In previous work a nonlinear neural framework, the Generalized Hopfield Network, was proposed as a means of solving nonlinear optimization problems. GHNs implementing the three most important nonlinear programming algorithms : Augmented Lagrangian, Generalized reduced Gradient and Successive Quadratic Programming methods were demonstrated. In the present work, the GHN construction is extended to discrete decision problems, specifically, mixed integer nonlinear problems such as arise in process design and scheduling applications. The proposed formulation does allow a natural parallelization of the optimization calculations, distributing the computational burden among a number of digital processors linear in the problem size. Furthermore, network solution appears to be amenable to the use of simple numerical integration methods with the integration in each processor carried out asynchronously. The use of a disjunctive constraint is shown to be superior to the sigmoid filter construction. The penalty parameter of the Augmented Lagrangian implementation appears to serve as a relaxation parameter which controls the extent of the domain within which alternative local optima are sought. Two examples are given to show the potential of the method for finding global optima to MINLPs.