Strictly digital neurocomputer based on a paradigm of constraint set programming for solving combinatorial optimization problems

The authors present an advanced neurocomputer system based on a programming paradigm of CS (constraint sets), together with an architecture of SDNN-3 (strictly digital neural network version 3) parallel processors. The paradigm of CS programming provides an interface between end-users and the neurocomputer for massively parallel processing, especially in combinatorial optimization. The paradigm helps to construct a systematically designed neural network with intersecting sets of neurons since it facilitates a direct representation of constraint conditions in a given problem. According to each of the conditions, individual sets can be designed using the k-out-of- n design rule. An SDNN-3 processor consists of three components that are an array of binary-synaptic connections, deterministically operated binary neurons, and set buffers of which each estimates an energy quantity in the constraint set