Growth of carbon nanotubes

The growth of CNTs was conducted in a customized quartz tube reactor placed in a horizontal tubular furnace. After the formation of catalyst islands described in Section 2.1, wafers were heated in pure H₂ for 5 ~ 15 minutes at 550°C to convert Co²⁺ to metal particles. Then the process temperature was raised to 850°C in Ar at a flow rate of 5000 sccm, and subsequently ethanol was introduced into APCVD furnace as the carbon source with Ar as the carrier gas at a flow rate of 3500 ~ 5000 sccm for 40 minutes in ambient atmosphere. After completion of the reaction, the power was turned off and the reactor was cooled down under flowing Ar. Finally, the suspended CNTs were grown between two neighboring catalyst islands by APCVD with ethanol/Ar (Fig. 5.2(b)).

After blanket-coating a photoresist layer, and using a photolithographic step with a negative source/drain (S/D) pattern mask to etch away the photoresist from the S/D area, a titanium (Ti) layer was deposited by radio frequency (RF) sputtering at room temperature. Next, an ultrasonic instrument and acetone were used for the Ti lift off process, leaving the Ti film only on the S/D regions (Fig. 5.2(c)). The spacing between the source and the drain electrodes was designed to be µm2 (i.e., the channel length of the devices).

A commercial HP-4155A was applied for measuring the Id-Vg transfer curves and I_d-V_d of the CNT-FETs. The raw nanotubes were examined by high-resolution transmission electron microscopy (TEM, JEOL JEM 2000EX).

5.3 Experimental Results and Discussion 5.3.1 Transition metal contamination

process to manufacture our catalyst islands and grow the CNTs. However, the catalyst metals, which belong to the transition metals, can potentially contaminate the expensive semiconductor equipments. As such the foundry is usually reluctant to allow such materials to enter their production lines. When the dimensions of integrated circuits become smaller, the thickness of the gate oxide is being reduced to virtually atomic levels. With oxide thicknesses being less than a few tens of angstroms, the metal impurities can have serious effects on the oxide properties. Generally, transition metals, such as Fe, Co, Ni are commonly used as catalysts for carbon nanotube growth. Unfortunately, these transition metals can be unintentionally left on the surface of the chamber at a number of processing steps including wet chemical etching, dry etching, photolithography and chemical vapor deposition. Therefore, trace transition metals analysis has become essential for the development of CNT processes that are compatible with silicon circuit technologies.

In order to clarify the contamination issue, monitors were set up to measure the cobalt ion concentration in the process chambers (including loadlock chambers and robot arms). The key machines monitored included dry etchers and the CVD machine (Table 5.1). In addition, particle levels were also monitored in our I-line stepper, chemical stations, and spin dryers. The monitor results showed that the monitored particle levels were quite normal (within specification). In our study, VG-PQ3, an inductively-coupled plasma-mass spectrometer (ICP-MS), was used to analyze the monitor wafers which were processed by the semiconductor equipments in question.

From Table 5.1, we can conclude that no noticeable increase in cobalt ion concentration is observed.

5.3.2 The diameter of cobalt catalyst

Fig. 5.3(a) shows the TEM picture of Co catalyst distribution in CMT powder.

The diameters of the embedded Co catalysts range from 10 to 20 nm, which is a parameter crucial to confining the width of the CNTs. Fig. 5.3(b) demonstrates the bundled-SWNTs that were synthesized from CMT powders with a diameter of 10 nm.

After the SWNTs were grown, some bridged SWNTs were found between the CMT powders, as shown in Fig. 5.3(c). These results suggest that the CNT bundle can indeed be naturally bridged between two CMT catalyst islands. It is worth noting that the diameters of Co catalysts are strongly related to the formation procedure of CMT described in Section 2.1 such as the stir time and process environment.

5.3.3 The adhesion of CMT layer on different bottom-gate dielectric layers

In this chapter, we investigate the dispersion/adhesion of our CMT on different bottom-gate dielectric layers (i.e., either thermal oxide or nitride). Fig. 5.4 shows that the CMT can disperse/adhere well and can be patterned well (by dry etching), as evidenced by the vernier structure (i.e., the Arabic numerals in the SEM pictures is an indicator for the resolution of photolithography and the finest pattern after dry etching) on thermal oxide layer. Unfortunately, the oxide bottom-gate dielectric layer with its hydroxyl (-OH) bonds is undesirable for many biosensor applications. To avoid the complication of hydroxyl bonds, nitride in lieu of oxide can be adopted as the bottom-gate dielectric layer. However, our data show that the adhesion of CMT layer on nitride is quite poor (Fig. 5.4(b)), so the direct spinning of the CMT solution on nitride film is not practical. In this work, we found that by adding a hexamethyldisilazane (HMDS) layer on the nitride film before spinning the CMT, good CMT catalyst islands can be formed on the nitride layer for biosensor purposes

(Fig. 5.4(c)).

5.3.4 Effects on CMT patterns by using dry or wet etching process

Fig. 5.5 illustrates the effects on the CMT patterns when growing CNTs using either dry or wet etching process. Fig. 5.5(a) shows the cross-sectional SEM picture of a pattern by dry etching. It can be seen that a suspending CNT is formed between the CMT patterns. In addition, sidewall polymers, which are frequently produced in the dry etching process, can be found all over the trench. These sidewall polymers will inhibit the formation of the CNTs. In contrast, no obvious polymer is found in the wet etching process by BOE (5:1) (buffered oxide etch solution, six parts 40% NH4F and one part 49% HF), as shown in Fig. 5.5(b). Nevertheless it is known that the control of the pattern profile and gap is quite poor in the wet etching process because of the isotropic etching nature. Since the CMT layer is vulnerable to severe over-etching because of its porosity, we eventually adopt a standard reactive ion etching (RIE) dry etching process for the formation of the catalyst islands.

In this work, post-etch cleaning was employed to remove the sidewall polymer (i.e., residue) and some cobalt/CMT debris spreads everywhere. In order to quantify the amount of the residue, we applied a voltage to two random positions on the blanket portion of the wafer and measured the current. In general, the current was expected to be very small (i.e., less than pico-ampere) on the blank region theoretically. Fig. 5.5(c) shows the distribution of leakage current level on the blanket portion of the wafer without receiving any post treatment after using dry etcher to form the catalyst islands. Fig. 5.5(d) indicates that the leakage current is in tens of µA range around the wafer center. Moreover, the dice with the leakage current in the µA range account for about 33.3% of the total dice (17 out of a total of 51dice). We

believe the high leakage current is caused by the conducting residues or the incomplete etching of the conducting CMT layer at the center of the wafer because the CMT is much denser there, so the wafer surface becomes conductive and the leakage current reaches the µA range (compared with the insulator whose current level should fall in the pico-ampere range). The high leakage on the wafer surface could result in the malfunction of the transistors. In order to minimize the leakage current, post treatment (i.e., descum process) after our dry etching process is necessary to remove these residues. To effectively remove the residue, the thickness of bottom-gate dielectric layer was increased and the over-etching time of our dry etching was prolonged.

5.3.5 Other factors that affect the length of CNTs

For CNT-FET applications, vertical distribution and lateral growth of CNTs by the proposed CMT method are critical and must therefore be studied thoroughly. It should be noted, however, that the length of the CNT is also important for some device applications. These include bio-sensors and dual/multiple-gated devices where longer CNT bundles are necessary.

In our reiterative experiments, the major parameter that affects the CNT length and the CNT tip-growth rate is found to be the concentration of carbon atoms. This parameter can be controlled by adjusting the reactive gas ratio (ethanol to Ar) [16].

Fig. 5.6(a) shows the results of the CNT length versus the composition of the reaction gas. For a given Ar flow rate of 5000 sccm, the length of CNT increases as the ethanol to Ar ratio increases from 0.25 to 0.5. This is ascribed to more carbon atoms being carried to the surface of the cobalt catalyst. A higher carbon concentration within the cobalt-carbon mixed catalyst will probably increase both the growth rate and the

length of the CNTs. However, when the ethanol/Ar ratio approaches unity, it is found that the CNT length does not increase any further, perhaps due to the accumulation of excess carbon atoms on the catalyst nanoparticles that eventually poison the catalysts.

When the total gas flow rate reduces from 5000 to 3500 sccm, the length of CNT increases to greater than 4.5 µm. Since our APCVD chamber pressure remains unchanged, when the gas flow rate is decreased, the exhaust speed is also reduced simultaneously. Consequently, the carbon atoms can stay much longer around the Co catalyst. Therefore, the solubility of carbon atoms into the Co increases, resulting in the increase of the growth rate and the length of the CNTs. Under this process condition, a CNT is found to bridge the 6µm-gap, as shown in Fig. 5.8(b).

Besides controlling ethanol/Ar ratio of the reactive gas, the density of CNTs can be tuned by the concentration of Co²⁺ ion in the CMT solution. Figs. 6(b) and 6(c) show SEM images of the CNTs grown by different Co²⁺ concentrations of 1.5 M and 0.5 M, respectively. When the Co²⁺ concentration is 1.5 M, the CMT solution is more viscous, and the final thickness of catalyst islands is about 150 nm. The resultant density of the CNTs is higher as shown in Fig. 5.6(b), in which CNTs grow on the catalyst islands’ sidewalls. However, when the Co²⁺ concentration is reduced to 0.5 M, the viscosity of the CMT solution is reduced and the thickness of the catalyst layer reduces to 20 nm. Obviously the density of the CNTs grown by the 0.5 M CMT solution is lower, and all the CNTs are located just in one single layer, as shown in Fig.

5.6(c).

In our experiments, for longer hydrogen (H2) reduction time, the growth mechanism favors the base-growth-type CNT. While for shorter hydrogen reduction time, the growth mechanism favors the tip-growth mechanism (Fig. 5.7), resulting in longer CNTs. So it appears that H₂ reduction time also plays a role on activating the

Co particle as catalyst in our CMT method. This result is not consistent with other group who used Co and carbon monoxide to synthesize CNTs.

The effects of the process gas flow direction in the APCVD quartz tube were also investigated in this chapter, no obvious difference was found when the gas flow is in either X- or Y-direction (Fig. 5.4(a)).

5.3.6 Performance of CNT-FET devices

Out of a total 1173 devices produced on one test wafer, 493 metallic bundled-CNTs (42.03%), 204 semiconducting-type CNT-FETs (17.39%) with an on/off ratio of less than or equal to two orders, and 17 semiconducting-type CNT-FETs (1.45%) with an on/off ratio ranging from two to six orders, were obtained in this study. It is worth noting that in a batch of as-grown CNTs, both metallic and semiconducting type CNTs were present [17]. When at least one metallic CNT is included in the bundle, the bundled-CNTs would show a gradual transition to metallic character. It is plausible that if the number of metallic CNTs is small, the bundled-CNTs will still exhibit weak semiconducting characteristics. On the contrary, if the number of metallic CNTs is large, the bundled-CNTs will show metallic characteristics. In general, when the on/off ratio of our CNT-FETs is about 10 ~ 100, less than 5 ~ 10 metallic CNTs are included in the channel (i.e., the space between two catalyst islands). Only when all CNTs in the bundle are semiconducting can a high performance in the CNT-FETs be expected.

Since our bundled-CNTs were exposed to the air (Fig. 5.2(c)), all the as-grown CNT-FETs manufactured in this study were p-type semiconducting in nature [29,30].

Fig. 5.8(c) shows the electrical properties (i.e., I_d-V_d and I_d-V_g) of an as-grown p-type CNT-FET with five orders of on/off ratio. The bottom-gate voltage applied in Fig.

5.8(c) varies from 0V to -10V (at a step of -2V) at Vds = -1V. It should also be noted that the CNT-FETs with one to two orders of on/off ratio are rendered acceptable to serve the role of sensors. However, in order to manufacture complementary CNT-FET structure (similar to conventional silicon-based CMOS devices) for broader applications, n-type CNT-FETs are indispensable. Previously, our group has successfully manufactured air-stable n-type CNT-FETs [26-28] without resorting to any additional and complex annealing process. After measuring the electrical properties of some wafers that depict generic p-type CNT-FETs in this work, a 300 ~ 400nm-thick silicon nitride film was deposited on the wafer as passivation layer by PECVD at 390^oC. Afterwards, the contact holes of the source/drain and bottom-gate regions were etched in the same MERIE dry etcher. Our experimental data confirmed that the generic p-type CNT-FETs are converted to air-stable n-type CNT-FETs, as shown in Fig. 5.8(d). The converted n-type CNT-FETs depict one to four orders of on/off ratio when the bottom-gate is biased from 0V to 10V (at a step of 2V) at Vds = 1V. This is ascribed to the use of PE-nitride film as the passivation layer whose deposition temperature is high enough to simultaneously remove the oxygen atoms from the CNTs or CNT/metal interface in the PE-CVD deposition chamber. The approach offers a feasible method to fabricate air-stable n-type CNT-FETs by converting the generic p-type CNT-FETs.

It is worth noting that theoretically both n- and p-type CNT-FETs can be fabricated on the same chip by selectively converting some of the generic p-type CNT-FETs on the chip, while leaving other p-type CNT-FETs on the same chip untouched. The design of a new photo mask set is currently under way which will allow us to test the feasibility of the idea.

Although the device performance of our proposed process has yet to be

optimized, our initial results are encouraging and suggest a viable method of fabricating functional CNT-FETs. Using the CNT growth method proposed in this chapter, we believe it is possible, with further process refinement, to manufacture semiconducting-type CNT-FETs suitable for many applications, especially sensors. As mentioned above, the CNTs synthesized in this study contain a high percentage of metallic-type CNTs, representing about 42.03% of metallic bundled-CNTs out of the total as-grown devices. We do believe however that after optimizing the growth conditions and applying plasma treatment on the as-grown CNT-FETs, a much higher percentage and a higher on/off ratio of semiconducting-type CNT-FETs are achievable [31].

5.4 Summary

To the best of our knowledge, only SWNTs can show the semiconducting-type behaviors. And judging from our electrical measurement results and TEM pictures (Fig. 5.3(b)), we deduce that the CNTs we synthesized are mostly bundled-SWNTs.

This result is also consistent with earlier study [32].

In order to synthesize long CNTs, it is critical to balance the rate of ethanol decomposition and the rate of carbon atom diffusion. In other words, the carbon supply route needs to remain open during processing. The rate of carbon diffusion will be dominated mostly by the temperature, and the ethanol decomposition rate will be affected by both the synthesis temperature and the flow rate of the carrier gas (Ar).

In short, we have developed an integrated circuit (IC) compatible process for fabricating CNT-FETs successfully with a shorter hydrogen reduction time in this study. Longer CNT length can be obtained by optimizing carbon ratio during synthesis, and the accumulation of carbon on the catalyst tip can be controlled by

mixing inert gas with the carbon source. This IC compatible process seems to be promising for fabricating a pre-aligned single-walled carbon nanotube matrix for both n- and p-type CNT-FETs.

References

[1] B. Q. Wei, R. Vajtai, and P. M. Ajayan, “Reliability and current carrying capacity of carbon nanotubes,” Appl. Phys. Lett., Vol. 79, Issue: 8, pp.1172-1174, August, 2001.

[2] S. Huang, X. Cai, C. Du, and J. Liu, “Oriented long single walled carbon nanotubes on substrates from floating catalysts,” J. Phys. Chem. B, Vol. 107, No.

48, pp.13251-4, December, 2003.

[3] S. Fan, M. G. Chapline, N. R. Franklin, T. W. Tombler, A. M. Cassell, and H. Dai,

“Self-oriented regular arrays of carbon nanotubes and their field emission properties,” Science, Vol. 283, Issue: 5401, p.512, January, 1999.

[4] P. A. Smith, C. D. Nordquist, T. N. Jackson, T. S. Mayer, B. R. Martin, J. Mbindyo, and T. E. Mallouk, “Electric-field assisted assembly and alignment of metallic nanowires,” Appl. Phys. Lett., Vol. 77, Issue: 9, pp.1399-1401, August, 2000.

[5] K. Yamamoto, S. Akita, and Y. Nakayama, “Orientation and purification of carbon nanotubes using ac electrophoresis,” J. Phys. D: Appl. Phys., Vol. 31, No. 8, pp.L34-L36, April 1998.

[6] J. Liu, M. J. Casavant, M. Cox, D. A. Walters, P. Boul, W. Lu, A. J. Rimberg, K. A.

Smith, D. T. Colbert, and R. E. Smalley, “Controlled deposition of individual single-walled carbon nanotubes on chemically functionalized templates,”

Chemical Physics Letters, Vol. 303, Issue: 1-2, pp.125-129, April, 1999.

[7] Y. Huang, X. Duan, Q. Wei, Lieber, M. Charles, “Directed assembly of one-dimensional nanostructures into functional networks,” Science, Vol. 291, Issue: 5504, p.630, January, 2001.

[8] Y. Huang, X. Duan, Y. Cui, L. J. Lauhon, K.-H. Kim, Lieber, M. Charles, “Logic

294, Issue: 5545, p.1313, November, 2001.

[9] Y. Zhang, A. Chang, J. Cao, Q. Wang, W. Kim, Y. Li, N. Morris, E. Yenilmez, J.

Kong, and H. Dai, “Electric-field-directed growth of aligned single-walled carbon nanotubes,” Appl. Phys. Lett., Vol. 79, Issue: 19, pp.3155-3157, November, 2001.

[10] K. H. Lee, J. M. Cho, and W. Sigmund, “Control of growth orientation for carbon nanotubes,” Appl. Phys. Lett., Vol. 82, Issue: 3, pp.448-450, January, 2003.

[11] B.-H. Chen, P.-Y. Lo, J.-H. Wei, M.-J. Tsai, C.-L. Hwang, T.-S. Chao, H.-C. Lin, and T.-Y. Huang, “Localized lateral growth of single-walled carbon nanotubes for field-effect transistors by a Cobalt-Mix-TEOS method,” Electrochem.

Solid-State Lett., Vol. 8, Issue: 10, pp.G290-293, August, 2005.

[12] W. E. Alvarez, B. Kitiyanan, A. Borgna, and D. E. Resasco, “Synergism of Co and Mo in the catalytic production of single-wall carbon nanotubes by decomposition of CO,” Carbon, Vol. 39, Issue: 4, pp.547-558, June, 2001.

[13] A. Lan, Y. Zhang, X. Zhang, Z. Iqbal, H. Grebel, “Is molybdenum necessary for the growth of single-wall carbon nanotubes from CO? ” Chemical Physics Letters, Vol. 379, Issue: 5-6, pp.395-400, October, 2003.

[14] B. C. Satishkumar, A. Govindaraj, R. Sen, C.N.R. Rao, “Single-walled nanotubes by the pyrolysis of acetylene-organometallic mixtures,” Chemical Physics Letters, Vol. 293, Issue: 1-2, pp.47-52, August, 1998.

[15] B. Kitiyanan, W. E. Alvarez, J.H. Harwell, and D. E. Resasco, “Controlled production of single-wall carbon nanotubes by catalytic decomposition of CO on bimetallic Co–Mo catalysts,” Chemical Physics Letters, Vol. 317, Issue: 3-5, pp.497-503, February , 2000.

[16] X. Z. Liao, A. Serquis, Q. X. Jia, D. E. Peterson, Y. T. Zhu , and H. F. Xu, “Effect of catalyst composition on carbon nanotube growth,” Appl. Phys. Lett., Vol. 82,

Issue: 16, pp.2694-2696, April, 2003.

[17] M. S. Dresselhaus, G. Dresselhaus, Ph. Avouris, Carbon nanotubes: synthesis, structure, properties, and applications: Springer-Verlag, Berlin, 2001.

[18] A. Bachtold, P. Hadley, T. Nakanishi C. Dekker, “Logic circuits with carbon nanotube transistors,” Science, Vol. 294, Issue: 5545, p.1317, November, 2001.

[19] S. J. Wind, J. Appenzeller, R. Martel, V. Derycke, Ph. Avouris, “Vertical scaling of carbon nanotube field-effect transistors using top gate electrodes,” Appl. Phys.

Lett., Vol. 80, Issue: 20, pp.3817-3819, May, 2002.

[20] J. Appenzeller, J. Knoch, V. Derycke, R. Martel, S. Wind, and Ph. Avouris,

“Field-modulated carrier transport in carbon nanotube transistors,” Phys. Rev.

Lett., Vol. 89, Issue: 12-16, p.126801, September, 2002.

[21] A. Javey, Q. Wang, W. Kim, and H. Dai, “Advancements in complementary

[21] A. Javey, Q. Wang, W. Kim, and H. Dai, “Advancements in complementary