A neural-type network for solving minimal energy path in real time

An analog neural type network is developed, which computes a minimal energy path in real-time under two-point boundary conditions. The network has many simple interconnected nodes operating concurrently in an asynchronous fashion. The energy equilibrium state of the network provides the solution. A mathematical formulation by the variational approach is first described. Then a mapping function is defined to convert the problem from a time domain to a spatial domain suitable for analog computing. A transfer function is derived and a node-connection weight matrix governing the evolution process of the network states is developed. Resistive building blocks and the structure of the network are designed which are suitable for analog VLSI implementation. Comparisons to the digital parallel computing are performed