From Defect Images to Die Prediction: How Intel Is Scaling AI in Advanced Manufacturing

2026-04-07 · Source: Big Data & AI News - EE Times · Field: Manufacturing & Industrial — Smart Manufacturing & Industry 4.0, Automation & Robotics, Manufacturing Operations & Management · Depth: Advanced, medium

Summary

Intel Foundry is strategically shifting its AI implementation from isolated analytics to "applied intelligence at scale" across its semiconductor manufacturing operations. This involves deploying production-grade AI systems continuously throughout the entire production chain, from technology development to advanced packaging and final tests. Rao Desineni, senior director of data analytics and AI at Intel Foundry, oversees a 300-person team managing petabytes of manufacturing data to support applications like defect inspection, scheduling, yield analysis, and anomaly detection. A key application is predictive die screening, which uses upstream data to identify marginal dies before costly packaging, preventing collateral damage in multi-die packages. Intel addresses challenges like imbalanced datasets and sparse sampling by using techniques such as conditional generative adversarial networks for synthetic data generation, enabling model training even with limited initial failure data.

Key takeaway

For AI Engineers and MLOps teams in manufacturing, this highlights the critical need to move beyond proof-of-concept AI to robust, scalable production systems. Your focus should be on integrating AI into existing factory automation, ensuring lifecycle management for model decay, and addressing challenges like imbalanced datasets with techniques like synthetic data generation to achieve tangible economic benefits and improved yield.

Key insights

Intel's AI strategy focuses on operationalizing AI at scale across semiconductor manufacturing to improve yield and efficiency.

Principles

• Prioritize production systems over experimental models.
• Balance quality improvement with economic optimization.
• Embed deep process knowledge into AI model design.

Method

Intel employs a "N−1/N+1" predictive die screening philosophy, analyzing upstream process data to flag suspect dies before expensive multi-die packaging, balancing quality against scrap.

In practice

• Use synthetic data to address cold-start problems.
• Integrate AI as decision-support, keeping humans in the loop.
• Focus on usability as much as model performance.

Topics

• Intel Foundry
• Semiconductor Manufacturing
• Applied Intelligence
• Predictive Die Screening
• Defect Classification

Best for: Machine Learning Engineer, Computer Vision Engineer, MLOps Engineer, AI Engineer, Director of AI/ML

Related on AIssential

• Scaling the Next Generation of Multi-Die Systems — The semiconductor industry is focused on overcoming practical scaling challenges for chiplet architectures in production, moving beyond initial adoption questions.
• Training the AIs' Eyes: How Roboflow is Making the Real World Programmable, with CEO Joseph Nelson — Computer vision, though advancing, lags LLMs due to real-world data complexity and demands efficient, task-specific model deployment.
• EE Times Europe Magazine – March 2026 — AI is fundamentally reshaping manufacturing and semiconductor strategies, driving demand for advanced memory solutions.
• The Apple Silicon Disruption — NVIDIA has displaced Apple as TSMC's top customer, shifting silicon industry leadership to data center AI.
• Intel is going all-in on advanced chip packaging — Advanced chip packaging is critical for AI-driven computing, offering a path for Intel to regain market share.
• From Factory Precision to Intelligent Robots: How Real-World Performance Shapes Industrial Edge AI — Production-ready industrial edge AI demands deterministic performance and integrated, concurrent compute for diverse workloads.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Big Data & AI News - EE Times.