University of Twente Student Theses

Login

An optimization approach between service level and inventory via simulation : an example from the semiconductor industry

Lingelbach, S.E. (2017) An optimization approach between service level and inventory via simulation : an example from the semiconductor industry.

[img]
Preview
PDF
5MB
Abstract:Company & Motivation This graduation project is conducted as part of the Industrial Engineering and Management master program in cooperation with Infineon Technologies AG. Infineon is a German semiconductor manufacturer producing chips, sensors, and microcontrollers. To stay competitive and satisfy customer demand quickly Infineon places inventory at various stock points within their supply chain. However, the more products are stored, the higher the costs due to the binding capital effect of stock. Thus, Infineon has to balance the trade off between high stocks (characterized by a high α-service level) and high costs when examining its supply chain planning processes. In this thesis, we concentrate on the planning process of two products: chips for contactbased and contactless payment of the Chip Card & Security (CCS) department. The relevance lies in their high production volume and revenue share of more than 25% of CCS‘s total revenues. Research objective The graduation project aims to solve the below stated research objective: Improve the supply chain planning process according to the service level and respective costs at CCS for two particular products considering the stocking strategies as well as the approach of quantifying the amount of wafers to be released to production. The stocking strategy concerns the decision where to place inventory and which amount to be stored. The production release approach examines the question how to quantify the amount of wafers (release quantity) to be started in production in advance. Usually, the production of wafers, which are thin slices of semiconductor material and serve as basis for many products, is started on forecast due to long processing times. This enables faster response to customer demand. A clever chosen approach of estimating the needed quantity helps cutting costs as stock levels can be reduced at the same α-service level. Methodology The existing simulation model (discrete event simulation) of Infineon‘s flexibility & econometrics team is used to study various scenarios. These combine different stock strategies and production release approaches among others the current practice. Before conducting the simulation study we require to parametrize the simulation model to the needs of the two exemplary products. This involves to ensure that the generated demand by the simulation model is similar to the observed demand of the products such that results are valid. An iterative approach is performed consisting of setting the parameters in the demand generation method and subsequently assessing the fit between the generated demand and observed demand. The fit is assessed by a modified Kolmogorov-Smirnov approach where we compare the total area between the cumulative distributions of the two demand series. The iterations are stopped when the total area is < 10% of the area below the cumulative distribution of the observed demand. To check the consistency of the chosen demand series we further apply the Chi-Square test. Analysis of current situation Infineons supply chain has three stock points (up- to downstream): the master storage, die bank, and distribution centre. The amount of products stored at these stock points is determined by the target reach. The target reach is defined as the safety stock in number of weeks. Currently, CCS has a target reach of 13 weeks at the master storage, and no stocks at the die bank nor the distribution centre since the customer order decoupling point (CODP) lies at the master storage and thus products become customer specific in the downstream manufacturing steps. Storing at the master storage employs the risk pooling effect. The stocks are managed periodically (per week). The production up to the master storage is done on forecast by using a four month moving average (MA) over the historical data. The remaining manufacturing steps are continued when a customer order arrives. The overall performance can be given by the α-service level. The α-service level becomes either 100% when all orders are satisfied by the on-hand inventory during period t, or it becomes 0% when demand is not satisfied completely from stock. Currently, the α-service level is 98%. Conclusion • Both new production release approaches: a simple MA over five weeks as well as single exponential smoothing (SES) outperform the current approach that uses a simple four months MA since they allow faster reaction in production as fluctuations are not as smoothed out as with a large time horizon of four months. • Applying either of these new approaches costs can be cut by 40% since the target reach can be reduced from 13 to five weeks while keeping an α-service level of 98%. • Comparing the simpleMA over five weeks and SES, the moving average performs slightly better. In addition, as it is easy to understand and to apply, we recommend to use the simple MA with a five week time window. That is, reducing the current time window from 16 to five weeks. • When keeping the current production release approach, the target reach at the master storage can be reduced from 13 to about eight weeks while having only a marginal drop by around 0.5% in the α-service level of currently 98%. Recommendations • Enhance the demand generation method of the simulation model such that it is able to create intermittent and autocorrelated demand which is currently not supported by the simulation model. Note, the products we are considering do not show autocorrelation nor are classified as intermittent, however there are autocorrelated and intermittent products at Infineon. • In addition, implement machine capacity and idle costs as currently capacity is unlimited and costs are solely evaluate according to the WIP and stock levels. However, in reality capacity is restricted and idle costs play an important role as machines are very expensive. • Last, more elaborated approaches of quantifying the amount of wafers to be started in production in advance such as advanced exponential smoothing techniques, Holt Winter procedure, or ARIMA models may be examined when the demand shows a trend, seasonality, or autocorrelation.
Item Type:Essay (Bachelor)
Clients:
Infineon Technologies AG, Neubiberg, Germany
Faculty:BMS: Behavioural, Management and Social Sciences
Subject:85 business administration, organizational science
Programme:Industrial Engineering and Management BSc (56994)
Link to this item:http://purl.utwente.nl/essays/74664
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page