Sustainability in the Semiconductor Industry: The Power of AI and Machine Learning
As a society, we have become inseparably dependent on semiconductors, yet the challenges associated with their production are formidable. With the current pandemic exacerbating already-strained supply chains, the ability to meet rising demand for these essential components is uncertain. Furthermore, the environmental toll of semiconductor production, including the high levels of CO2 emissions, energy consumption, and water usage, is a critical concern.
To address these issues, our team has developed an innovative AI/ML-based solution that facilitates early identification of production failures by analysing and classifying spatial wafer profiles. We believe that this technology represents a crucial step forward in making the semiconductor industry more sustainable and resilient.
As we move towards 2030, it is projected that a staggering 20% of global energy consumption will be attributed to the ICT sector, with the semiconductor industry playing a significant role. While semiconductor industry has surpassed the automobile industry as larger CO2 polluter, the negative environmental impact of semiconductor production, such as excessive water consumption, has become a pressing issue.
Semiconductor plants are usually called fabs, and it is essential to recognise that a significant portion of the energy consumption within these fabs is allocated towards the maintenance and infrastructure needs such as air filtration, ventilation and water treatment , accounting as large as 50-60% of the electricity consumption. The remainder is dedicated towards actual manufacturing of the chips. The total energy consumption of a single fab can reach up to 100 MWh per hour, emphasising the significance of implementing energy-efficient strategies and sustainable practices within the sector.
The sustainability problem created by semiconductor fabrication is manyfold. The advancement of each new generation of microchips requires increased amounts of energy and water, thereby exacerbating the already substantial carbon footprint of the sector. This problem is exemplified by the fact that a significant portion (1/3) of the carbon dioxide emissions generated by ubiquitous devices such as the Apple iPhone can be attributed to microchip production.
The current global landscape is facing a significant challenge in the form of semiconductor shortage. Long-predicted to occur due to the ever-increasing demands of an increasingly digitised world, the global pandemic has served to exacerbate this issue exponentially. As such, the industry must now concurrently address both the need to increase production capabilities and the imperative to implement more sustainable manufacturing processes in order to overcome this dual-faceted obstacle and ensure a viable future.
There exists a prevalent skepticism surrounding the potential for the semiconductor industry to attain a level of sustainability that is both meaningful and feasible. However, such doubts fail to take into account the vital role that semiconductors play in enabling the production of the very technologies that hold the key to mitigating the impending climate crisis, such as solar panels. Furthermore, semiconductors are indispensable components of virtually every electronic device that is essential to our modern way of life, underlining the undeniably crucial nature of the industry. In light of these facts, it is clear that a sustainable future for the semiconductor industry is not only necessary, but entirely achievable.
As the world progresses towards Industry 5.0, we find ourselves on the brink of a revolutionary breakthrough in the form of AI/ML powered Statistical Process Control (SPC) systems. These powerful systems, enabled by the integration of cutting-edge Artificial Intelligence and Machine Learning technologies, have the potential to dramatically increase yield and reduce scrap rates within fabs. This is an opportune moment for industry players to harness these technological advancements and steer the industry towards a sustainable and profitable future.
Navigating Deviation: Challenges and Opportunities
The key to revolutionising semiconductor manufacturing and promoting sustainability lies in Integrated Circuit (IC) production on silicon wafers. These wafers, comprised of thin slices of semiconductor material, serve as the foundation for the microelectronic devices that drive our modern technological landscape.
The complexity and intricacies of Integrated Circuit (IC) manufacturing, encompassing over a hundred process steps and spanning multiple months, generates a vast trove of data. This data serves as a powerful tool for identifying production inefficiencies and maximising the yield of each individual wafer. Harnessing the insights derived from this data enables us to not only improve production efficiency but also to stay ahead of the curve in an ever-evolving industry.
Wafer mapping represents a vital method for capturing and visualising the spatial characteristics of wafer data. The process involves taking precise measurements of key surface points across a wafer, resulting in a map that illuminates the uniformity and deviation of the surface from ideal specifications. While traditional approaches to wafer profile analysis have relied on summary statistics and sampling measurements, wafer mapping offers a more holistic and nuanced understanding of wafer quality, enabling us to push the boundaries of performance and efficiency in semiconductor manufacturing.
Wafer mapping, while an integral aspect of Statistical Process Control (SPC), is not without its limitations. The use of summary statistics inherently leads to a loss of granular information, and the inability to distinguish different shape characteristics such as convexity and concavity, or identify violations of radial symmetry in wafer profiles, are notable limitations.
Mapping is also highly specific to individual product-types due to the different location and dimensions of the chips being manufactured. As the measurement coordinates vary the identification of location, shape and sharpness of the variation patterns becomes more difficult and the comparison of profiles across the different product types is prevented.
Due to this information loss, the traditional methods of monitoring and ensuring the stability of the semiconductor manufacturing process, such as control charts such as EWMA charts and Shewhart’s charts, which focus on monitoring process variables and attributes within established limits, are no longer sufficient in the ever-evolving and highly competitive semiconductor industry. These techniques provide limited insights and do not effectively contribute to process optimisation and enhancement. The industry demands a more sophisticated and data-driven approach, leveraging cutting-edge techniques and advanced analytics to gain deeper insights into the production process and stay ahead of the curve.
This is where the integration of advanced Artificial Intelligence (AI) and Machine Learning (ML) solutions come into play. By leveraging these cutting-edge technologies, it is possible to gain a holistic understanding of spatial profiles across the entire product portfolio, and use this insight to dynamically adjust production in the event of a defective or anomalous product detection. AI/ML-driven approaches enable improved level of pattern and shape recognition in complex data formats such as time-series data or images, unlocking insights beyond the capabilities of current systems and paving the way for next-generation process optimisation and quality control in the semiconductor industry.
Advancing Beyond the Basic SPC Framework
Our cutting-edge AI/ML solution, the Critical Profile Detector, represents a paradigm shift in the field of semiconductor manufacturing. By integrating advanced real-time online AI analysis into a Statistical Process Control (SPC) framework, we are able to empower intelligent decision-making and automate corrective actions, driving productivity to new heights. This innovative solution represents the future of process optimisation and quality control in the semiconductor industry.
The Critical Profile Detector is at the forefront of Industry 5.0, embodying the principle of human-machine collaboration within a digital ecosystem. By leveraging the power of AI and ML, it enables process optimisation without human intervention, while empowering human operators with smart solutions, ultimately maximising efficiency, productivity, and sustainability. This solution represents a major step forward in the field of semiconductor manufacturing and provides a significant competitive advantage over other semiconductor fabs still reliant on basic SPC solutions.
Unlocking the full potential of this solution requires a holistic approach that encompasses the pillars of explainability, interactivity, and intelligent action.
Explainability, enabled by advanced Machine Learning algorithms, enables the analysis of data flow to detect deviations and anomalies that may otherwise go unnoticed by human observers. The results of this analysis are presented in a clear and concise manner, allowing for easy interpretation.
Interactivity empowers process experts to make informed decisions by providing them with the ability to evaluate the significance of the anomalies identified by the AI model data. Furthermore, intelligent actions are derived from the capabilities of AI to analyse historical incidents and predict potential future occurrences, thus bridging the gap between expert insights and real-time process correction.
Leveraging advanced automation techniques, such as the Out-of-Control Action Plan (OCAP) process, seamlessly integrated with real-time AI/ML analysis, can further optimise the process.
The AI-assisted analysis encompasses a comprehensive examination of all process data, including numerical information and operational components such as equipment, through the use of multivariate discriminant analysis. This allows the AI to identify the key drivers of out-of-control actions, effectively pinpointing the root cause of any issues.
As the AI continuously learns and adapts to the intricacies of the process, the OCAP processes will become increasingly accurate and efficient, leading to a more streamlined and optimised workflow.
AI: The Key to Improved Productivity
Our cutting-edge SPC framework for Critical Profile Detection, LineWorks SPACE, coupled with its advanced add-on, eCAP (Electronic Corrective Action Plan) enables powerful AI/ML solutions. LineWorks SPACE serves as the foundation, providing a comprehensive data source for all measured data including product quality and equipment sensor data, within a robust ecosystem of chart plug-ins and add-ons.
The SPACE eCAP add-on represents the vanguard of AI/ML integration in wafer production, enabling the development of sophisticated machine learning models that can drive significant gains in productivity through the implementation of automatic corrective action plans.
By leveraging advanced AI/ML algorithms, this innovative solution provides real-time monitoring and analysis of the entire wafer production process, including anomaly detection, prediction of system performance, process deviation detection and more. Furthermore, it covers entire manufacturing chain with different product lines. With eCAP, the system is able to proactively respond to observed deviations, implementing intelligent corrective actions that ensure optimal performance and efficiency, at all times.
The power of AI/ML lies in its ability to delve deep into the underlying causes of process glitches, identifying the most effective corrective actions and equipping humans with the insights they need to make informed decisions. This proactive approach enables organisations to stay ahead of potential issues, preventing them from escalating and minimizing the impact on overall performance. This capability not only drives efficiency but also enables the organisation to be nimble, and to be able to take advantage of opportunities to optimise and improve the process.
Through the application of advanced Machine Learning techniques, our AI system is able to gain an in-depth understanding of the typical actions and decisions applied to wafers with similar critical defective profiles. Utilising this knowledge, the system is able to initiate eCAP processes automatically in the event of an anomaly detection, executing the same set of actions that have been successful in the past.
In cases where multiple options for corrective actions are available, the system can request human input to determine the most appropriate course of action. This integration of human decision-making allows organizations to strike a balance between the efficiency and adaptability, and to make sure that the appropriate corrective actions are taken every time.
The integration of AI/ML solutions for Critical Profile Detection represents a paradigm shift in semiconductor production, enabling significant enhancements in yield and sustainability. As global demand for microchips continues to rise, the need for advanced solutions like this becomes increasingly pressing in the pursuit of a sustainable future and effective industry-wide efforts to combat the climate crisis.
By leveraging cutting-edge technology, this approach can greatly improve the efficiency and sustainability of production processes in fabs, ultimately allowing organisations to stay competitive in today’s fast-paced and constantly evolving business environment. It can also serves as an important step in building a greener and sustainable future for the semiconductor industry.
Kalle Ylä-Jarkko is working as a Senior Data Scientist in Elisa IndustrIQ and is developing AI/ML models for camLine’s flagship LineWorks SPACE for complex operational systems in manufacturing, telco, and semiconductor industries. Kalle has a long back-ground in start-ups in laser industry and he holds 6 patents and is the author of over 30 papers in the field of laser technology, laser machining, machine learning, and AI applications.
Alan Crawford, Ian King, Debby Wu: The Chip Industry Has a Problem With Its Giant Carbon Footprint (Bloomberg, 2020) – https://www.bloomberg.com/news/articles/2021-04-08/the-chip-industry-has-a-problem-with-its-giant-carbon-footprint
Udit Gupta, Young Geun Kim, Sylvia Lee, Jordan Tse, Hsien-Hsin S. Lee, Gu-Yeon Wei, David Brooks, Carole-Jean Wu: Chasing Carbon: The Elusive Environmental Footprint of Computing (2020)
Bryan Ng, Nils Knoblauch, Kalle Ylä-Jarkko, Rasmus Heikkilä: Machine Learning in an SPC Framework Drives Productivity (2022)
Future-Fit Heatmap: Semiconductor Manufacturing (Future-Fit, 2020) https://futurefitbusiness.org/wp-content/uploads/2020/09/Semiconductor-Manufacturing-Future-Fit-Heatmap.pdf