The core premise of smart manufacturing is to minimize waste and increase operational efficiency through a connected network of sensors. The data from these sensors can be processed in real-time to diagnose points of failure, detect cause and effect relationships and assess supply chain risks. Unfortunately, given the investment involved in building a data and analytics infrastructure, coupled with privacy and security concerns, smart manufacturing does not aways correlate with lean manufacturing.
Electronic design automation based on computational modeling and simulation of materials, processes and devices has for a long time offered a virtual platform to plan, verify and optimize manufacturing yields. Given the complexity of semiconductor devices today, multi-scale and multi-domain modeling can be challenging, time-consuming, and computationally prohibitive to reliably map the end-to-end manufacturing pipeline from materials to systems.
In this talk, I will present a case for establishing a middle ground between multi-scale modeling and connected factories to draw correlations between materials, processes and devices. I will discuss the benefits and drawbacks of a novel methodology that leverages natural language processing and data mining of unstructured literature to map materials to systems. Like the cyber-physical requirements for connected factories, the proposed data mining approach also faces challenges associated with knowledge silos within the industry. However, unlike complete automation, the proposed approach has a potential to empower the next generation workforce, create a knowledge network based on existing information and identify gaps in data which do require additional computation and/or measurement.
Finally, I will share a strategy to demonstrate a minimum viable product based on this approach.
