Pore Processing Results
I have been using MatLab to analyze the pores in the SC wafers. Here are the results from my analyses:
To take a first look, I simply plotted four characteristics: average pore size, porosity, cut-off pore size, and mean roundness against the wafer number. Notice that the blue dots are the b wafers. Excel would not let me plot wafer numbers like “b8,” so I plotted them as a different series to avoid overlaps.
Looking at the above graphs, there seems to be a connection between the average pore size and porosity and the cut-off pore size. The mean roundness appears almost linear, with a patch of consistently round pores between 100 and 130 or so.
Here is a first attempt to quantify some of the data:
This graph is pretty noisy, but there does seem to be an upward trend. The outliers come from a few wafers that have a myriad of very small pores and wafers that have a sparse number of very large pores.
There seems to be a nice, linear trend here.
Here is the Excel spreadsheet:
Is the outlier at nearly 19% porosity really that porous, or possibly an artifact? Which wafer number is it?
The plot of the average vs. cutoff pore size is very interesting. I had not realized how well-defined that 50nm cap has been, although it makes sense.
The trend of pores becoming slightly more elliptical is something that I have noticed, but it’s great to have it quantified. It’s unclear why this is happening, but it is certainly real.
Thanks for compiling this data – is the Excel file posted somewhere?
Why does the cap at 50 nm make sense? I hope this isn’t an artifact of our histogram plots typically ending out around 50 nm.
I inserted the data into this post as a picture because I was not sure how to upload an Excel document. If there is a better way, please let me know.
It looks like that outlier is SC 143 with a porosity of 18.6%. Look at the TEM image for that wafer. To me, it looks very porous compared to the others. See what you think.
Update: I figured out how to upload the Excel spreadsheet (thanks to Jess), so now the actual spreadsheet is linked to this post.
Carrie –
I took a look at your processed SC 143 image and it appears you interpreted a lot of the background texture as pores, which explains the high calculated porosity. This is a tough image to process because of the high texture. To me it looks like very few pores in this membrane go all the way through…
Dave – When you get a chance please make a list of the wafers that you think have similar problems with including texture in Carrie’s results. We need to redo these until we get all the data correct.
Dave,
After re-looking at the image, I see what you mean. If you don’t mind making a list of wafers that look problematic, I can go back and redo them. If it helps, I uploaded the pictures with the red lines, so you can see exactly what I counted as a pore. You can either e-mail me the list or just give it to me on Tuesday, whatever is easiest for you. Thanks!
Yeah, 143 has almost no porosity. There are probably many wafers like this, so I’m not sure it’s a good use of time to go through 300 posts and simply make a list, since each one will need special interpretation, if useful data is to be extracted. Generally, a true through-pore is bright white with no significant texture, since there is nothing within it. Obviously, if there are features that overlap each other, they cannot be through pores. I’d be happy to help with any confusing images and general interpretation.
Regarding the 50nm cap, it makes sense b/c this is the metric that I usually use to determine how large pores are. Since there is usually a 50nm scale bar, I typically scan for pores with comparable size, and for 15nm thick material, I don’t recall ever seeing a 50nm pore. However, the pores in the b-series do have some larger pores, so if these are included in that graph, then maybe a 100nm scale bar got mistaken for a 50nm scale bar. B21 has several pores >50nm, for example.
Okay. Thank you very much.
I re-did 143 and ended up with a 7.02 nm average diameter, a cut-off of around 13 nm, and a 0.429% porosity. Does that sound more accurate? I uploaded the new histogram and bound image. I am going to look through my bound images and re-do ones where I mistook the surface texture for pores. Then, I will re-upload the Excel sheet. Hopefully that will make my results will be more accurate.
Also, I opened up the _r.mat file for B21 and it has quite a few pores that have diameters over 50 nm (actually, there are about 10 that are over 100 nm). I guess it had enough smaller (around 17 nm) pores to bring the average pore size down to 32.2 nm. Even so, I would have expected the average to be much higher.
To tell you the truth, I think that wafers that look like 143 will be impossible to process with our software, so should be omitted. The spots that are isolated by the program are not likely to be real pores, but bright imaging artifacts caused by diffraction. Samples with this much texture do not have any obvious pores to me, so I am not sure how to quantify them.
Regarding the pore sizes, it sounds like the 50nm ceiling is some type of programming glitch, then. If the data file has values over 100nm, shouldn’t this have been reported as the cut-off pore size, or is a different value used? 100nm is too large, as this is the major axis of a few of the highly elongated pores. Perhaps an average of the major/minor axis or an area-equivalent diameter would be more accurate?
hmm… That is interesting. First, I want to make sure that I am not reading the program wrong. To get the diameters of each pore, I clicked on 143 and then typed, “results.pores.EquivDiameter” into MatLab. Is that correct?
Oh. Maybe that gives a reading in pixels?
results.pores.EquivDiameter is in pixels. You need to divide by results.pixpernm to get it in nm I believe.
‘results.nmstats’ holds all the statistics in nm
Looking at the stats, in nm this time, the histogram seems correct. The _r.mat file has a mean pore diameter of 7.0 nm and a cut-off diameter of 12.4 nm.
Also, because the pore data a quoted above is actually in pixels and there is a conversion of 2 pixels/ nm for this wafer, the 100 pixel diameters are actually just around 50 nm. Sorry for the confusion.
Okay! Jess and I think we solved the mystery of the universal 50 nm cut-off point! We looked into the histogram program and saw that its default bins were too small- the default range was 5nm to 50nm. We changed the limit of the program to be 100nm because the largest pore diameter I found was 90.9 nm (sc b015). I am going to go back and fix the histograms and plots now.