_Revolutionizing semiconductor manufacturing: How a chemical company optimized filtration with ML and IoT 

Summary 

An Asian chemical company operating in the semiconductor manufacturing sector, aimed to enhance their production quality assurance processes by implementing advanced data analytics and anomaly detection solutions. TT PSC’s goal was to ensure higher consistency in product quality and minimize manufacturing errors by leveraging machine learning (ML) models and advanced IoT-based data analytics. The quality control process in the chemical industry is closely linked to the filtration process, which is essential for achieving the proper composition of products, free from impurities that could affect their effectiveness. 

Business Challenges 

• Lack of holistic data insights: The absence of a unified data platform made it difficult to monitor and analyze filtration processes efficiently. Data was scattered across multiple spreadsheets and systems, hampering the ability to track process anomalies or deviations from established standards. 

• Inefficient data analysis processes: Manual data analysis and reporting processes were time-consuming, reducing productivity and leaving limited time for higher-value tasks. 

• Quality assurance and process control: Despite their mature semiconductor production processes, our customer sought more reliable tools to ensure consistent, high-quality production batches and meet customer expectations for repeatable product quality. 

• Absence of real-time insights and monitoring: The company struggled with real-time visibility into key production metrics and could only perform post-production analysis, delaying potential problem detection and resolution. 

Solutions 

Our team developed a tailored anomaly detection application to address our Partner’s challenges. 

• Unified data management: implemented an automated data pipeline using Airflow platform to consolidate, transform, and contextualize data for efficient analysis. 

• Scalable and cost-effective deployment: leveraged AWS infrastructure and MLFlow for managing model lifecycles, enabling seamless deployment across testing and production environments. 

• Custom models: developed advanced models focused on filtration anomaly detection, including flux prediction and filter resistance estimation, helping our customer identify deviations that previously went unnoticed. 

• Dashboard and visualization: a custom dashboard was built within ThingWorx to provide our Partner’s production team with clear, actionable insights, showing deviations and anomalies visually in real-time, reducing manual data processing efforts. 

Main Benefits and Results 

• Enhanced data efficiency: our Partner reduced the time spent on manual data analysis, improving productivity and enabling their team to focus on strategic activities rather than labor-intensive data processing. 

• Real-Time insights and decision-making: with real-time data available through dashboards, our customer can make more informed decisions, leading to faster problem detection and resolution, reducing production defects and errors. 

• Improved Quality Assurance: the app ensures consistent product quality across batches, enhancing customer trust and satisfaction by identifying potential anomalies earlier and taking corrective actions proactively. 

• Increased productivity and cost efficiency: The automation and integration of ML models eliminates the need for our Partner to build separate models for each product, streamlining the process and reducing costs associated with manual model training and deployment.