Artificial intelligence applications are expected to drastically change semiconductor manufacturing, exemplified by an estimated 17% reduction in manufacturing cost [1]. Such applications of AI are expected to impact all levels of semiconductor manufacturing, from the operational and tactical to the strategic level [2-4].
The development and deployment of AI often occurs in several phases, for risk and change management purposes. In phase 1, AI improves existing decision support systems with minimal disruption to allow for gradual acceptance. In phase 2, AI informs existing automation through incorporation of frequent, but low-risk, predictions or control points into existing systems. Finally, in the last phase (not depicted in Fig 1.), AI automatically takes high-level impactful decisions at the fab automation level.

For example, at the strategic level, supervised learning allows for improving forecasting of market demands [4] and cost & energy of manufacturing (phase 1), while reinforcement learning allows to dynamically and automatically select demand forecast models (phase 2) [5].
At the tactical level, preventative maintenance planning [6] allows to prevent major disruption to the fab, or more accurate lead time prediction for improved planning and scheduling [7].
While, at the operational level, applications range from supervised learning for wafer defect detection [8] and virtual metrology [9], unsupervised learning for equipment health monitoring [10-11], to improving scheduling and dispatching through the incorporation of all these AI-based insights.
As expert knowledge and data-extracted insights are increasingly encoded into AI models, the integration and automation of processes across the operational, tactical, and strategic levels improves. For applications in AI, the sum is therefore more than its parts.
Our fab scheduling solution, minds.ai Maestro, builds upon these advances in AI. It optimizes user-specified fab-level KPIs through dynamic coordination of a fab’s existing lower-level scheduling and dispatching systems. It leverages the power of Deep Reinforcement Learning to incorporate knowledge from all levels in order to dynamically adapt to the changing fab environment (see Fig. 2).

References
[1] Scaling AI in the sector that enables it: Lessons for semiconductor-device makers
[2] Machine learning in manufacturing: advantages, challenges, and applications
[3] Machine Learning for industrial applications: A comprehensive literature review
[4] Data-driven intermittent demand forecast framework to empower supply chain resilience and an empirical study in electronics distribution
[5] Deep reinforcement learning for selecting demand forecast models to empower Industry 3.5 and an empirical study for a semiconductor component distributor
[6] Opportunistic maintenance scheduling with deep reinforcement learning
[7] Lead time prediction using machine learning algorithms: A case study by a semiconductor manufacturer
[8] A Review on Machine and Deep Learning for Semiconductor Defect Classification in Scanning Electron Microscope Images
[9] Virtual Metrology in Semiconductor Manufacturing by Means of Predictive Machine Learning Models
[10] A Scalable Deep Learning-Based Approach for Anomaly Detection in Semiconductor Manufacturing
[11] Deep Learning and Its Applications to Machine Health Monitoring: A Survey