By the year 2030, 78% of economic growth will be driven by advancements in materials. Simultaneously, developing new materials is becoming more expensive as they become more complex. While digital materials development and testing holds promise to reduce the time to develop new materials from 20 to 5 years, digital material models are often too expensive or inaccurate to be used. PHIN's mission is to develop accurate and cost-effective materials models to enable digital materials development.
Over the past 5 years, AI has decreased the time to develop material models from multiple years to a few months. Machine learning models, however, introduce a new issue: their accuracy cannot be trusted. While the accuracy can be continually increased by expanding the dataset to create a materials foundation model, this approach does not improve trust in the models and will still require extensive validation for each new material it is used on. PHIN has developed trustable machine learning models that co-predict their accuracy. With this approach our software can automatically generate accurate and cost effective models for any material to predict its properties. We have used our models to simulate surface decomposition, catalytic activity, and mass transport and are harnessing the decades of simulation protocol development to predict many more properties,. With PHIN, scientists are able to develop their materials digitally, no longer do they need to worry about the accuracy of the models or wait for results. Instead, they can focus on developing materials.
Why Materials Development
Over the last century, materials science and engineering has matured to become a cornerstone in the development of modern technology. Nearly all products on the market today rely in some form on materials advancements – stronger alloys, cheaper manufacturing techniques, or simply novel materials with functional properties that enable new kinds of devices. Computers and batteries, products most of us use on a daily basis, would not exist without deep efforts to understand the fundamental nature of material behavior, and the corresponding application of this knowledge to design and optimize material properties. Moore’s law, the doubling of transistor density every two years, would not hold true without continued progress in silicon processing technology. Similarly, electrolytes and cathode developments have improved lithium ion battery safety and energy density to facilitate greener transportation with consumer electric vehicles. These outcomes have highlighted the importance of materials development, which is increasingly becoming critical across every industrial vertical, including metals, polymers, and composites. Technological progress will come hand-in-hand with materials advancements – those who can capitalize on the growing need for materials development will find great success in enabling tomorrow's technology.
Climate change is driving the development of a new host of products, and consequently, new materials development. Although the materials properties that will be needed have been forecasted to address these outside forces, the decades-long timelines of materials development will struggle to deliver materials that fulfill those requirements. In order to prevent materials from becoming the limiting factor in product development, materials development timelines need to be reduced from decades to align with product development timelines of five years.
At PHIN Materials, our experience in developing battery materials has given us a first hand look at the many new developments that aim to provide the energy storage requirements for addressing climate change. There are innovations in the anodes with silicon and lithium, innovations in the electrolyte with ceramic, sulfide, and polymers, innovations in iron phosphate and conversion cathodes, and innovations in charge carriers with sodium (Na), zinc (Zn), magnesium (Mg) and potassium (K). With battery material systems maturing, we can observe an exponential growth of materials technologies that are needed to diversify and optimize end products. Many of these technologies have already been in development for 10+ years and yet still need significantly more time before they are in mass production.
This issue is compounded by the growing complexity of materials systems. Unfortunately, running more experiments with more scientists cannot scale to address the escalating complexity and has driven a sharp rise in the number of scientists required to make the same progress. For example, Moore’s law has been slowing down despite increasing investment. Today it takes 18x more scientists to achieve the same gains in transistor density than were required in the 1970’s. A similar analog can be seen with lithium-ion batteries and will only become more apparent across industries as increasingly complex materials become the bottleneck. Each industry will encounter their own slowing of Moore’s law and require ever increasing resources to achieve the same performance improvements.
We started Physics Inverted Materials to address the exponentially rising avenues of materials development and complexity of these materials. Without a fundamental change to the materials development paradigm, even massive increases in funding will not be sufficient to address the challenges in materials development.
Why Digital Materials Development
Digitization has been key to overcoming design complexity in engineering, but digital tools have been hard to adopt in materials development. In the product design and electronics fields, the proliferating improvement of computing since the 1950's has created digital tools for product and circuit design such as computer aided design (CAD) and electronic design automation (EDA). These tools have improved worker productivity and reduced the cost and time for developing new and complex products by enabling rapid and low-cost prototyping, automated testing of hypotheses, and creating digital blueprints for manufacturing and communication.
The use of digital tools in materials development requires two components: (1) a digital testing procedure and (2) a model of the material. While a large library of established digital testing procedures has been developed over the past half century, the availability of accurate and low-cost models is very limited. The models needed to predict performance is where digital materials development differs from product design or electronics. The most complex products contain millions of components and circuits contain billions of transistors. Materials by contrast contain Avogadro’s number (a trillion trillion) of atoms. Since it is computationally impossible to model this many particles, materials models must be segmented by scale, where each scale only predicts properties for a million to a billion particles. The need for a model not just for each material but also for each scale adds further complexity. And since it takes a team of experts years to develop just one of these models, the coverage of materials models is very sparse, significantly limiting digital materials development.
Where accurate models do exist, digital tools have been shown to accelerate materials development. Density Functional Theory (DFT) has allowed researchers to quickly identify new candidate materials. Classical interatomic potentials have allowed researchers to find mechanical and chemical properties of materials. Kinetic and reaction diffusion models are used in process engineering to optimize manufacturing methods. Device models combine a number of materials properties together to predict the performance of a system of materials. However, despite the success of these methods, these digital tools required tremendous effort and cost to develop and implement, making their use limited to a select few materials.
We started Physics Inverted Materials to provide machine learning models that are accurate and cost-effective to materials scientists. We can generate machine learning models in a matter of months, allowing them to focus on developing new materials while leveraging decades of computational materials science knowledge.