Variable Binding via STP

One of the big criticisms of neural nets is that they can't do variable binding.
A simple example is the red square and blue circle problem. If these are seen, red, blue, square and circle ignite, but how can the system (brain) know which colour the square is?
The most common way to do VB in simulated neural systems is via synchrony.
I think this has some capacity problems, but Shastri says it doesn't, but I think he's fooling himself.
VanderVelde has also done it with special circuits.
You can do both of these with fLIF neurons, but I've done it via short term potentiation (STP).
I also did it with LTP, but this seems to run into the stability plasticity dilemma.
The idea is that two items are bound, but the binding is erased relatively quickly (seconds or minutes).
There is evidence for STP.
It also doesn't require the CAs to keep firing to keep the binding.
You can also use all four types of binding in one system.
So, you could do binding of visual features via synchrony, but parse using STP.