A. WO₃ etch and Co site density

To grow an array of MWNTs in AAO with maximum site density, one must start with near 100% Co site density. The key to achieving this is the W underlayer. When Al is anodized it results in a test tube-like structure with a barrier oxide thickness similar to the wall thickness at the bottom of the pores above the underlying substrate or unanodized metallic aluminum, as marked with arrows in Fig. 1(a). To successfully electrodeposit catalysts at the pore bottoms, this barrier oxide should be as thin and uniform as possible. For MWNT array applications, this is normally achieved via barrier thinning by decreasing the applied anodization voltage ^{Reference Sklar, Paramguru, Misra and LaCombe11} and a pore-widening step, if desired. If an underlayer is used that oxidizes under the same conditions as the Al film and has a higher ionic conductivity than alumina, as W does, the underlayer's oxide will grow into the pore and form an oxide plug. ^{Reference Oh and Thompson13,Reference Limmer, Yelton, Siegal, Lensch-Falk, Pillars and Medlin15} This is shown in Fig. 1(b), where the arrow points to one such WO₃ plug. By selectively etching the WO₃ plug, the underlying metallic W is exposed at the pore bottoms, as shown in Fig. 1(c). This metallic underlayer facilitates electrodeposition of the Co catalyst and provides a backside electrical contact, even on an insulating substrate.

FIG. 1.

(a) SEM image of the test tube-shaped pore structure and barrier oxide at the pore bottoms after anodizing an Al film without a W underlayer. The barrier oxide is marked by arrows. (b)–(c) SEM images of the tungsten/pore bottom interface before and after etching of the WO₃ plug, respectively. The arrows point to an example WO₃ plug and a resulting void after removal. All scale bars are 200 nm.

To investigate the Co site density resulting from the W underlayer technique, ∼750 nm Co was electrodeposited into the pores followed by ion milling of the AAO surface in a Precision Ion Polishing System (Gatan) at an ion beam angle of 2°. A Co height of 750 nm was determined to yield reliable MWNT growth, which is discussed further in Sec. III. B. The ion milling, which etches a wide-angle cone as the sample rotates, continues until the AAO in the center of the cone is completely removed. In this configuration, SEM easily distinguishes between the filled and empty pores along the etched cone since the surface of the AAO and the Co nanowires are now flush with each other. The Co-containing pores appear bright relative to the empty pores. Furthermore, we can gain some insight into the height uniformity of the Co nanowires by comparing the fraction of full and empty appearing pores, as we image radially out from the center of the etched cone.

For a given distance from the etched cone center, the apparent pore filling ratio is constant. SEM images in this regime for WO₃ etch times of 10, 60, and 120 min are shown in Figs. 2(a)–2(c). Etch times longer than 120 min and up to 15 h yield neither further improvement nor template failure upon Co electrodeposition. The 10-min etch time leaves both empty, isolated pores surrounded by full pores as well as clusters of empty pores. Increasing the etch time to 60 min decreases the number of clusters and the total number of empty pores. After 120 min, all empty pore clusters are eliminated, with only a few isolated empty pores remaining. Figure 2(d) shows the Co site density as calculated from representative SEM images for each of the three WO₃ etch times. With this technique, the upper limit of Co site density is ∼98%. Controlling the WO₃ etch time, therefore, provides a new technique to tune the site density of the Co catalyst and thus that of the MWNTs. This approach is also relevant for controlling the fabrication of nanowire arrays of other materials using AAO templates.

FIG. 2.

(a)–(c) Representative SEM images of Co site density in ion-milled pores for 10-, 60-, and 120-min WO₃ etch times, respectively. Scale: pore diameter is 75 nm. (d) Co site density versus WO₃ etch time. The line is a guide to the eye.

B. Carbon nanotube growth and characterization

When C₂H₂ is introduced to the CVD furnace at temperature, there are two competing processes: (i) CVD growth of carbon nanotubes from the Co catalyst and (ii) amorphous carbon deposition via C₂H₂ decomposition on the AAO surfaces. ^{Reference Chen, Chang, Kuo and Pan19,Reference Lee, Gu, Kim, Jeong, Bae and Suh20} Catalytic CVD growth of MWNTs from Co under experimentally similar conditions is known to occur by tip, as opposed to base, growth, evidenced by transmission electron microscopy imaging of Co particles at the ends of MWNTs, which emerge from the pores. ^{Reference Chen, Chang, Kuo and Pan19} However, as the amorphous carbon deposits on the pore walls, it can prevent upward lifting of the Co catalyst, halting MWNT growth and leaving short MWNTs inside the pores. MWNTs that do emerge from the pore openings are free to continue growing longer with time. Adding H₂ to the gas mixture during nanotube growth has been systematically shown to etch away amorphous carbon and help MWNTs emerge from the pores. ^{Reference Chen, Chang, Kuo and Pan19} In addition to blocking the Co particle uplift, if the concentration of the carbonaceous feed gas is too high, the Co catalyst can be deactivated by amorphous carbon deposition. This results in shorter MWNTs above the AAO surface at which point amorphous carbon growth on the AAO is no longer interfering with the MWNT growth. Decreasing the aspect ratio of the open pore above the Co catalyst is another technique to increase the fraction of pores producing MWNTs. Varying the aspect ratio by more than a factor of 4 was shown to vary the emerging MWNT density by 4 orders of magnitude, underscoring the sensitivity of this competition. ^{Reference Jeong, Lee, Lee, Oh and Park10}

All of these mechanisms are relevant to this work. The presence of amorphous carbon on the AAO surface is confirmed by Raman spectroscopy (not shown) of an AAO template that underwent an identical CVD process without the Co catalyst. Additionally, a maximum aspect ratio ∼10 was established as necessary for at least 1% emerging MWNT site density as determined by SEM imaging. Explicitly stated, for the pore diameter of 75 nm used in this work, the depth of the open pore above the Co needs to be less than 750 nm for MWNTs to emerge from the pores. Because the Al film expands by nearly 50% upon anodization, the deposited 1-µm thick Al film produces a 1.5-μm thick AAO template and requires electrodeposition of 750 nm of Co. Using a thinner template layer decreases the pore aspect ratio linearly for a given Co deposition height, but it is observed that a minimum AAO thickness is necessary for structural integrity of the AAO array following dense MWNT growth. Templates thinner than 1 µm tend to have regions of delamination after MWNT growth due to stress on the template from the presence of MWNTs in the pores. Using a thicker Nd-doped Al layer together with longer electrodeposited Co catalyst nanowires mitigates this problem. However, with longer electrodeposition times, it becomes increasingly important for the pore bottoms to have uniform electrical contacts with the underlying W metal to ensure Co height uniformity. Catalyst overgrowth of even a small percentage of the pores results in uncontrolled carbonaceous growth that is detrimental for the practical use of MWNT/AAO arrays. The implementation of the W layer and attention to WO₃ etch times described above provide the necessary uniformity.

SEM images of the as-grown MWNT arrays are shown in Fig. 3. Once the MWNTs emerge from the pore openings, they lose their lateral confinement and tend to take random gently curved paths resulting in a loosely intertwined network of MWNTs. However, the growth is generally free from kinks and bends, which are evidence of severe microstructural defects in the nanotubes. Figure 3(b) provides a magnified inspection near the AAO surface and reveals that MWNTs emerge from the pores straight and with uniform diameter; however, not every pore yields a nanotube. With further optimization of the gas composition and pore aspect ratio, higher MWNT site density should be achievable. Note that the MWNT diameters appear to be slightly larger than the pore opening, in agreement with previous reports. ^{Reference Jeong, Lee, Lee, Oh and Park6,Reference Suh and Lee8}

FIG. 3.

SEM images of (a) the MWNT array overgrowth above the AAO surface and (b) the MWNTs emerging from the AAO pore openings.

To gain insight into the crystalline quality of MWNTs resulting from this template growth process, Raman spectra are acquired on (i) the overgrown array and (ii) an individual tube dispersed on a silicon substrate after being sheared from the template during sonication. To obtain a representative sample, 200 separate spectral acquisitions of the overgrown array are gathered over a length scale of 100 μm. Since the AAO has a nominal pore size of 75 nm and the diameter of the Raman spot is ∼350 nm, the average spectrum shown in Fig. 4 resulting from these 200 acquisitions is the composite signal resulting from a sample of ∼1000 separate MWNTs. No appreciable change in the Raman signal was observed between any of the 200 acquisitions. In a similar fashion, investigation of an individual MWNT occurred via spectral acquisition every 30 nm over a 2 × 3-μm area. Little variation is observed over the 500 Raman spectra acquired across the nanotube. In addition, little variation occurred from tube to tube. Spectra extremely similar to that shown in Fig. 4 were acquired on 10 separately imaged individual MWNTs. Therefore, the average response shown in Fig. 4 is representative of the individual MWNT as a whole.

FIG. 4.

Raman spectra of the MWNT array overgrowth above the AAO and an individual MWNT cut from the same array overgrowth by ultrasonication. Insets show SEM images of the material sampled for each spectrum.

The Raman response of both the array and individual tube are marked by four distinct spectral features characteristic of MWNTs, namely the first order D (∼1350 cm⁻¹) and G modes (∼1582 cm⁻¹) along with the second order 2D (∼2677 cm⁻¹) and D+G (∼2930 cm⁻¹) modes. ^{Reference Dresselhaus and Eklund21} While observation of the D and D+G modes is indicative of crystalline disorder, the persistence of the 2D-mode accompanied by an I(D)/I(G) ratio of 1.9 implies that this disorder is moderate and in line with previous reports of non-graphitized CVD-synthesized MWNT arrays grown without the use of templates. ^{Reference Kang, Yoon, Kim, Hong, Han, Cha, Song, Jung, Lee and Kim22–Reference Chen, Chen and Chen25} Due to this similarity with reported non-templated growth, we conclude that use of the AAO template is to first order benign to the synthesis of the MWNTs. Furthermore, since the Raman spectrum of an individual tube is very similar to that acquired over an array of tubes, we deduce that the template approach results in nominally uniform MWNTs across the array. Note that it should be possible to improve the MWNT crystalline quality with a post-growth graphitization annealing step, ^{Reference Mattia, Rossi, Kim, Korneva, Bau and Gogotsi26,Reference Huang, Wang, Luo and Wei27} which is the subject of ongoing work.

C. Nanotube array planarization and characterization

Figure 5 shows SEM images of the MWNT array in AAO after ultrasonication in acetone for 1 min, after which nearly every nanotube is cut to a few hundred nanometers above the AAO surface. For illustration purposes, a region with higher than average MWNT site density is shown. Because of the sensitivity of the MWNT density to the factors outlined in Sec. III. B, some variation may be observed across a given sample. The MWNTs do not fully support each other laterally, since their site density has yet to be optimized, and tend to clump together or lean over. In future work, this can be improved by drying the array in supercritical CO₂. ^{Reference Liang, Zhen, Zou and Xu28,Reference Limmer, Yelton, Erickson, Medlin and Siegal29} Figure 5(b) shows a higher magnification image of the ends of the cut MWNTs. The insert shows an image of the MWNT ends before cutting. By comparison, it is clear that the growth produces carbon nanotubes and that ultrasonication results in open-ended tubes.

FIG. 5.

(a) Low- and (b) high-magnification SEM images of a MWNT array after ultrasonic cutting. The inset in (b) shows the ends of MWNTs before ultrasonic cutting.

However, an open question remained as to whether or not ultrasonication degrades or destroys the electrical contact between the MWNT, the Co catalyst, and the W-coated substrate or even causes breaks in the MWNTs themselves inside the template. To answer this question, we used c-AFM to study the current path between the exposed MWNT tip and the W underlayer. For this measurement, contact to the W underlayer is made through the Ti strip that is located on one edge of the sample surface as mentioned above. The Ti strip makes contact with the W underlayer through the non-anodized Nd-doped Al that is under the strip. The W underlayer extends across the entire substrate and makes contact with every Co catalyst nanowire. Topological and current maps are acquired simultaneously in non-contact/intermittent mode using a conductive tip with a nominal spring constant of 18 N m⁻¹. Non-contact mode is necessary due to mechanical damage and nanotube cleaving produced by physical contact between the tip and the MWNT. The tip will, however, have momentary contact with protruding MWNTs on the side of approach before the feedback is able to move the tip over the MWNT.

Figure 6(a) shows a current map generated by c-AFM. The current recorded varied between 0–250 nA with the highest current coming from a small number of hot spots. The current in the figure is cropped at 32 nA to increase the contrast and improve the clarity of the image. During current mapping, a static bias voltage is applied to the tip, and the current through the sample is recorded as a function of tip position. The current is only detected when the tip is in intermittent contact with a MWNT. As expected, no current is seen from the AAO. Intermittent contact occurs when the non-contact topological feedback begins to move the tip over the MWNT. Since the non-contact feedback is not perfect, there are brief moments when the tip touches the MWNT. This is verified by merging both the topological and current maps (not shown) to reveal that the current flow occurred at the approaching edge of the MWNT.

FIG. 6.

(a) Conductive AFM map of current between the top of the AAO/MWNT array and the underlying W layer. (b) I–V data from three individual MWNTs in the array showing Ohmic conduction through the MWNT, the MWNT-Co interface and the Co-W interface after ultrasonic cutting.

Figure 6(b) shows I(V) data collected on three individual MWNTs in contact mode with a loading force of ∼1.6 μN using the same tip as the non-contact imaging experiments. The electrical resistance varies from ∼5 to 25 kΩ, depending on the MWNT being measured. Recall from earlier in this section that this resistance includes the series resistance contributions of the unanodized Nd-doped Al under the Ti contact strip, W underlayer, and Co catalyst nanowire, as well as the contact between the metal layers, the contact between the Co and MWNT, and the contact between the MWNT and tip. In every case, Ohmic I(V) behavior is observed, implying that the W-Co and Co-MWNT interfaces have not been seriously compromised. This measurement does not explain the origin of the range of total resistance measured, but the contact resistance and contact geometries between individual MWNTs and the c-AFM tip are potential sources of variability. It is also not known how many shells contribute to the electronic conduction for each measurement and hence how many parallel current paths through the MWNT are in use. The MWNTs are open at the c-AFM tip end, suggesting that the tip could be in contact with several shells at once. However, if the tip is only in contact with the outermost shell, for instance from a MWNT edge, only that shell would contribute significantly to the conductivity since adjacent shells have negligible interlayer electronic coupling. ^{Reference Collins, Arnold and Avouris30}

Ohmic contact between the MWNTs and the underlying metal is a significant finding. One of the benefits of the W underlayer technique discussed above is the ability to have electrical contact to the entire MWNT array regardless of the chosen substrate's electrical properties. This offers flexibility in device design not possible when MWNT/AAO arrays are fabricated from Al sheets. Confirming the retention of Ohmic contact to the MWNTs after ultrasonic cutting allows for a simple, scalable technique to cut all MWNTs in the array to a uniform height. Good electrical contact also suggests the possibility of good thermal contact. Thermal conductivity of the MWNT arrays and the thermal interfaces between the MWNT and underlying substrate are not studied here but are the subject of an ongoing effort.