Force directed self-organizing maps for L-shaped cell placement using delta learning rule

In a previous paper we have described a force directed self-organizing map (FDSOM) based on Kohonen´s algorithm for resolving the cell placement problem using the circle cell model. We show here how an FDSOM with an additional hidden layer can be designed to resolve the placement problem with L-shaped cells. Moreover, the actual-shaped cell model is considered. The proposed network model easily models an L-shaped cell as a pair of rectangles using hidden neurons. Using the collective computing property of neural networks, these L-shaped cells can correctly interact with each other and finally converge to a good placement result. Placement results of L-shaped test circuits are presented which demonstrate that the proposed method is better than previous methods. The CPU time of the proposed method running on a sequential machine is also faster than that required by the simulated annealing method. The quality of the solutions obtained using the proposed model is also discussed. The proposed algorithm succeeds in yielding a proper placement when different initial configurations are used