The SciDB project illustrates that there is no general case solution for semantic identity.
If we distinguish between IRIs as addresses versus IRIs as identifiers, IRIs are useful for some cases of semantic identity. (IRIs can be used even if you don’t make that distinction, but they are less useful.)
But can you imagine an IRI for each tuple of values in the some 15 petabytes of data annually from the Large Hadron Collider? It may be very important to identify any number of those tuples. Such as if (not when) they discover the Higgs boson.
Those tuples have semantic identity, as do subjects composed of those tuples.
Rather than seeking general solutions for all semantic identity, perhaps we should find solutions that work for particular cases.