Technical Resources
Technical resources on semiconductor process control, Virtual Metrology, AI-based manufacturing and production operation.
How to predict and control process outcomes
Why process control is harder to build and operate
than the algorithm itself
Compare equation-based control, single algorithms and Amfibian from the perspective of model building, pre-deployment validation, production operation and reuse.
Are EPD and OES enough?
Finding an endpoint and keeping final quality on target are different control problems.
Run-to-Run APC: Feedback + Feedforward
How repeated control updates the next run from the previous result and current system state.
Wafer Profile Control and multi-zone process control
Model the relationship between zone actuators and spatial process results to manage target profile and WIWNU, not only wafer averages.
Feed position-dependent batch variation
into the next recipe
A control structure for calculating the next batch's zone conditions from actual wafer outcomes in batch equipment.
Virtual Metrology: fill metrology gaps
without breaking the control loop
Estimate process outcomes where metrology is sparse or delayed and connect those estimates to the next decision and control action.
Why TTTM is not enough: the role of APC
in wafer manufacturing
Matching tool baselines and keeping wafer results within target over time are different problems.
How to operate manufacturing AI
in production
Why manufacturing teams keep
solving the same problems again
How hard-won process knowledge can move the starting point forward for the next tool, product and process.
Tacit knowledge has an expiration date
Why Excel macros and personal automation rarely become organizational knowledge, and why past decisions need to be reinterpreted in today's context.
AI models need production engineering too
Manufacturing separates process development from mass production for a reason. Data and AI products also need a system that connects preparation, model development, deployment and continuous operation.
Why manufacturing software should be operated
as a product, not an SI project
Software used every day in production and expanded across processes needs a product operating model rather than one-off project delivery.
Do you need SPC and FDC first before starting manufacturing AI?
The limiting factor is often not the number of infrastructure layers, but whether the system can be operated continuously with limited engineering resources.
What does Physical AI look like in semiconductor manufacturing?
Physical AI connects software understanding of a physical system to real operational decisions and actions.
Beyond diagnostics and monitoring: the next step is process control
Why detecting anomalies, predicting quality and changing process conditions solve different manufacturing problems.
What separates a fab that uses AI
from a fab that runs on AI?
In production, the important question is not whether a model exists, but whether it can keep operating reliably as the real process changes.
Why do so many PoCs fail to reach production?
The gap between PoC and production is often driven more by operating trust, data consistency, change management and workflow integration than by model accuracy.
Fab IT modernization: from system projects to event-driven data platforms
How manufacturing IT is moving from MES-centered system architecture toward event-driven data flows and lakehouse-based platforms.