AFM for Membrane Analysis

Dozens of poly(carbonate) membrane images have been analyzed at all four oil-contaminate levels over a variety of random locations and at multiple "magnifications." Only the most representative and enlightening ones are presented along with surface roughness statistics. Oscillating mode was employed for all image generation. Since the polymers were cast on float glass slides, the top surface exposed to air was expected to reveal marked differences compared to the bottom surface cast against the glass. Membrane films also tend to curl away from the bottom side when drying. This identification method is by no means fail-safe and results in the necessity of careful sample labeling, especially for clean polymer samples.

Several different length scales were examined for both sides of the membranes. The underside of all membranes was nonporous and relatively smooth, modeling the float glass surface. Because of this similarity, all images of membrane undersides were lumped together for analysis. No statistical differences of the underside were found for varied oil contamination or over image ranges from 30 mm to 50 mm on a side; the surface deviations are constant over this range. The 95% confidence interval for the overall average RMS roughness of the undersides was 23.4 + 3.8 nm after linear flattening of 36 images.

It must be pointed out that one or two images in each batch were discarded as outliers. These were eliminated by falling outside of the raw mean by more than five standard deviations; all of these images had RMS roughness of the order of 100 nm and had clearly visible defects that were not representative of the polymer cast against smooth glass. If the area of interest for taking roughness measurements were on the millimeter scale, these features would not be justification for outlier elimination. Here, we are focusing on the microstructure only, not the effects of dirt contamination, crazes, and cracks.

It was initially thought that the RMS roughness of the bottom side would tend to be lower than the top, providing a quantitative basis for assisting in the differentiation of the top and bottom sides. While it is obvious from the images and visual inspection that oil contamination increased surface roughness, the variability in RMS was not consistent. The 95% confidence intervals for membrane topside roughness after linear flattening are 73+20 nm (clean membranes - 42 images), 33+8 nm (103 ppm oil - 16 images), 40+22 nm (104 ppm - 8 images), and 75+17 nm (105 ppm - 26 images). Values of RMS roughness based on ~10_10 mm2 images are 3.4+1.8 nm (clean - 18 images), 16.2+6.3 nm (103 ppm - 20 images), 7.4+0.9 nm (104 ppm - 18 images), & 117+40 nm (105 ppm - 20 images). This emphasizes the importance of image size relative to feature size of interest in doing AFM surface roughness analysis.

These collective data indicate statistical differences in the membrane batches. However, it is unclear why the supposedly clean membrane was essentially as rough as the most highly oil-contaminated membrane. For large AFM images, the trend of increasing roughness for 103, 104, & 105 ppm is real but not sufficient for a blind study. This is supported by the lack of consistent trend in the smaller images. Most importantly, the apparent discrepancies imply that there are other factors of importance. Ultimately, roughness values alone are not a sufficient means of sample classification for this experiment and must be approached with caution in any study.

The individual images must be examined for further insight into structural property effects from oil, and this is where the more interesting data are found. During the membrane drying stage, solvent escapes into the air and roughens the topside. This also aids the migration of oil toward the free surface. On the macroscopic scale, the higher contamination levels (e.g., 105 ppm) appear cloudy and dull to the unaided eye, while the bottom side is glossy. Scans of the 105 ppm oil-contaminated membranes revealed numerous pits of 0.5 to 1.5 µm (Fig. 1) responsible for the dull appearance, due to light scattering.

The 104 ppm samples also displayed the pitted effect but to a lesser degree, with pits approximately 100 nm in diameter and much more numerous (Fig. 2).

The pits were notably smaller than at the 105 ppm level, possibly indicating differences in oil solubility in polymer solution or the decreasing probability of oil droplet coalescence at lower volume fractions. Figures 3 and 4 show cross-sectional views for the 105 and 104 ppm oil concentrations, respectively, and provide an extended comparison of crater dimensions.

The holes at 105 ppm appear to be an order of magnitude wider and deeper than those at 104 ppm. However, the nature of AFM skews the depth observation. Since the AFM probe tip is conical-sometimes pyramidal-in shape, it cannot access very narrow or deep features. Therefore, we can only accurately measure diameter at the surface and glean minimum crater depths from the images.

Craters were also discovered in the 103 ppm doping (Fig. 5) with much less frequency and of similar size to those at 104 ppm.

Figure 5 begins to reveal the source of what might be considered the true polymer roughness at the nanometer scale. The larger, high (white) features reveal the level of non-oil contaminant (i.e., dirt). Small bumps only 1-5 nm high are resolved at this scale in the relatively clean regions between craters. This is coming near the noise threshold of these scanning conditions. The clean membrane consistently showed a lack of significant features for various resolutions, from square images of 4 to 80 µm over several sample locations.