Global placement of macro cells using self-organization principle

A new neural network approach is presented for the global placement of macrocells. This algorithm is based on a learning algorithm for neural networks proposed by T. Kohonen (1988), called the self-organization principle, which has the property of topology-preserving mapping. Due to this property, topologically close circuit modules are located closely in the target placement region. Compared to earlier work on standard cell circuits, finite sizes of modules are considered during the self-organization process to reduce the overlaps among modules, which makes this algorithm applicable to the macrocell placement having large variations in module size and shape. Gradual expansion of the module size is adopted to maintain the original placement result during the overlap reducing process