Any problem is solvable with two layers.
The first layer consists of 1 node per input pattern, with activation and weights such that only one node is active that matches the pattern. And a second layer of weights that provides the output per input pattern.