Yield and allocation management in a continuous make-to-stock system with demand upgrade substitution

Yield management is crucial for high-tech industries to reduce over-production. However, it is very challenging for a manufacturer to make a yield decision under the circumstance of uncertain demand and stochastic product quality. For example, in semiconductor manufacturing industry, products are produced in batches within certain production cycle time. Even if the same raw material is put into the production system, the product qualities (measured by performance) usually are not identical due to random yield in the manufacturing process. Since products have overlapped performance specifications, products of the same batch are classified into different classes according to their qualities at the allocation stage. Products with higher performance can substitute for demand for products with lower performance. This type of production and allocation system can be specified as a make-to-stock system with single material, multi-products, multi-periods, upward demand substitution under random yield and uncertain demands. In order to maximize profit, the manufacturer needs to decide on the optimal input material quantity and the optimal allocation policy. The problem is formulated as a stochastic dynamic programming model. Because manufacturer׳s profit is affected by the allocation policies, ATP-r allocation policy is proposed to help manufacturer make real time allocation decisions. A numerical experiment shows that ATP-r allocation policy performs better than the traditional Myopic allocation policy. To reduce the computation burden involved in ATP-r allocation policy, we further propose a simplified algorithm and our experimental studies provide proof of the effectiveness of the proposed simple algorithm.