Data Analytics Enables Significant Improvement of Robustness in Chemical Vapor Deposition of Carbon Nanotubes Based on Vacuum Baking

Abstract

The root causes underlying run-to-run variations in the synthesis of vertically aligned carbon nanotubes (VACNTs) by chemical vapor deposition can be attributed to the sensitivity of the process to small uncontrolled/unmeasured quantities of gaseous species in the reactor. Hence, universally applicable processing steps are needed to ensure consistency. Here, we quantitatively test the effectiveness of various processing heuristics in reducing growth variability. Statistical analysis of 95 VACNT samples grown by 11 different recipes demonstrated that pumping with mild baking at 200 °C prior to catalyst formation resulted in a significantly reduced coefficient of variation of forest heights (by a factor of 6). In contrast, other processing steps such as vacuum pumping without heating and adjusting the He to H 2 ratio during catalyst formation did not significantly affect growth variability; we could not reject the null hypothesis at any reasonable level of significance. Atomic force microscopy analyses suggest that variability in VACNT height is not caused by variations of nanoparticle size distribution; thus, we can conjecture that the variability might be caused by variations in the chemical state of catalyst nanoparticles.

Reducing Variability in Chemical Vapor Deposition of Carbon Nanotubes Based on Gas Purification and Sample Support Redesign

“…Cold-open refers to the cases in which enough wait time for cooling was allowed after baking to reach temperatures in the range of 30-80 C before opening the reactor to load the substrate, while hot-open refers to the cases in which the temperature was in the range of 130-200 C when opening the reactor. These slight differences in processing parameters should not have a considerable effect on the variability of forest growth based on the previous study of our group [20]. Figure 3 shows that vacuum baking prior to IR heating improves the variability in the height of the forests.…”

“…It also shows that changing He cylinder could affect the height of the forests. This is the reason why we focused on CV when comparing different populations of CNT forests grown according to different conditions, in our previous work [20]. Another limitation of that previous study is that it focused on height measurements only and did not study the variability in the density of the forests.…”

“…It is worth noting that in a decoupled recipe, the growth temperature (T g ) and the catalyst formation temperature (T c ) can be equal, but there is a cooling and reheating step between catalyst formation and growth steps. Therefore, decoupled recipes are generally more sensitive to small variation in other growth conditions, possibly due to longer exposure of catalyst particles prior to growth [20]. Therefore, coupled or semicoupled recipes, wherein temperatures go directly from T c to T g without cooling down first, are more consistent for fundamental studies of growth.…”

“…We used two statistical tests to compute test statistics and p-values: asymptotic test by Feltz and Miller [26] as well as modified signed-likelihood ratio test (M-SLRT) by Krishnamoorthy and Lee [27]. More details about the statistical hypotheses can be found elsewhere [20]. Test statistics and p-values listed in Table 5 reveal that adding He purifier along with using the new support wafer (Fig.…”

“…Previous work from our group showed that pumping with mild baking at 200 C prior to catalyst formation resulted in a significantly improved process consistency by reducing the coefficient of variation of forest heights (by a factor of 6) [20]. The use of coefficient of variation as a statistical measure of consistency allowed comparing different means of populations and proved to be an effective measure under different procedures used in the study.…”