Recent developments have aroused the interest of researchers in the application of chaotic neural networks to combinatorial optimization problems. In this paper, we introduce a new approach, which is termed as Sequential Chaotic Annealing. The approach combines chaotic neural networks and ideas from the theory of nonlinear optimization. The proposed neural networks are adaptive in the sense that the network "learns" the right cost or energy function to optimize. Sequential Chaotic Annealing is applied to multilayer channel routing using the reserved wiring model and restricted doglegging. We show that the proposed approach improves convergence to valid solutions and reduces the sensitivity to the initial states of the neurons.