I have a semiconductor customer that is starting a JMP Pro evaluation to see if FDE can help with a common issue across the industry called random telegraph noise (or burst noise). Essentially they have current by time curves and are testing different conditions in order to remove noise which are bursts or jumps in the current. We are going to try functional DOE with golden curve analysis where the target curve has no noise. The variability they see across the range of conditions can be a couple jumps in current to very frequent (100s) jumps where the jumps are of different magnitudes. Frequency and magnitude of the "jumps" is important. I found the image included below after doing a google search for random telegraph noise. It represents about the midrange of variable that they have in their data (their data seems to have a consistent baseline).
Wondering if anyone has experience with this semiconductor use case or if anyone has an opinion about the FDE approach or if there is another approach to try?
The approach they have tried is to fit multiple multimodal distributions of current to determine the number of magnitudes of the jumps then they would process the data into step functions then count the number of jumps for each magnitude. As you can imagine, this is a lot of data processing through scripting and then they are stuck with how to evaluate the conditions they tested.. There must be a better way.
