Renowned influencer James Moyne joins Samantha Duchscherer in an engaging discussion exploring the importance of integrating additional information and advanced technologies like Artificial intelligence (AI) into the semiconductor industry’s scheduling and dispatching process. The comprehensive series consists of four parts focusing on various subjects such as the importance and advantages of data, AI, and human involvement. It also delves into the obstacles faced and offers insights into the role of digital twin technology.
In this fourth and final article of the series, they discuss two key aspects of digital twins—what they are versus common misconceptions and the concept of a digital twin framework.
Definition
Example of Digital Twin Technology
James: Sure. Let’s consider an example on replicating the degradation of the filament of a lightbulb. If you remember, this type of bulb gets bright before it burns out. If I’m monitoring the temperature of that light bulb, or maybe the brightness of that filament, and predicting when that bulb is going to break, the model is used in a digital twin to predict its failure. In this example, we wouldn’t just model the theoretical failure of the light bulb, but we’d be synchronizing that model with an actual light bulb.
Sam: So, it seems like a key aspect of digital twins is the synchronization between them and their real counterparts?
James: Yes, digital twins are synchronized with their real counterparts in a time critical fashion. And this is important to note; it’s not necessarily in real-time, but rather is time critical. For this light bulb example, maybe I need to synchronize my model with this bulb every second because I need to predict within 60 seconds of the light bulb going out.
However, moving to the semiconductor manufacturing environment, in the case of dispatching and scheduling, I just need to synchronize every time a new wafer shows up.
Sam: What about the level of confidence in predictions? In our last blog we talked about the importance of this; how does that come into play when talking about digital twins?
James: The key outputs of the digital twin are along the lines of a prediction or detection (such as something is broken or is going to break). But, yes, just like we talked about in the last segment it also must provide information about its accuracy. If a digital twin tells you that a light bulb filament will fail, it must indicate when it will fail. If it says 60 seconds or ±5 seconds with a probability of 95%, you can use that information to order a replacement bulb.
Framework
James: Right! For example, people don’t realize we’ve been using digital twins in run-to-run control in semiconductor manufacturing since the early 90s and now it’s pervasive. It takes a model of a particular piece of equipment and then tries to predict a recipe to improve the quality or throughput of that equipment. Run-to-run control is a form of model-based process control which also uses a digital twin of the process.
Predictive maintenance, which has now been around for 10 years, predicts some aspects or some failure mechanism such as in our earlier light bulb example. So, predictive maintenance also uses digital twin technology. Virtual metrology is another type of digital twin; a virtual metrology twin takes measurements from the tool, tries to predict metrology values and synchronizes with the real metrology tool to update the model.
So, as you point out, it’s important to emphasize that digital twins have been around for a very long time, and if we’re going to develop a digital twin framework for the industry, we have to have a framework to accommodate all these existing applications.
James: Essentially, we need to be able to both reuse digital twins and combine them.
First, I’ll discuss their reuse. Let’s say I’ve got a digital twin of an Applied etch tool that’s going to predict the machine’s throughput. I could develop a model that’s applicable to all etch tools that tells me what inputs are needed, but it won’t have high accuracy. If I refine that model for one Applied etch tool brand or a particular etch tool instance, the digital twin model will have more specificity. For example, it could determine what additional sensors are needed or the equation needs to be modified for the specific etch tool. This is what we call generalization hierarchy.
Combining digital twins is the other piece of the puzzle, which is particularly beneficial for scheduling and dispatch. Let’s say we have several existing digital twins: a scheduling and dispatch digital twin that is rule-based and tells us when to schedule different wafers at different tools, run-to-run control based digital twins indicating the quality of each tool that I’m sending wafers to, and a predictive maintenance digital twin that tells me when particular tools might fail in the future and with what probability. If I then had a scheduling dispatch digital twin that aggregates information from these twins on the quality of production of all my tools, when they’re going to fail and the probability of their failing, I can build a better scheduling and dispatch solution. That’s called the aggregation of digital twins.
Sam: Machine Learning and/or AI must benefit from Digital Twins – right?
James: Yes! Developing common interfaces, common ways in which these models can talk to each other, creates a playground for performing so many applications.
Digital Twins can take machine learning from different components and bring them together to create a much better system.
Conclusion
Digital twins are purpose-driven replicas of aspects of a system, such as processes, equipment, or products, that are synchronized with their real counterparts. They have been utilized for a long time in various applications. To develop a digital twin framework, it is essential to acknowledge and accommodate these existing applications while defining clear technical definitions for digital twins and the framework itself. We can leverage the power of digital twins to drive innovation and improve various applications – even AI.
About Dr. Moyne
Dr. James Moyne is an Associate Research Scientist at the University of Michigan. He specializes in improving decision-making by bringing more information into the scheduling and dispatching area. Dr. Moyne is experienced in prediction technologies such as predictive maintenance, model-based process control, virtual metrology, and yield prediction. He also focuses on smart manufacturing concepts like digital twin and analytics, working towards the implementation of smart manufacturing in the microelectronics industry.
Dr. Moyne actively supports advanced process control through his co-chairing efforts in industry associations and leadership roles, including the IMA-APC Council, the International Roadmap for Devices and Systems (IRDS) Factory Integration focus group, the SEMI Information and Control Standards committee, and the annual APC-SM Conference in the United States.
With his extensive experience and expertise, Dr. Moyne is highly regarded as a consultant for standards and technology. He has made significant contributions to the field of smart manufacturing, prediction, and big data technologies.
