Neural somatotopical mapping for VLSI placement optimization

Describes how a typical combinatorial optimization problem, the module placement problem in VLSI design, can be rapidly solved by neural networks based on somatotopical mapping. Models to solve this problem were designed using principles acquired from an understanding of the competitive and self-organizing properties of the neural networks. It is also shown that such networks can work well for the two-dimensional optimization problem. In comparison with conventional approaches, the network has shown distinct advantages. The results obtained seem to surpass those of classical placement algorithms in efficiency and computation time. The algorithm based on the neural somatotopical mapping is a general optimization technique and appears very promising for solving other problems in the areas of VLSI CAD (computer-aided design) such as routing and compaction