What is ICME?
Integrated Computational Materials Engineering (ICME) is a holistic approach that aims to link materials modeling at different length and time scales to design products, the materials that comprise them, and the associated materials processing methods. A primary focus of ICME is placed on the materials and the process-structure-properties-performance relationship (PSPP), i.e. understanding how the processes produce material structures, how those structures give rise to material properties, and how do those properties produce certain performance, to enable better selection and development of materials for a given application.
What is the need for ICME?
A single computational modeling tool is incapable of capturing all the information that is needed to understand the PSPP relationship. First principle-based tools like Density Functional Theory (DFT) and Ab-initio Molecular Dynamics (AIMD) are only able to model systems as small as a couple of 100 atoms. On the other hand, classical tools like Molecular Dynamics (MD) or Finite Element Methods (FEM), while they cancan model larger systems (> 100,000 atoms) and capture long-range properties, they require some empirical description of the material system derived experimentally. Other tools like CALPHAD are also used to model the thermodynamics of a material system that is useful in evaluating the processing conditions but are heavily dependent on empirical data as well.
ICME aims to link the processes described above in a way that the results from one process can be used in the next one, the sequence of which is decided based on the length and time-scales regime of each modeling tool. This results in an understanding of PSPP relationship which helps researchers make a more informed decision about which material systems to further investigate through experimentation, as well as the processing conditions needed to produce the desired structure-property effect.
Phaseshift’s take on ICME with the addition of Machine Learning
Phaseshift has combined a more general ICME approach with a precursor Machine Learning (ML) system to further aid the material selection process. Generally, a material system is chosen for further investigation based on a previous understanding of it and its constituent elements. Since it becomes impossible to do an exhaustive search of a large design space through manual iteration, Phaseshift uses Machine Learning models trained on data available through literature and previous in-house developments to aid the material selection process. Not only does it speed up material selection, but ML is also able to identify completely novel material systems that have not been previously reported in the literature.
Suggested material systems from ML are subsequently further investigated through DFT. Based on the principles of quantum mechanics, DFT is one of the most versatile tools to investigate material properties through the use of a functional that describes the spatially dependent electron density of a system. The only limitation is the computational resources it requires to effectively simulate a complete system, which in turn limits the size of the system to a couple of 100 atoms. This is not sufficient to capture material characteristics that are observed over a longer range. For that, tools such as classical MD are used that are capable of handling over 100,000 atoms. However, MD requires a descriptor, known as Interatomic Potential (IP), specific to each material system and specific temperature ranges. Developing IP is an intensive process that requires experimental data, and therefore it is not available for every material system that could be investigated. To address this, Phaseshift has developed a process to develop IP from DFT data using ML. This rids the need for any empirical data to simulate material systems through MD, enabling one to get a glimpse into the material properties of a completely unique material system.
Phaseshift also uses thermodynamic modeling through CALPHAD based assessment to identify ideal processing conditions for the material systems and further establish the processing-structure-property relationship. The information about the structure-property relationship can then be fed into FEM models to evaluate the use of a particular material system for a specific use-case. The incorporation of all the models described here provides a complete understanding of the PSPP relationship of a material system, as is the goal of ICME.
Learn more about Phaseshift’s framework in this blog post.