Wafer fabrication does not welcome shutdowns. All it takes is a problem with one semiconductor tool, and a fab can come to a halt. For semiconductor manufacturing, in which every second counts, a halt in production can also be significantly costly.
The Internet of Things - commonly known as IoT - a network of connected objects embedded with sensors and other technologies, is making equipment failures easier to identify and act upon. In advanced manufacturing settings, IoT networks are laying the basis for predictive maintenance systems. Data harvested using IoT can also be the foundation for reducing chip production processing time, using artificial intelligence (AI) to set new benchmarks for efficiency and output. A 2021 study forecast that AI could make the semiconductor industry operational gains of $85-95 billion (£68-76 billion) per year by 2025.
In this article, we ask how IoT, combined with AI, is helping to optimise preventative maintenance and operating processes in semiconductor manufacturing.
IoT and AI in predictive maintenance
IoT networks in fabs are simplifying data collection that can feed into AI used for predictive maintenance. From power consumption to vibration levels; IoT networks are instrumental in gathering the evidence that AI can use to detect any operational issues such process deviations or equipment problems. This ability to monitor and update devices is key to predictive maintenance that can ultimately help to avoid the scenario of a shutdown.
A report from Deloitte states that predictive maintenance enhanced by AI can increase equipment uptime by around 10-20%. Other significant findings of the report include the potential time that can be saved (50%) in planning maintenance with AI, and the reduction of material spend (10%).
Jason Shields, Vice President of Equipment Intelligence at Lam Research, a provider of wafer processing solutions for chipmakers, is at the heart of the AI revolution within semiconductors. He describes AI as helping achieve “first-time-right goals in troubleshooting and service tasks” and optimise “tool fleet management… and predictive maintenance”.
IoT and AI in increasing operational efficiency
The evolution of ‘smart fabs’ hasn’t just translated into advantages in problem solving. Increasing production output is also possible thanks to the machine-to-machine and machine-to-human conversations that are enabled by IoT and AI.
Chip production processing times is one of the biggest challenges faced by the semiconductor industry, particularly in mind of the recent supply chain crisis. There is a pressing need to slash the time it takes from initial processing to final product. Efficiency also relies on cutting yield losses which can eat away at production costs. Thanks to the information gained by IoT, and analysis by AI applications embedded into the production cycle, operating processes are continually being refined and optimised.
Within semiconductor manufacturing, AI platforms are being used for; data harvesting, taking information ingested by IoT networks and performing pre-processing that addresses duplicates and outliers; developing AI models, training models with machine learning techniques such as deep learning; predictive analysis, predicting KPIs including cycle time, yield and defect rate; and reporting, with platforms being able to produce interactive displays of various metrics in real time.
The results of these implementations have been outstanding. Chipmakers have benefitted from AI-enabled improvements in yield, with the specific process steps that lead to losses being identified. There has been a reduction in cycle time, thanks optimisation of the sequencing and scheduling of tools, speeding up time-to-market.
Defects have been reduced, with causes such as contamination or tool wear being identified and rectified more quickly. Processes have become more controlled and stable, with KPI variability being reduced and product performance made more consistent. This is owing to the monitoring of parameters in fab processes that wasn’t previously possible.
The future of IoT and AI in semiconductors
Despite the leaps and bounds taken by the technology, there is room for improvement. The very nature of AI’s learning models means that processes will be refined, including the tightening of the IoT networks that gather data.
As expert Jason Shields of Lam Research concludes: “The industry must continue the transition from today’s loosely stitched boxes to nodes on a network where data flows back and forth, and algorithms from various parties contribute to achieving greater control at a lower cost of ownership.”
