J. Foucher, S. Martinez, H. Ozdoba, J. Baderot, O. Cru, T. Ziraoui and S. Girard
Pollen Metrology,
France
Keywords: metrology, process control, artificial intelligence, deep learning, image processing, process optimization, yield
Summary:
The complexity of semiconductor industry development requires industries to digitalize their metrology and defectivity process to reach shorter time to market cycles. Process Engineers are confronted with obtaining more accurate data to better characterize their processes. The available data is highly heterogeneous, coming from different imaging and non-imaging techniques. Consequently, more complex analyses take more time and are human dependent for fine tuning. The impact is huge on R&D costs raising and production yield long ramp-up. Though, automation and robustness are key topics. The industry is facing several key challenges: firstly, versatility of metrology and defectivity capabilities has a major impact to handle a large variety of heterogeneous data. Secondly, specifications of measurements are evolving very quickly. A flexible solution where users can customize their own software to fit their specific environment is required and leverages internal data science work that is becoming more strategic. It enables users to be more efficient while maintaining consistency and quality of the extracted data. This final aspect also preserves the confidentiality of the data as the whole process can be performed locally. We have developed a software platform that proposes a solution for these challenges. For the versatility of analysis, we powered the system with a generic AI pipeline that can recognize objects taught by the user with few examples and a system of geometric constraints to define a custom measurement recipe to characterize accurately the object. We validate the approach on a wide range of objects from different applications. For instance, the semiconductor industry includes logic and memory from early stage to late integration. In case of highly internal specialized needs, the technology can be turned into a SDK (Software Development Kit) that allows to customize the software at all levels, from algorithmic capabilities to full GUI perspectives to easily tune the platform for environments such as R&D, production, built-in software for manufacturing equipment or microscope. It also gives an answer to scalability needs. Finally, the platform offers all capabilities to read, process, transform and export data. Most imaging equipment can be handled to load the data (SEM, TEM, Optical, Acoustic, etc), so the AI pipeline is helping for the automation of key raw metrology and defectivity data extraction. In a second time, this data can be displayed, aggregated, used for statistics or exported in reports. It can also be connected to Manufacturing Execution Systems or simulation software to improve the characterization process and reach targeted material performances in R&D and production environments. We demonstrate the capabilities of the software for both metrology and defectivity applications on several use cases, including logic and memory for the semiconductor, AR/VR materials, micro-LED displays, chemistry application and micro battery to show the versatility of the AI pipeline. We demonstrate the capabilities of the platform to process data into valuable characterization information. Finally, we show how the platform deployed into multiple industries benefit semiconductor industry to accelerate time to market in comparison to developing a pure semiconductor industry solution.