng, Q. Zhang, J. Li, and N. Liu, "Influences of ac electric field on the spatial distribution of carbon nanotubes formed between electrodes," J. Appl. Phys., vol. 100, no. 2, 2006. [34] M. Zheng, A. Jagota, M. S. Strano, A. P. Santos, P. Barone, S. G. Chou, B. A. Diner, M. S. Dresselhaus, R.S. McLean, G.B. Onoa, G. G. Samsonidze, E. D. Semke, M. Usrey, and D. J. Walls, "Structure-based carbon nanotube sorting by sequence dependent DNA assembly," Science, vol. 302, no. 5650, pp. 15451548, Nov. 2003. 109