In this paper we have addressed the problem of state assignment of Finite State Machines (FSMs). The approach used by us to solve state assignment problem is based on hyper-cube embedding. We have designed a new technique to efficiently solve the hyper-cube embedding problem by integrating two different techniques; one of these is gradient projection method while the other is a variant of Kernighan-Lin algorithm. The gradient projection method operates in continuous space and improves an initial feasible solution iteratively by tracing a search path in gradient descent direction. Kernighan-Lin algorithm operates in discrete space and also improves an initial feasible solution iteratively. We have integrated both techniques in such a way that output from gradient projection method is fed to Kernighan-Lin style algorithm. The effectiveness of the proposed technique is shown by comparing its results with another technique on a number of MCNC benchmark examples for logic synthesis and optimization.