Sức mạnh của AI trong việc chuyển đổi thiết kế và sản xuất chất bán dẫn

Artificial intelligence and machine learning (AI/ML) have immense power to transform semiconductor design and manufacturing for a variety of broad and far-ranging applications. Just consider the volume of data generated by design and manufacturing each year. With increasingly complex products, machines, processes and supply chains, the overall amount of data associated with semiconductor making is exploding.

Great opportunities arise at the intersection of AI and manufacturing execution systems (MES) data, sensor-based fault detection and diagnostics. The integration of physics-based models with a digital twin of factory floors, equipment and end-to-end processes of the supply or risk prediction at various points of failure are also opportunities for new AI-based solutions.

So far, the potential of AI/ML and data-driven innovation has been limited by secure, scalable data sharing challenges in either offline or real-time verification, validation and uncertainty quantification (VVUQ) capabilities, hindering the ability to adapt to changing conditions with embedded AI intelligence.

The semiconductor industry’s smart manufacturing of 2024 will be different as sophisticated and highly advanced AI tools, including GenAI, will be available to analyze large datasets and offer new insights.

While semiconductor manufacturing and test generate massive amounts of data, traditional analytics methods focus on a small subset of that data. The right data infrastructure was not always available to handle the volume of data created and analytics models were developed based on a small number of parameters. That meant insufficient parameter coverage, and predictive modelling and insights were not able to consider all available data. An AI-based approach will be able to consider all available data to determine which parameters are the most important to derive meaningful insights.

Another limitation of traditional analytics methods is the focus on one single domain for design, manufacturing or test. AI’s ability to handle much larger data sets offers the opportunity to look at a larger data space and address questions over multiple domains.

What sets AI apart is its ability to create a unique and superior model. While many AI efforts focus on finding “better models,” they stay at the pilot stage and are never deployed at scale. A few important elements are needed to make AI impactful in a large-scale semiconductor context. The first is a data infrastructure that can bring together large amounts of high-velocity data from design, manufacturing and test. The second is an extensive semantic model to pull together and align data from different process domains. The ability to develop or leverage the best models, including models developed by third-party teams and organizations, is third.

And finally, to be effective, AI must support not only the development of models, but also the ability to deploy and monitor these models where they will be used for decision making and process control, often in real time. For example, unless AI models for test can be deployed where tests are performed, they will not contribute to improving device quality in the globally distributed supply chain of complex 3D hybrid systems. This overall AI approach and infrastructure combining data, semantic, model development and deployment and life-cycle management is known as ModelsOps.

Massive investments are going into the creation of new semiconductor facilities. These new fabs will often be smaller than the largest and most productive existing fabs. AI will be critical in accelerating learning and integrating processes to make advanced fabs as productive as the established and larger fabs.

As the semiconductor industry faces a talent challenge as the existing employee population ages, the need for new talent is outpacing graduation rates. AI will be critical in “democratizing” analytical decision making and allowing a larger set of employees to use the best AI models to make optimal design and manufacturing decisions without data analytics experts.

PDF Solutions and other industry experts will discuss tangible applications of AI in the Semiconductor Industry at the AI Executive Conference Thursday, December 12, in San Francisco. Registration is open.

John Kibarian

  (all posts)

John Kibarian is president, CEO, and Co-Founder of PDF Solutions. He has served as president since 1991, and CEO since 2000. Kibarian received a Bachelor of Science degree in Electrical Engineering, a Master of Science and PhD degrees in Engineering Computer Science from Carnegie Mellon University.

source

Facebook Comments Box

Trả lời

Email của bạn sẽ không được hiển thị công khai. Các trường bắt buộc được đánh dấu *