You can totally reference non-free software in your work. The licence under which it is released has no impact on the technical and scientific merit of its algorithms and features. The only issue is that it is harder to know exactly what the system does if the code is not open, and that it will be harder (and probably more expensive) to reproduce any research based on it.
There are a lot of papers out there (often coming from private-sector reserachers rather than from academics, but still) which are exclusively focused on industrial products under proprietary licence. What is important is what is said about them, to which extent they are analysed, whether they are properly compared to other existing systems... Of course, it is easier to do that if you have inside knowledge of the product (i.e. you have participated to its design for instance), but it will remain difficult (not impossible) for the community to validate or refute your claims.
In a state of the art, if you know that a non-free system has a relevant feature, or solves a related problem, or contribute to the field, you just cannot ignore it, it would be a deep mistake, scientifically speaking (*). However, it might be useful to take precautions about the verifiability of the system's claim.
For what it's worth, the software development attached to my PhD thesis relied on a proprietary framework. It's not ideal in my opinion, but it doesn't invalidate the corresponding research, as long as the choice is duly motivated (for instance, no other framework provides feature X).
(*) "Listen to me: someone who doesn’t know the truth is just thick-headed. But someone who does know it and calls it a lie is a crook." (Berthold Brecht, Life of Galileo, scene 9 - you want to be a researcher? This play is a must read).