Research

Human-AI Teaming for Systems Science and Engineering to Accelerate Scientific Discovery and Engineering Innovation

System-driven AI Modeling and Optimization Framework to Accelerate Complex Systems Design 

Complex systems design especially involving uncertain high-dimensional parameters is the fundamental research issue commonly appearing in the science and engineering, such as physics, chemistry, aerospace, semiconductor and automobile. The design process is extremely time-consuming and cost-intensive. Extensive experiments and testing need to be conducted for identifying the optimal system design parameters to achieve the best system performance. This research is to develop a system-driven AI modeling and optimization (SAIMO) framework for the accelerated complex engineering systems/experiments design. The proposed SAIMO framework consists of three main components. First, by utilizing the data observations from both digital models and physical experiments, a machine learning based predictor will be developed to provide the predictive system performance accurately. Second, by scientifically identifying the candidate design set, a sequential optimization and control technique will be developed to evaluate and capture the best design. Third, an uncertainty quantification tool, with comprehensive consideration of model bias, computational error and experimental error, will be customized to measure the risk of decision-making on the system parameters design. The SAIMO framework is a closed-loop decision support system that integrates the cutting-edge optimization, statistics, simulation and machine learning techniques. With a limited number of high-fidelity simulated and experimental observations,  the SAIMO framework can build the emulator for complex system and further accelerate the system design and scientific discovery.

Geometry-based Computing Framework for Large-scale Wafer Inspection Optimization

Optical and e-beam inspections are currently two major defect inspection techniques in semiconductor manufacturing.  With the rapid technological development, defects in the wafers are becoming smaller and smaller to the nanometer level, most of which cannot be detected by optical inspection due to its low resolution. E-beam inspection (EBI) is capable of detecting those smaller defects and much needed than ever to be used for in-line inspection to further improve the quality and yield of wafers. However, due to EBI's high resolution and extremely large-scale optimization issue in wafer inspection, it takes much longer time to scan the wafer and suffers from very low throughput compared with optical inspection, which significantly limits its in-line application. This research refines the scanning process of EBI technique as a fundamental computational geometry problem and will develop a  partition-shifting computing (PSC)  framework for this problem to address the extremely large-scale optimization challenge in wafer inspection. The proposed PSC framework partitions the original problem into predefined amount of subproblems with the guarantee of strictly theoretical accuracy, which yields to a polynomial-time approximation scheme (PTAS) under certain conditions. The PSC framework will also achieve the high performance computing through its well-designed parallel computing mechanism and has the potential to be integrated into the state-of-the-art inspection techniques.

Intelligent Decision Support and Analytics Framework for Complex Manufacturing Systems 

Modern industry is stepping into a new era that heavily focuses on the interconnectivity through industrial internet of things (IIoT) and cyber physical systems (CPS). A more holistic and better connected ecosystem for manufacturing firms needs to be created by marrying the physical production system with digitalization technology, big data, cloud computing and artificial intelligence (AI). System-wide integration of industrial software is also becoming more and more seamless and immediate. Those advanced technologies are giving birth to the next industrial revolution, and meantime significantly increasing the complexity and challenge of manufacturing systems management. The challenge derives not only from the scale of systems and customer demands, but also from the fact that a large amount of multi-source industrial data in manufacturing is not sufficiently fused, utilized and mined in a more efficient way to enhance the real-time decision making and analysis. The research is to develop a data and model-driven system framework, denoted by intelligent decision support and analytics (IDSA), to address the fundamental research issues of complex manufacturing systems, such as design, operations, quality and maintenance. By utilizing optimization, statistics, simulation and computer science, an AI-based algorithm automation method will be developed as the solver engine of the proposed IDSA framework. With the fusion of large-scale multi-source industrial data, domain knowledges and digital twin production systems, the IDSA framework will be capable of providing efficient intelligent decision and analytics algorithms for different industrial scenarios as a toolset to serve the complex manufacturing systems. The proposed framework can also be extended to address the decision and analytics issues in other complex systems.