Hi,
Oh yes, i see where you are coming from now. It takes a little experience with these kinds of things, where the one thing is relative to another thing rather than just one set value.
For example, although the strike voltage may change between the same exact part numbers made by the same manufacturer, they will not show that data the same way, but will use a statistical plot instead that gives us an idea how many pieces can be expected to deviate from the mean where the mean is like how "most" of the units behave. For example, out of 100 units maybe 90 will fit close to the mean by some error margin, while 80 will fit within a larger error margin. They dont always show that data though in the data sheet itself, you have to look on the manufacturer's site to see if they publish this data.
When you read the main data sheet, you assume they are talking about the mean. Later when you do the design, you may want to refer to the statistical data in order to ascertain the risk associated with going by the mean and perhaps make adjustments to the circuit.
As far as a semi log plot, they use that when one variable is better shown on a linear scale and the other varies widely. For the linear plot the data is usually of interest over the entire range without too much difference in significance between points, while the log axis is used to show a lot of range but without showing every single data point. For example, the linear scale may run from 0 to 10 in steps of 1, while the log scale may run 1,10,100,1000,10000, etc. That's because the difference between 10 and 11 on the log scale is not as interesting as the difference between say 10 and 20 or 10 and 100. This could be frequency for example, where if we know what the amplifier does at 10 Hertz we more or less know what it does at 11Hz (so we dont need 11 Hertz plotted with as much resolution), but we still want to know what it does at 20Hertz or 100 Hertz.
So the log plots are used when the significance between data points is not as great as between larger groups of points (like 10 and 100) and this helps to cover a much wider range of data. For example, if we had to plot the response for frequencies from 1Hz to 1Mhz we'd have to plot a million points, and even more significant is the fact that the response does not change that much between points like 999998 to 999999 Hertz, but it does change significantly between points 100000 and 1000000 Hertz or maybe even between 100000 and 200000 Hertz. So instead of having a plot that includes every single point we just use 1,10,100,1000,10000,100000,1000000.
Note that we can plot between 1 and 10 without too much difficulty:
1,2,3,4,5,6,7,8,9,10
But when it comes to doing 1 to 100 we'd be there forever, and if we shorten it to decades (linear):
10,20,30,40,50,60,70,80,90,100
which is not too bad, but then we run into the same problem when we try to do the same for 10 to 1000.
Using the log scale means we only have to show a few points per decade to get the main idea of what is going on across.
One of the things you can do to help understand this kind of thing is you can work through several examples of a design, trying to make it work the way it should, then try to figure out what is important by comparing designs. Try to figure out what can go wrong and how you can make it better.
I know it's not as easy as having a drawn out procedure, but sometimes that's the way it goes