Based in Oxford, UK
Digital transformation; Gradiant; FTD Solutions; Farnsworth Group;
Digital transformation: opportunities and challenges for the semiconductor industry
UPM spoke with three experts in the field about the benefits and difficulties of integrating artificial intelligence into semiconductor manufacturing facilities in the form of digital twins. Brandon Ekberg, Principal, Facilities Science and Technology at Farnsworth Group, Josh Best, Vice President of Innovation at FTD Solutions, and Hiep Le, Chief Technology Officer – Eastern Region at Gradiant gave their thoughts on how artificial intelligence will shape the future of the semiconductor industry.
The role of artificial intelligence (AI) and the digital transformation has rapidly made its way to the forefront of the global consciousness, and it plays a key role in semiconductor industry productivity. Many see machine learning and artificial intelligence as the ideal opportunity to resolve ongoing challenges for the semiconductor industry – enabling automation and alleviating workforce restraints, facilitating real-time monitoring, and avoiding expensive disruption to the production pipeline caused by equipment malfunctions. A digital twin provides a high-fidelity model of a physical object which can facilitate decision making by using real-world data to produce simulations of how said object will be affected by real-world conditions. However, as with any new technology, there are unavoidable risks which may impede uptake of AI tools like digital twins.
What role will the digital transformation play in the future of semiconductor manufacturing?
The implications of artificial intelligence for the future of semiconductor manufacturing are bringing a new buzz to the industry.
Hiep Le, Chief Technology Officer – Eastern Region at Gradiant proposes that artificial intelligence will be most valuable in semiconductor manufacturing for forecasting and predicting areas which see a lot of changes in operating conditions, and areas where predictive maintenance is valuable. Three key improvements that digital transformation can provide are: reliable operation of the plant, optimising energy consumption, and optimising consumables (such as chemicals, filter replacement parts and other maintenance materials). From an operational perspective, AI advances could be transformational for semiconductor facility efficiency where carbon and water footprints reduction could be achieved.
Brandon Ekberg, Principal, Facilities Science and Technology at Farnsworth Group, sees that the digital transformation will be useful for design and construction of semiconductor facilities. He highlights that the speed of technological advances can leave facilities under construction scrambling to update floorplans. The digital transformation could optimize the reviewing process and change the turnaround for approval of these changes from months to days. Moreover, Ekberg notes that digital modelling has huge implications for water balances: “with data modelling one can connect client changes at recipe and tool level so that it automatically changes the water balance accordingly.”
Josh Best, VP of Innovation at FTD Solutions, agrees that the first stage of adoption of digital tools will be in the design phase, which presents the most flexibility for testing whilst minimizing immediate risk. He expects they will turn into a day-to-day decision-making tool.
What sets digital twins apart?
A digital twin model of a facility’s mass flow balances for water is extremely valuable for construction and maintenance. It can be used to design axis clearances, space needed to operate a valve, moving and installing tanks, among other things.
One thing everyone agrees on is the vital importance of maintaining a digital twin in order to retain its value. Digital twins are as useful as they are accurate – in fact, the moment a digital twin becomes outdated, it ceases to be a true ‘twin’.
“Your plant is an ever-changing physical reality - system operating efficiencies increase and decrease. Digital twin maintenance is necessary so that the outputs are representative of the physical reality of the plant on a given day,” says Josh Best.
What are the current limitations of digital twins?
As much as machine learning and artificial intelligence provide key opportunities for growth, they must also be treated with caution. As a technology in its infancy, digital twin software has limitations which are impeding uptake.
A key concern is that of maintaining the integrity of intellectual property in the face of cybersecurity threats. Creating a digital twin model could leave facilities exposed to hacking, and so creating robust standards for cybersecurity is vital.
The time it takes to gather sufficient data to demonstrate the true value of the digital twin may be a challenge, particularly if semiconductor manufacturing companies do not take a chance in piloting the software.
How is best to tackle data quality control?
With so much data being processed by digital tools pertaining to water flow balances, Hiep Le notes that “data ingesting is an important aspect of AI; good data in means good predictions out, and bad data leads to bad predictions.” The perspective that high quality data is a crucial factor in the success of a digital twin, is a popular one among industry professionals. So how will the ‘good’ data be separated from the large quantities of data being transmitted?
Some believe that early adoption of AI will furnish semiconductor manufacturers with a tool to scrub outlier datasets and analyse trends to ensure that incoming data is representative of a water flow balance. Further to this, Hiep Le believes that experience and domain knowledge is key to facilitate swift identification of outliers. Brandon Ekberg, however, believes that rather than scrubbing, leveraging data is the answer – accessing data and making connections creatively is best enabled by keeping the full picture visible, and will make plants smarter in the long run.
How will the digital transformation be enacted in this industry?
Given the unknown variables of this new technology that remain, it is probable that uptake for AI in semiconductor manufacturing will be done with caution. Some early adopters will press on with AI and digital twins ahead of the curve, but the industry will likely take a more measured approach to adoption.
Any downtime in semiconductor manufacturing facilities can cost billions of dollars, which means an incremental approach is preferable to avoid costly interruptions. A phased approach can mitigate some risk, incorporating change management and balancing technological advancement with the needs of business.
The implications of large scale, innovative tools such as digital twins are an exciting prospect for the future of the semiconductor industry. With AI on an upwards growth trajectory, there will surely be a sweet spot between jumping the gun before risk management systems are in place and lagging behind out of an overabundance of caution which leaves companies playing catch-up.