The basic idea behind Kelvin (or four terminal) sensing is very simple, and is explained here (or in the posts above):
The devil is in the detail, though.
When you have IC design, for example - power transistor, you want to measure the current or voltage on it.
Now, power FET usually uses multiple metal layers (from 1-2-3 in old technologies, to ~20 in latest technologies), and multiple "fingers".
Where you "connect" in this massive metal network, is very important.
The sense point should be selected so that it provides kind of an average of the whole device.
The information about where this average is, can be obtained from proper simulation.
Such methodology for selecting an optimal sense point - where you connect sense or replica (small) device - to measure the current in the large device - is very important for sensing accuracy.