AbstractA new procedure for the optimum design of optoelectronic devices is explained in this paper and an automatic search is made simultaneously for the structure satisfying various demands. The feature of this procedure is in the introduction of cost by which the quantitative evaluation of the structure becomes possible and the global search for the required structure by simulated annealing can be carried out.First, the definition of cost and details of the optimization procedure are clarified. In optimization, in addition to convergence to the minimum point of the cost function (the optimum configuration from theoretical viewpoint), the convergence also is possible in structures with great tolerance to fabrication errors (neighborhood cost (NC) and finite temperature annealing (FTA) methods).Next, these three proposed methods are used in the design of a pnpn differential optical switch and the effectiveness of the methods is verified. The method of cost expression, the relation between annealing parameters, and convergence are investigated. It is shown that cost expression with large degree of freedom improves the search for high‐performance structures and the initial temperature of annealing or the fixed temperature of FTA method is the important parameter which sets up the probability of acceptance. Further, it is shown that the convergence cost is inversely proportional to the time spent in annealing. These results are useful guidelines in the optimum design of arbitrary optoelectronic device