The type of
complexity something must have to be a
problem for
evolution to explain.
Michael Behe is the one who coined this term, and he defines it as "a single system composed of several well-matched, interacting parts that contribute to the basic function, wherein the removal of any one of the parts causes the system to effectively cease functioning."
The definition proposed by Behe is not always workable. Study of the biology of various living things gives indications that many features of those living things have resulted from what might be called "jury-rigged design" - where something that was serving one purpose has mutated slightly and now either performs an additional purpose, or perfoms a new purpose instead.
True irreducible complexity would require finding a feature that not only stops working for the intended purpose* when one part is removed, but that serves no purpose whatsoever, and could not conceivably be adapted for a new one with minor changes.
* Technically, the term intended purpose makes no sense when discussion evolution, because there was no designer to create the feature for a specific purpose, it just happens to have evolved to perform whatever it does.