To confirm what you said: yes, only a Y-DNA test will do what you want. Ancestry DNA used to sell Y-DNA tests years ago. I assume this would be how the gentleman from Switzerland would have conducted the study. Are you sure it was on Ancestry DNA?
If there has been a significant study, then there is probably a surname project on FTDNA right now. If you reveal the surname in question, then I can check for you. Or you can just Google for yourself "Smith surname study FTDNA", but replace "Smith" with your desired surname. You can poke around in the study pages without testing at FTDNA or otherwise having an association with them. That will give you an idea of the size of the study, which should be a factor in deciding whether you want to proceed.
What you want to find is the "chart" for the study. The chart shows the DNA test results for men in the study, anonymized, but showing the most distant paternal ancestor for each person in the study. The study administrators will cluster the testers into groups with similar test results. Hopefully what you will see is a group labeled "Switzerland" and another labeled "Winterthur", or at least, that area of Germany.
If that is the case, then you may have success testing your theory. Test your uncle, and either he will come back as a decent match to one cluster, or a decent match to the other, or not a match to either. If you share a common ancestor 500 years ago, then you have a good chance of detecting this with an STR test.
You mention that Y-DNA haplogroups aren't very specific. That is true. Thee are two ways to test a Y-chromosome. You can test STRs, or you can test SNPs. The basic FTDNA test tests STRs only. 23andme estimates your haplogroup by instead testing just a few SNPs.
STR mutations happen much more frequently than SNP mutations. That means you may match exactly the same small collection of SNP mutations with your paternal ancestor from 2000 years ago, as will a bunch of other people around Europe, so you can't get genealogical use out of this. Your STRs, on the other hand, will probably differ considerably from your ancestor from 2000 years ago. So if you share a lot of STRs with a match, your common ancestor was probably more recent. 500 years is a reasonable time-frame to be looking at to test your theory with STRs. If you find a match in that time-frame, however, you will not know exactly how many generations back the common ancestor is. I should warn you that adoption of surnames happened in many places only around 500 years ago or more recently, so it could be that your paternal ancestor back then did not have the "surname" that you think. It may happen then that you have matches with a different surname than you are expecting.
For completeness: there is a much more expensive test that will test your STRs and a lot of SNPs. Or you can even get a Whole Genome Sequence, which will test everything. If you test a lot of SNPs, then you will have a much more precise subclade of your haplogroup. My impression is that most of these subclades still formed more than 500 years ago, but that in some cases it is possible to pinpoint the subclade around 400-500 years ago, which starts to make those SNP tests genealogically useful. There is a huge "y-tree" that just keeps getting more detailed as more people test, and the lower branches are starting to enter the "genealogical timeframe". The dream is that this will eventually be the case for a large chunk of the tree.