Joint Production and Substitution Strategies under Yield Uncertainty
|School||Shanghai Jiaotong University|
|Course||Management Science and Engineering|
|Keywords||Random Yield Random Demand Demand Substitution Product Depreciate Customer Lifetime Value Available to Promise|
Many high-tech products, especially the semiconductor products,tend to lose more than60%of their values in first six month afterlaunched, in that the life cycle of which are very short (one or two yearsin general). Meanwhile, due to dynamic market demand, long lead time,expensive facility cost and continuous technology innovation, thesemiconductor industry suffers from intensive capital cost and hugemarket risk. Therefore, the effective yield management of manufacturingis critical to reducing the waste of capital, lowering the risk of failureinvestment, while improving the profitability of company.The manufacturing processes of the semiconductor products areextremely complex, which lead to the highly unstable performances ofthe semiconductor products even if they are produced in the same batch(especially for the high-tech products). The products with differentperformances are usually divided into different grade at the allocationstage. However, at the allocation stage, the demands for different gradeproducts are downwards substitutable; namely, the shortage of inferiorproducts can be satisfied with superior ones. As a consequence ofrandomness in both yield rates and demands for classified products, itbecomes a huge challenge for semiconductor manufacturers to makedecisions on the optimal input of raw materials at the production stageand the optimal allocation policy at the allocation stage.The manufacture system with randomness in both production andallocation stages are named Random Yield and Allocation (RYA) System. In this paper, we focus on the raw material input and the allocationproblems of a RYA system with multi-selling period, multi-product andupgrade substitution. Mathematical programming, and intelligentoptimization theory are used to find the relationship between thecontinuous demand and the decision on the production and the sellingpolices. Stochastic process and CRM theory based on the value ofcustomer life cycle are studied in the thesis. In order to analyze problemand test the validity of model and conclusion, instance statistics andsimulations are used. The research mainly focuses on three aspects:(1) The problem of the random system with unsynchronizedproduction and allocation processes. We focus on the raw materialquantity and the optimal allocation policy problem of the RYA systemwith single production period and multi allocation periods. The allocationstage is divided into multiple allocation periods, and demand for eachproduct in each period is independent with each other. This problem canbe formulated as a multi-parameter stochastic dynamic programmingmodel with the objective to maximize the total profits in the productionand allocation stages.The theoretical analysises are used to find the optimal substitutepolicy in the allocation stage. Based on the comparison of several relatedmodels with different allocation polices; the simplified solution algorithmis setup. The example studies show that the algorithm is effective ineliminating the unimaginable workload.(2) The problem of the random system with asynchronizedproduction and allocation processes. This study focuses on themulti-periods production and multi-periods allocation simultaneously,that is, the products is made in certain batches, the selling periods begin after the first batch of the product is done. The length of one productionperiod equals to that of one allocation period. The allocation decision ofthe products is difficult because of the continuity of the demand. So ATPtheory (available to promise) is adopted in this research. The allocationpolicy basing on the ATP theory is setup through an experiment on a RYAsystem with two products, and multi-periods. Then, the RYA system withmulti-product, multi-period is examined by simulations. The solution forcorresponding optimal raw material input, and the simulation are givenand the numerical example shows that this solution is feasible.(3) The yield management problem considering the customerlifetime value. The customer lifetime value model is formulated in thesystem dynamic simulation framework, and the simulation result showsthe relationship between the customer lifetime value and the raw materialinput quantity. After that, the production quantity optimization problemfrom the side of raw material supplier is analyzed in the suppliercontrolled supply chain. The objective is to maximize the customerlifetime value of the whole supply chain, and this research can be ofsomehow helpful to encourage manufactures to pay more attention onlong-term profits.The complexity, randomness, dynamic and the substitutability of thesemiconductors leads to huge difficulty in modeling and solving theproblem of decision making of a RYA system with multi-product andmulti-period. Thus, the simplified algorithm and crucial technology aregiven to lower the difficulties.Comparing with the existing literature, the contributions of thisresearch are:(1) The existing literature focuses separately on yield quantity or the allocation policy, while the joint studies on the decision making are madein single period. The joint research on both raw material quantity and theoptimal allocation policy of a RYA system is done in this thesis.(2) The problem is formulated as a stochastic dynamic programmingmodel. The proposed allocation policy is much optimal than the newsboyallocation policy and the myopic allocation policy. Then the bounds ofthe decision parameters are given and an effective algorithm by modelcomparison is setup. The experiments show that the proposed algorithmis effective.(3) All the related factors to the yield management of asemiconductor manufacturer, such as human resource, capital, cost andmarketing are considered in a system dynamics model. After severalrounds of simulation, the relationship between the yield quantity and themanufacturers profit is found. The simulations also show that thesemiconductor manufacturer’s profit decreases quickly in a relative shorttime period after the products are launched. Comparing with thetraditional optimization method, system dynamics simulation can showthe results more visibly.(4) In a RYA system with synchronized production and distribution,the real time allocation decision is difficult to be made. Thus the ATPtheory is introduced to build an allocation policy. The comparison withthe ideal solution by numerical studies verifies the effectiveness of theproposed allocation policy.