Hardware annealing for fast retrieval of optimal solutions in Hopfield neural networks
B.J. Sheu, B.W. Lee, Chi-Yuan Chang · 2002
Due to the feedback characteristics of Hopfield networks, the solutions often get stuck at local minima where the objective functions have surrounding barriers. The theory and procedure of hardware annealing, which can retrieve the optimal solution in parallel, have been developed. In hardware annealing, voltage gains of output neurons in asynchronous VLSI neural chips are increased from an initial low value to a final high value in a continuous fashion. Hardware annealing can be applied to pure analog and mixed analog-digital neurocomputing systems. It achieves a speed-up factor of more than 10000 times over simulated annealing on a SUN-3/60 workstation.>