We are living in an era where every industry is attempting to adopt artificial intelligence (AI) and automation into their manufacturing environment. However, most of the semiconductor backend operations are still heavily dependent on humans. This limits the efficiency of the manufacturing processes and imposes quality risks on final products. Although IoT devices and traceability are being widely used in the semiconductor environment, there are still many structured and unstructured useful data being hidden or scattered around. It requires an ecosystem to unleash the full potential of these data. Quite often, end users struggle to apply data mining techniques to discover trends and patterns in the data due to the lack of such ecosystem. Questions that are commonly asked are how artificial intelligence techniques can be applied in backend semiconductor environment using supervised and unsupervised machine learning or deep learning models. What kind of use cases can be established to help the end users to improve efficiency, predict process failure for yield improvement and make real time decisions? Without a proper ecosystem, many are just performing general data analytics, simple predictive maintenance on equipment and did not fully utilize the power of AI. Despite all the challenges, many Integrated Device Manufacturers (IDMs) have embarked on their digital AI transformation journey. Therefore, in this paper, we describe a complete ecosystem that covers the data analytics lifecycle from data acquisition, AI model development, model deployment to continuous improvement. The processes have been automated with logistic automation solutions to reduce human interaction and errors to ensure data collection efficiency and accuracy. To illustrate the benefits of this ecosystem, two AI applications are discussed in this paper with real production use cases. In one application, we developed a supervised machine learning model to predict the quality of the bonded wires in real time using the process signals streamed out from wire bonder equipment through an edge device. The other application is to detect the abnormalities in the process signals during production using an unsupervised neural network model. Both use cases have yielded promising results. This ecosystem is evolutional in transforming the traditional human operated manufacturing into an integrated smart manufacturing. It vastly improves production efficiency through automation solutions and production quality with the implementation of AI.
