I have many posts on the NEWSGROUP, ANSWERS and comp.ai.neural-nets re XOR. Most can be retreived by searching on
The minimal configuration has a 2-2-1 topology with Nw = (2+1)*2+(2+1)*1 = 9 unknown weights to be estimated with only 4 equations. Consequently, there are an infinite number of solutions.
Nevertheless, I recall a success rate of only ~ 70% when training from a random set of initial weights generated by MATLAB's default NW algorithm.
So, just try 10 or more different random weight initializations. You should get at least 5 successful solutions.
Hope this helps.
Thank you for formally accepting my answer.