I was reading about a few studies on estimating functional connectivity between brain areas using fMRI signals. However, as far as I know that fMRI has a very poor time resolution, roughly in the order of 1 second. Individual neurons communicate with delay less than 100 ms, and many brain functions happen within a second, in which case fMRI wouldn't have enough time resolution to determine if things happened simultaneously, or in a sequence at all. This problem would be even greater for causality analysis such as Granger causality framework.
So my question is, what is the theoretical, and practical limit of time resolution in fMRI on human?