A genetic algorithm approach for mixed-macro and standard cell placement that directly incorporates cell membership information (circuit layout)

The objectives of the VLSI circuit design process are to integrate a system on a chip such that the chip size is minimal and the timing requirements are met, and to complete this process in minimal time. An important part of the design process is the placement stage, which determines the arrangement of the circuit components on the chip. The placement problem is NP-hard, thus contemporary placement tools use a variety of approaches in an attempt to meet the design objectives. Many modern VLSI circuits use a combination of macro cells and standard cells to implement systems on chips. These cells are pre-designed circuit components that perform specific system functions. Since the placement problem is NP-hard, contemporary placement tools use heuristic optimization methods. Simulated annealing is a popular optimization method for placement; however, recent applications of genetic algorithms have yielded improved results. Each major system function is often implemented by a group of cells. Cell membership information identifies the functional groups and their cells. Contemporary placement tools often use this information to arrange cells on the chip according to groups. The current approach applies a separate optimization stage to each group. However, this approach is inefficient with respect to meeting the design objectives. This dissertation describes the development of a new application of cell membership information to the placement process. Instead of applying a separate optimization stage to each group, the cell membership information is incorporated directly into the optimization method. For this research study, a genetic algorithm for mixed macro and standard cell placement was developed such that the algorithm uses the cell grouping information. Another genetic algorithm for mixed macro and standard cell placement was developed to use the current application of cell membership information. The performances of both tools were compared: the new approach was found to produce better placement solutions in faster time.