Productivity enhancement in optical semiconductor manufacturing: Early warning of failures in BHet laser fabrication

There are unique challenges in the fabrication and testing of optical semiconductor devices because, unlike typical silicon semiconductor devices, which can be tested cost-effectively on the wafer, many optical devices can only be 'fully' tested once the individual die is bonded to a heat sink. As a result, both the manufacturing process and test strategy need to be capable of predicting yields and product quality attributes based on limited sampling from a batch of the product. The large number of product quality and process variables that are measured for each device when it is manufactured makes it almost impractical to manually analyze them for valuable decision-making information. There is a need for understanding the complex, interactive effects of process variables on the product quality variables as well as automated analysis and discovery tools for extracting useful knowledge from the raw data. Such knowledge could have a significant impact on productivity and quality improvement. The objective of the present study is to identify useful correlations amongst the numerous process variables and to develop simple empirical models to predict important process quality indicators. Quality is assessed using statistical data analysis, focusing on the primary functions of major failures. Validation of the data set also demonstrated that linear models were accurate in predicting new data points for some of the output variables, whereas the variation of some output variables could not be explained using the available industrial data bank. Some models were powerful in making predictions and to provide a clearer insight in determining the key factors in manufacturing of BHet. BHet is a directly modulated laser operating at 2.5 Gb/s and reaches up to 360 km. Application of this research for failure prediction at an early stage of the manufacturing line could result in a dramatic reduction in the number of defective wafers that are completely processed and thereby lowering the overall manufacturing cost