中文 English

Hyperscaling Cyber-Physical Systems

Digital twins are being utilized across the semiconductor industry, from models of SoCs to the fabs that make them.

popularity

As consumers, we are not always aware of all the data created around us by our cars, mobile devices, computers, and day-to-day consumer products. In most cases, we are even less aware of what’s going on during the development and production of today’s devices and systems that we have gotten so used to. During the most recent DATE 2021 conference, the special day on cyber-physical systems gave some fascinating insights on digital twinning, AI/ML, and embedded development for automotive and industrial applications. It was rounded up by an exciting keynote by STMicroelectronics’s Philippe Magarshack and a panel on EDA’s role.

Prof. Enrico Macii from the Politecnico di Torino chaired the special day, and I was his co-chair to bring in the industry view. As an extension of traditional embedded systems, cyber-physical systems combine deeply intertwined physical and software components to build sensor-based, communication-enabled autonomous systems. They represent the key enabling technology in production systems for automotive and industrial applications, networks of consumer smart home and robotics applications, systems-of-systems applications for transportation, and autonomous driving.

Philippe Magarshack’s keynote was the highlight of that day. As ST’s group VP, strategy technology & systems architecture, we recruited him for his unique perspective on this topic. “life.augmented” is the tag-line within ST’s logo, indicating its impact on enabling every consumer’s day-to-day activities. ST also uses cyber-physical systems within their production.

Philippe identified three CPS-related target areas as a focus for ST: smart industry, industrial internet of things (IIoT) and 5G, and smart cities. For smart industry, cyber-physical systems improve operations’ agility and autonomy, make environments safer, optimize energy efficiency, and improve productivity. In the area of IIoT and 5G, Philippe described how 5G is accelerating IIoT’s pervasiveness, IIoT data’s impact on fueling cloud services, and smarter autonomous objects. End-to-end security and AI/ML processing at near and far edges will be instrumental here.

In the “smart cities” domain, cyber-physical systems enable more efficient city infrastructures and enhanced services like smart parking and e-bike rentals. Given the vast scale of cities, controlled LED lighting’s energy efficiency is also a critical topic. The requirements for sensing, actuating, and embedded processing for autonomous decisions while considering security and various connectivity types are quite diverse. The resulting product portfolio of 32-bit general-purpose MPUs and MCUs offers 17 different designs meeting the various needs for high and mainstream performance, ultra-low power, and wireless connectivity.

The results can be quite profound. For instance, in a sheet metal factory in Connecticut, USA, the productivity improvement was 17% to 20%. The asset tracking using centimeter-accuracy localization with UWB technology enables geofencing for areas that workers can access near equipment and improves safety with navigation, anti-collision, and traffic management. The data analytics optimize the allocation and location of workers, tools, and materials.

Philippe also described ST’s own usage of cyber-physical systems in their 300mm fab in Crolles, France. At 6,300 wafers per week, this fab generates 3TB of data per day used for shop floor optimization, advanced defect classification, optimization of equipment productivity, process control and optimization, managing yield ramps, and achieving design for manufacturability. They have applied their own chips (of course) to monitor wafer handling and local AI processing for vibration analysis on vacuum pumps, fans, and compressors. ST cut wafer scrap per month in half, and that just seems to be the beginning.


Three main digital twin application areas. (Source: Frank Schirrmeister, Cadence Design Systems)

Digital twins play a crucial role in cyber-physical systems as well. To me, they fall into three different classes: development of systems with a focus on building better products, process optimization of production, and predicting maintenance issues and life cycle effects for life-cycle management of a product. Digital twins of the SoCs allow for unlimited visibility and enable validation of counter-measure strategies and the assessment of reliability, safety, and aging. The illustration of how digital twins extend from the SoC to the factory level was quite impressive. ST’s fab features 150 vehicles that move 480 tons in 60,000 transports per day, driving on average 45km per day per vehicle. A digital twin of the fab allows for real-time optimization over millions of combinations and has a real-time GUI connecting the digital and physical twin for supervision.


Digital twins for SoCs, equipment, and fabs. (Source: Philippe Magarshack, STMicroelectronics)

ST is in good company in applying digital twins. DATE 2021 saw car-as-a-service optimization using digital twin models presented by Charles Steinmetz (Hamm-Lippstadt University of Applied Sciences) and a description of cognitive digital twins for manufacturing systems by UC Irvine‘s Mohammad Al Faruque. Sara Vinco (Politecnico di Torino) presented extensions of digital twins to optimize communications and energy consumption, and Thomas Markwirth (FhG ISST Fraunhofer-Institut für Software und Systemtechnik) illustrated results of dynamically introducing fault structures into digital twins without the need to change virtual prototype models.

Unsurprisingly AI/ML plays a key role as well. The DATE 2021 special day featured presentations on bin-picking of randomly filed 3D industrial parts (Sukhan Lee, Sungkyunkwan University) and AI/ML for defect detection in additive manufacturing for medical, aerospace, and automotive applications (Davide Cannizzaro, Politecnico di Torino). Zijie Ren, South China University of Technology, discussed how machine learning improves digital twins’ cognitive, reasoning, and decision-making abilities. Finally, Michael Huebner, Brandenburg University of Applied Sciences and his team presented how artificial neural networks automate performance-efficient realizations of mass spectrometry and nuclear magnetic resonance spectroscopy used in industrial chemical processes.

In the third session of the day, Prof. Harald Schenk delivered an embedded tutorial in which the use of microsensors and sensor nodes is practically demonstrated in the context of the “Innovation Campus Electronics and Micro Sensor Technologies Cottbus” (iCampμs). Prof. Rolf Ernst (Technische Universität Braunschweig) presented in the context of automotive and AUTOSAR on run-time environments for system-level “logical execution time” (LET) programming. Birgit Vogel-Heuser (Technical University of Munich) closed this session with a presentation on managing variability and reuse of control software in cyber-physical production systems, focusing on the challenges in documenting the dependencies of software parts and their variability using family models.

Cyber-physical systems are a vital part of our increasingly hyper-scaled world. The volume of data they produce, transmit, and process further drives the need for computational software and domain-specific semiconductor components.

Brave new world, here we come!



Leave a Reply


(Note: This name will be displayed publicly)