Current Status and Future Perspectives of Carbon Nanotube Interconnects Kaustav Banerjee, Hong Li, and Navin Srivastava Department of Electrical and Computer Engineering, University of California, Santa Barbara, CA 93106, USA E-mail: {kaustav, hongli, navins}@ece.ucsb.edu Abstract Abstract- In this paper, we review the current status of CNT interconnect research, from both fabrication and modeling aspects. The fabrication issues of vertical and horizontal CNT interconnects and remaining challenges are discussed. State-of-the-art in both SWCNT and MWCNT modeling and performance analysis are presented. In addition, high-frequency effects, off-chip application, and process variation of CNT interconnects have also been discussed. Keywords: carbon nanotube, interconnect, performance analysis, high-frequency, process variation. 1. Introduction The resistivity of Cu interconnects in current and imminent technologies is increasing rapidly under the combined effects of grain boundary scattering, surface scattering and the presence of a highly resistive diffusion barrier layer [1], [2]. The steep rise in parasitic resistance of Cu interconnects not only increases interconnect delay (degrades electrical performance) but also leads to significant reliability issues [3] due to decreasing thermal conductivity of low-k dielectrics [2] and increasing current density demands from interconnects [4]. Due to their long mean free paths (MFP), high current carrying capability and high thermal conductivity, Carbon Nanotubes (CNTs) are expected to be a very good alternative material for future nanoscale interconnects, which can enhance the electrical performance as well as eliminate electromigration reliability concerns that plague nanoscale Cu interconnects. CNTs are sheets of graphene rolled up as hollow cylinders. CNTs can be classified as Single-Walled (SWCNTs, with only one shell and diameter ranging from 0.4 nm to 4 nm) and Multi-Walled (MWCNTs, with several concentric shells and diameter ranging from several nm to tens of nm). While SWCNTs can be either metallic or semiconducting depending on their chirality (the direction in which they get rolled up); giving rise to zigzag (mostly semiconducting), armchair (metallic) or chiral nanotubes (mostly semiconducting), MWCNTs are always metallic. A comparison of the properties of Cu, single-walled (SWCNT) and multi-walled CNTs (MWCNT) is shown in Table I. Specifically, Fig. 1 shows the resistivity comparison among Cu wire, SWCNTs and MWCNTs for different lengths. 2. CNT Interconnect Fabrication and Integration For interconnect applications, chemical vapor deposition (CVD) methods are most suited since they have the capability of selective growth, large area deposition, and aligned CNT growth. Moreover, the high resistance associated with an isolated CNT (greater than 6.45 KΩ) [8] necessitates the use of a bundle of CNTs conducting current in parallel to form an interconnection [10]-[13]. The current state-of-the-art of on-chip integration of CNTs as interconnects focuses on vertical interconnects (vias) [10]-[13], while growing long length horizontal CNT interconnects remains challenging. The demonstration of CNT bundle growth in a horizontal direction uses the fact that CNT bundles always tend to grow perpendicular to a surface [14]. The orientation of nanotubes is directly controlled by the direction of gas flow in the CVD system [15]. The substrate needs to be rotated to get different orientation of the CNTs. Other approaches like electric field induced alignment [16] or fluidic methods are not suited for large scale integration [17]. Until recently it has been difficult to grow dense bundles of SWCNTs because the fertility of the catalyst particles for SWCNT growth was low. Although 84% catalyst activity has been reported in [18], the grown SWCNT bundle is very sparse, only occupying 3.6% of the total volume. Moreover, the lack of control on chirality means that it is difficult to ensure that the SWCNTs forming a bundle are all metallic. Although 87% metallic SWCNT bundle has been reported [19], it is not suitable for large scale integration and the density of bundle after separation is quite low. Since it is very challenging to obtain densely packed bundles of metallic SWCNTS, all of the CNT bundle fabrication work has been focused on MWCNTs which can guarantee metallic behavior. Two Fig. 2. (a) Process sequence for CNT vias. (b) The first CNT via by Kreupl et al., [11]. (c) Vertical and horizontal CNT bundles from Fujitsu [12]. TABLE I. Comparison of properties among Cu, SWCNT, and MWCNT. Cu SWNCT MWCNT Max. current density (A/cm 2 ) <1x10 7 >1x10 9 [5] Thermal conductivity (W/mK) 385 5800 [6] 3000 [7] Mean free path (nm) @ 300K 40 >1000 [8] >25000 [9]* * MFP of MWCNTs depends on their diameters. The value shown here is for the MWCNT with outmost shell diameter of 100 nm. 1 10 100 1000 1 10 100 Cu: W=32nm, ρ =4.83 Cu: W=22nm, ρ =6.01 SWCNT(D=1nm), All Metallic SWCNT(D=1nm), 1/3 Metallic (Random Chirality ) Cu: W=14nm, ρ =8.19 MWCNT,D=32nm MWCNT,D=22nm MWCNT,D=14nm Resistivity [ μΩ -cm] Length [ μm] Fig. 1. Comparison of resistivity among MWCNTs with various diameters, Cu wires with different dimensions, and SWCNT bundles with different chiralities. Dimension of Cu wires are adopted from ITRS. SWCNT bundles are assumed to be densely packed (interval is 0.34 nm as shown alongside). Invited Paper