Materials Intelligence: Harnessing Data and Digital Innovations to Redefine Material Science – EE Times
Microchips have been hailed by both technology and financial sectors as the new oil — the scarce, vital resource underpinning our modern world — because of their fundamental role in economies and their impact on key industries, innovation, and global competitiveness.
While much has been written about the groundbreaking innovations of semiconductor companies like Nvidia, AMD, and Intel, as well as electronic design automation (EDA) and fabrication technologies, little attention has been given to the increasingly complex, high-purity materials and chemicals essential for the manufacture of semiconductors.
A select few companies provide the specialty materials, gases and equipment that are crucial to the intricate and time-consuming process of manufacturing most of the world’s semiconductors. These materials are critical components in the fabrication of integrated circuits and chips, which are the heart of virtually all electronic devices, from smartphones and computers to advanced automotive systems and more.
The future of digital technologies and high-performance computing hinges on the continuous availability of high-quality computer chips. Without these chips, the digital world stalls, and without advanced materials, these essential chips cannot be produced.
As the semiconductor industry progresses, we envision boundless opportunities. Recognizing the pivotal role of materials in fabricating next-generation semiconductors, we are positioned to unlock new frontiers through data, elevating performance and efficiency, and ushering in an era of unparalleled innovation in semiconductor technology.
We call this “Materials Intelligence,” and it involves not only the scientific understanding and engineering of materials at the atomic and molecular level but also the integration of digital technologies to optimize material properties, performance, and manufacturing processes. It is the systematic analysis of vast amounts of data related to material characteristics, production processes, and performance outcomes.
By applying AI and machine learning (ML) algorithms to this data, the semiconductor industry can predict material behaviors under various conditions, identify optimal material compositions for specific applications, and enhance manufacturing efficiencies. It is the unique capability to make the right material with the right quality, at the right place and time. Ultimately, this enables us to innovate and improve the materials used in semiconductor manufacturing and other electronic applications.
The potential of analyzing vast amounts of data goes beyond company-internal work and there are immense opportunities in data-sharing-based improvement collaborations across the electronics industry.
For example, through co-optimization projects with device makers, our quality teams are continuously pushing the boundaries with our customers to quickly resolve undesirable performance drifts, increase yield and reliability, and ultimately enable our customers’ journeys towards contamination-free manufacturing. Using ML techniques, a new level of excursion-avoidance can be reached by moving from analyzing and controlling tens of parameters on a certificate of analysis (CofA) to monitoring thousands of parameters from the supplier and fab including their interaction effects and identifying the ones that truly matter to performance.
The possibilities for digitalization, the Internet of Things, and AI are influenced not only by the availability of materials but also by how we harness their potential.
To advance AI innovation, we must develop computer chips and semiconductors that fulfill significantly increased computing power and storage capacity requirements.
Increasingly powerful chips are therefore needed for the next phase of AI. After all, we have been rapidly approaching the fundamental limits of physical possibilities. Making chips even more powerful, while further shrinking their structures, is becoming increasingly difficult. For this reason, various approaches are being employed simultaneously. With the evolution of electronics and the implementation of AI, advanced chips will need more advanced materials on the front end and more capacity of materials at the back end of the wafer.
Multi-die technology, also known as chiplet-based design, involves integrating multiple small chip components, or chiplets, into a single chip. This method enhances component efficiency and substantially boosts performance. Another method is the implementation of vertical structures instead of making flat structures smaller, similar to high-rise buildings. This technology has been very successful in recent years, especially for flash memory (NAND). These advances make it possible to further increase chip performance while reducing cost and energy consumption.
These technological approaches share a common requirement: new materials that have not previously been used in chip production. The new 3D structures necessitate a different layering of materials, shifting from horizontal layers to vertical structures.
In addition, the properties of many commonly used materials change dramatically when shrunk down further. For example, copper does not conduct electricity well at just a few nanometers in size. Mechanical and thermal properties are also becoming increasingly important, as modern chips produce more heat on their surface than a stovetop. Dissipating heat is increasingly challenging with layered structures. Therefore, developing new materials that better meet these requirements is crucial for the chip industry. The challenge of discovering new materials is immense, as it involves exploring numerous combinations of various elements into diverse three-dimensional structures. However, new tools running on today’s chips could help revolutionize the chips of tomorrow.
To envision the future, we can look at the prediction of crystal structures, a challenging computational problem. In November 2023, Google’s AI lab, DeepMind, introduced a new AI tool called GNoME (Graph Networks for Materials Exploration), which predicted 2.2 million new crystals. This tool could eventually expand to other material classes and predict new theoretical substances.
According to the GNoME calculations, 380,000 of the new crystals should be stable materials that could power future technologies. That said, these new substances exist only in theory, we don’t know their properties, or if they can be produced in large quantities.
Before GNoME and new data and digital technologies, discovering new materials was extremely time-consuming and expensive. Some materials scientists have compared their previous experiments to the iterative and laborious work of Thomas Edison, who tested more than 3,000 different materials to find the ideal filament for the first commercially viable electric light bulb. This “trial and error” method was time-consuming, prone to errors, and heavily dependent on the individual experience. Moreover, there was little connection from one experiment to the next.
By harnessing large volumes of data, today we can gain valuable insights and identify patterns that help accelerate the discovery and development of new materials. Advanced analytics and machine learning techniques enable us to analyze complex data sets and predict material properties and performance, guiding our research efforts. Automation and robotics in data collection and analysis streamline R&D workflows, freeing up researchers’ time for more strategic and innovative tasks.
For instance, through multivariate analysis, our R&D team recently uncovered unique interactions that were previously unconsidered and predicted their impact on material performance. Understanding the effect of multi-component interactions on output parameters and predicting the performance of millions of virtual formulations enabled us to optimize the concentration of key components, expediting the selection of the best-performing formulation that precisely meets our customers’ requirements. The acceleration of learning cycles by more than 30%, combined with the ability to quickly generate new formulations in response to updated targets, enables us to address undesirable characteristics, improve cost efficiency, and accelerate our customers’ time-to-market.
As the semiconductor industry advances, our vision of boundless opportunities becomes a reality through “Materials Intelligence.” This approach harnesses the scientific and digital integration of materials at the atomic level, leveraging vast data to optimize properties, performance, and manufacturing processes, to usher in a new era of innovation in semiconductor technology. Join us as we embark on this transformative journey, setting new standards and shaping the future of the industry.
Merck is a leading science and technology company that uniquely combines three specialized, innovation-driven businesses: Life Science, Healthcare, and Electronics. Using our multi-industry expertise, we help to address a broad range of global challenges. From providing products for scientific research and biotech production to therapies against serious diseases to high-tech materials and solutions for the semiconductor industry and displays – Merck works to make a positive impact on millions of people’s lives every day. Merck’s Electronics business operates as EMD Electronics in the United States and Canada
You must Register or Login to post a comment.
This site uses Akismet to reduce spam. Learn how your comment data is processed.
Advertisement
