Speaker

Machine learning prediction models for semiconductor fabrication

Yield and quality of semiconductor chips are two main performance criteria for a semiconductor production. Based on historical manufacturing data, models can be trained that are able to give predictions of electrical performance measurements. Such predictions are useful to gain a real-time monitoring of the production performance, find corrective actions to optimize yield results and support an accurate production planning. Different sets of input data and various machine learning approaches were tested to achieve a high prediction performance.