AI and the big data revolution have already made a huge impact in sectors such as retail, healthcare, education, and autonomous driving.
Now, manufacturing is experiencing rapid transformation thanks to a combination of AI and advanced technologies like the Internet of Things.
The transition has been so significant that it has been deemed the Fourth Industrial Revolution; commonly referred to as Industry 4.0. The result is highly digitized and connected smart factories that are able to run almost autonomously and, thanks to AI, have the capability to self-correct.
Machine learning algorithms and applications are giving manufacturers access to disruptive new business models, allowing them to optimize their operations and improve product quality, all while reducing costs.
Proactive businesses are getting ready for the change in their competitive landscape and embracing machine learning, but is it really worth it?
Here, we look at a case study which shows the dramatic effect that machine learning is having on manufacturing, and more specifically on yield improvement in the semiconductor industry.
The problem of yield losses in the semiconductor industry
In complex manufacturing environments, such as those found in the semiconductor industry, productions processes typically include several stages and can span many weeks to months. Yield losses, where products need to be reworked or scrapped because of defects, have a significant impact on production-run times and profitability. A report by McKinsey shows that in the industry, it is common that:
The production cycle from the first processing of the wafer to finished chip includes a range of intermediate quality control tests.
Yield losses and testing costs can comprise as much as 20 to 30 percent of the total production costs.
Data is highly available, but it is not systematically analysed, and data sources across tool groups are not linked.
The role of machine learning in providing the solution
The high degree of automation and advanced production equipment used in semiconductor fabs means that detailed production data spanning years is potentially available. Using AI engines to link quality control and yield data with process control data gives semiconductor manufacturers the foundation for identifying yield losses, as well as homing in on their root causes.
This capability allows manufacturers to adjust production processes and chip designs in order to avoid problems, often using applications to monitor and make adjustments to subprocesses in real time.
Specialized analytics companies are rising to the challenge faced by semiconductor manufacturers and harnessing the power of data to effect yield improvements:
Qualicent Analytics, for example, is building AI engines which can determine the optimal process or product operating conditions to drastically reduce manufacturing defects.
Motivo offers systems which can anticipate the location of yield detractors based on both supervised and unsupervised machine learning, in order to optimize chip design.
AI-based Algorithms are used to break down known problems into key components, which can then be identified in new and existing designs to uncover potential problems in individual microchips.
The results of employing AI
The impact of using AI in the semiconductor industry has been huge. Overall, yield detraction has been reduced by up to 30 percent, and the cost reduction benefits of using AI-based algorithms for testing cover a variety of areas:
The ability of AI to identify root causes can reduce scrap rates which improves yield.
AI can improve the overall effectiveness of equipment by lessening the requirement for equipment and maintenance.
The cost of test procedures is reduced when they are AI-optimized.
A reduction or stabilization in the flow factor can lead to higher throughput.
This case study shows a glimpse of the endless power of machine learning and what is possible for businesses. In the semiconductor industry alone, manufacturing yields have increased by up to 30 percent, scrap rates have been reduced, and fab operations have been optimized.
However, this is just a snippet of the enormous potential for machine learning to revolutionize manufacturing. Amongst a host of other use cases, predictive maintenance of industrial equipment using AI is one of the fastest growing niches in manufacturing.
Deloitte reports that AI-enhanced predictive maintenance can improve equipment uptime with increases of around 10-20 percent, reduce the time spent planning maintenance by up to 50 percent, and reduce material spend by 10 percent. This would have the effect of reducing overall maintenance costs by 10 percent; a potential saving of millions of dollars for large companies.
AI is already changing the way we do business, and its importance is set to grow in the coming years. The effects of machine learning will be far reaching, not only in the semiconductor industry, but in transforming manufacturing as a whole. Companies of all sizes should be prepared. Embracing Industry 4.0 is going to become increasingly critical as businesses try to hold on to their competitive advantage and improve their bottom line.