VLSI placement using uncertain costs

Many objective functions used to evaluate placement quality contain uncertain parameters (e.g. channel width, wire length). While these values can be estimated, they cannot be precisely known until the layout is finished. As a result, current automatic placement algorithms use ´expected´ values in their objective functions and return one possible estimate of placement quality. An algorithm that uses the full range of potential values when computing placement cost can yield a more credible prediction of placement quality and reveal more about the structure of optimal configurations. An interval-based approach to modeling uncertainty in automatic placement is proposed. It is used to illustrate the authors´ methods by implementing an interval branch and bound placement algorithm.<>