Computational tools have unlocked entirely new ways to construct, simulate, and test new materials for commercial use – evolving rapidly over the last few decades to yield new ways to model materials and processes. The successful creation and commercialization of metal-organic frameworks, or MOFs, would not be possible without such computational modeling.

At NuMat, we are using state-of-the-art computational methods to transform images into insights. Atomistic modeling and material informatics can rapidly predict material performance, enabling us to design new MOF structures before performing a single experiment. Our computational workflow includes:

  • Discovering new materials through high-powered atomistic simulations;
  • Identifying and testing new properties through material informatics;
  • Predicting materials performance through first-principles modeling; and
  • Modeling complex systems.

This computational approach accelerates research and development and IP generation, while improving outcomes at a fraction of the cost. Combined with rich experimental and system design capabilities, our world-class team is leveraging computational tools in new ways to create novel solutions in the microelectronics, life sciences, chemicals, industrial, and defense sectors.

Figure 1: Metal-organic frameworks form ordered 3-D structures that are perfect candidates for atomic simulations.
Figure 1: Metal-organic frameworks form ordered 3-D structures that are perfect candidates for atomic simulations.

Breaking New Ground in Modeling MOFs

The origins of computational modeling for atomic structures date back to the 1950s when high-performance supercomputers came onboard at national laboratories and elite universities. At the U.S. Los Alamos National Laboratories, for example, a computer called MANIAC performed the first classical simulation of a liquid. Built entirely of vacuum tubes, MANIAC took several hours to run this 224-particle Monte Carlo probabilistic simulation. By contrast, the same simulation could be run on today’s laptops in the time it takes a researcher to drink their morning coffee.

The rapid growth in computing power has enabled researchers to model more complex structures and predict their interactions within larger systems. Beginning in the 1980s, researchers moved beyond dense liquids to begin computationally modeling vapors and gases that are absorbed into pores – which differ significantly in their behavior.

In 2001, Japanese researchers were among the first to report the application of quantum chemistry, Monte Carlo simulations, and classical mechanics to study the charge density, adsorption, and movement of small molecules in MOFs – a group of nanoporous materials that are the foundation of NuMat’s technology. National and academic laboratories would spend the next decade refining the algorithms and studying intermolecular interactions to model adsorption and diffusion of gas and vapors inside MOFs.

While universities and governments continue to build new supercomputers, including one at Argonne National Labs a short drive from NuMat, access to such systems is no longer required to perform advanced computational chemistry simulations. Cloud computing behemoths, such as Amazon Web Services, Microsoft Azure, and Google Cloud, have changed the equation, offering supercomputer-scale power with affordable on-demand pricing.

NuMat is the first company in the world to leverage this set of computational tools to engineer and commercialize MOFs for applications that advance computing power. So, NuMat is using the microprocessors of today to make the microprocessors of tomorrow.

The algorithms NuMat uses are based on the latest developments in quantum chemistry, classical mechanics, and probabilistic modeling. NuMat is the first company in the world to leverage this set of computational tools to engineer and commercialize MOFs for applications that advance computing power. So, NuMat is using the microprocessors of today to make the microprocessors of tomorrow.

Discovering New Materials

Our team at NuMat has pioneered the use of computational tools to identify the right MOF for the right application. Computationally driven high-throughput screening can take the infinite world of MOFs and narrow it down to only the most promising candidates to be validated in the lab. This requires sophisticated algorithms that can run fast, high-powered simulations on millions of MOFs at a time.

The development of such tools has paralleled the development of high-throughput screening in the pharmaceutical world. Just as researchers are continually searching billions of drug-like compounds to discover the best candidates for binding to viruses, bacteria, rogue cells, or disordered proteins, the NuMat team is constantly looking to identify MOFs that interact with small molecules for separation, purification, or catalysis applications.

Unlike the large biological structures used for pharmaceutical discovery, however, MOFs are relatively small and highly ordered, making their modeling less constrained. They are also broadly chemically diverse – encompassing essentially the entire periodic table – giving us an enormous canvas for creating an ever-larger catalogue of hypothetical MOF structures.

NuMat’s patented algorithm takes the building blocks of MOFs – organic linkers and metal nodes – and combines them into novel structures. With the use of reproducible workflows and data management best-practices, NuMat has identified more than 10 million possible MOF structures.

These numbers dwarf those of zeolites, the incumbent materials technology. While 1 million or so hypothetical zeolite structures have been generated, we estimate that there could be more than 100 million MOF structures. In the meantime, the Cambridge Structural Database only contains approximately 81,000 experimentally synthesized MOF structures and approximately 200 zeolites.

After generating hypothetical structures, NuMat can screen potential MOF structures for particular properties using computational tools. For example, to identify novel materials for medical oxygen delivery, our team screened 10,000 hypothetical MOFs in a matter of days and identified two candidates with high deliverable-oxygen capacities. Although these adsorption predictions were later confirmed experimentally, our team would have never identified these candidates in such a rapid product development cycle without computational modeling.

NuMat will continue to lead the MOF computational field by pushing the boundaries on generating unique MOF structures and implementing high-fidelity screening tools, while continuously updating computational models based on data from experimental studies.

Identifying New Properties

At NuMat, we are constantly refining our models to identify the unique properties of individual MOFs not captured in the initial discovery and screening process. With a dedicated chemistry team and rapid experimental isotherm equipment, NuMat has generated synthesis, structural, stability, and adsorption data for thousands of MOF materials. Leveraging the power of material informatics, the data helps direct computational chemistry simulations that lead to new insights.

Collecting the data to drive this process is now easier than ever with the Internet of Things. For example, at NuMat, Raspberry Pi single-board computers beam data alongside programmable logic controllers. NuMat has also developed internal software to log samples, equipment data, and laboratory reports. Our computational team can improve simulations based on experimental results in near-real time. These efforts generate rapid insights that turbocharge the materials engineering process.

Figure 2. Predicting gas isotherms – the amount of gas a MOF stores at a given pressure – is a classic example of the simulations performed on MOF structures at NuMat.
Figure 2. Predicting gas isotherms – the amount of gas a MOF stores at a given pressure – is a classic example of the simulations performed on MOF structures at NuMat.

In one notable example, our ION-X® development team experimentally measured the adsorption properties of phosphine in a series of MOFs as a precursor to modeling. Then, our researchers ran unsupervised machine learning on this dataset based on material-property descriptors to provide suggestions on which MOF properties would lead to the largest deliverable capacities. With the results in hand, the research team analyzed structural models of other MOFs and selected a MOF outside the original set that had the optimal pore size to maximize stored deliverable capacities.

With more than 20% higher deliverable capacity than the previous candidate, the MOF holds the world record among MOFs for phosphine deliverable capacity – a key advantage in microelectronics sector, in which more deliverable capacity makes the process of making semiconductor chips much more efficient.

A second example of these capabilities in action involves the high-dimensional optimization of atomic interactions from experimental isotherms. An industrial partner approached NuMat regarding the separation of two gases with diameters that differed by just 1.0 angstroms and nearly the same molecular weight and polarizability. Our chemists measured experimental isotherms, and our computational research team then optimized energy function data to fit the experimental data. The resulting description of atomic interactions informed a screening of several thousand MOF structures from which we identified a suitable MOF for production. This material was validated at pilot scale and now serves as basis for a commercial plant design.

With NuMat’s growing databases, churning through the data will continue to require the most advanced computational architectures. We work on the leading edge with specialized graphics processing units such as those offered by Nvidia for vectorized data sets (for example, blockchain), and are looking toward adopting technology such as field-programmable gate arrays.

Predicting Material Performance

Figure 3. Modeling the properties of MOFs allows NuMat’s team to predict product or process performance, engineer MOFs with enhanced properties, and explain experimental observations.
Figure 3. Modeling the properties of MOFs allows NuMat’s team to predict product or process performance, engineer MOFs with enhanced properties, and explain experimental observations.

While much of NuMat’s computational research is focused on selecting a MOF with an optimized property, other parts of our computational workflow aim to accurately describe the performance of a material in a product or process. This capability involves understanding a combination of material properties, which can be difficult to obtain from experiments alone.

A recent project at NuMat involved understanding the degradation mechanisms of a group of unstable, toxic chemicals prone to spontaneous ignition, making the transport, storing, and handling of these materials especially hazardous. Our team needed to understand these chemicals’ decomposition mechanism and the conditions that could trigger decomposition within the cavities of a MOF framework holding them.

Our computational team was able to use Born-Oppenheimer molecular dynamics in combination with algorithms such as metadynamics – an algorithm to explore reactions of interest – to rank MOFs based on their stability in the presence of these unstable chemicals. The simulation predictions were confirmed through experimental testing. The same process could be applied to finding MOF scaffolds for abating and capturing such toxic chemicals.

Reaction mechanisms are not the only performance characteristics examined. NuMat can also simulate binding energies – how strongly a gas, vapor, enzyme, etc., sticks to the surfaces of a MOF – as well as mass transfer of gases through the pores of a MOF material. Such characteristics are important for evaluating, for example, the effectiveness of a gas mask. Other performance metrics, such as mass transfer, dictate the cycle speed on a pressure swing adsorption (PSA) unit or whether a MOF membrane can kinetically sieve a multicomponent mixture of gases.

Modeling Processes

Figure 4. NuMat’s modeling hierarchy approach.
Figure 4. NuMat’s modeling hierarchy approach.

NuMat not only creates innovative MOFs but also integrated systems such as PSA and temperature swing adsorption (TSA) units for separating and purifying gases. Using process modeling tools allows our researchers to predict optimal working parameters for the technology.

Process modeling embodies the fundamentals of thermodynamics, heat and mass transfer, separations, and fluid dynamics. While much the physics and governing macroscopic equations are well established, providing meaningful parameters is difficult. Many of the material properties we simulate can be directly used in our process models, or experimental measurements can be extended with physics-based theories.

As an example, with an industrial partner, NuMat’s team designed and created a pilot-scale version of a MOF-based separation unit to strip a minority component chemical from an impure gas stream. Its success relied on inputting accurate parameters that would validate the operating model. Our computational tools provided a multi-stepped solution to displace the incumbent technology with a 1000-fold reduction in the impurity.

Before the pilot scale, our team experimentally measured how the gases in question would be adsorbed into our MOFs over a range of temperatures and pressures. NuMat’s computational research team then fit temperature-dependent equations to the data, allowing them to calculate adsorption behavior across the entire temperature operating range. They also ran simulations of mass transfer, which suggested that the process could operate under equilibrium conditions.

The resulting process model enabled NuMat’s team to calculate expected product gas purities and recovery. Once the pilot TSA unit was operational, few operating design parameters needed alteration in order to achieve the required gas purity and recovery levels.

Emerging Technologies and Challenges

Figure 4. NuMat’s modeling hierarchy approach.

As NuMat continues to innovate on the leading-edge of MOF discovery, development, integration, and commercialization, new opportunities and challenges are emerging.

Quantum computing offers one opportunity. Using quantum computing to perform electronic structure calculations on MOFs would allow for an incredible acceleration of research but much work remains to make that a reality. At NuMat, we need to simulate systems with more than 100 atoms, something that has yet to be achieved on current quantum computing hardware.

Beyond the computing power, NuMat is working to make MOF simulations as accurate as possible, both to create new structures using a combinatorial approach and to understand and predict MOF behavior under various conditions. For example, we are working to find ways to model flexible MOFs – MOFs that like to wiggle. Predicting protein conformations enabled by this flexibility is extremely challenging, requiring microsecond-scale molecular dynamics simulations. Likewise, MOFs undergo structural changes upon external stimuli or exhibit kinetic flexibility of their building blocks, which affects gas adsorption or mass transfer properties. These movements introduce another dimension of complexity that our modeling needs to address.

Nevertheless, atomic-based computational modeling does not always have to be accurate to several decimal places but rather can provide qualitative results that drive experimentation. That is the true power of NuMat’s computational approach: the idea that we can rapidly model atomic-level interactions to inform our research and development team in a way never previously possible. This approach is creating high-value, high-performance new materials and processes that are reinventing materials technology and informatics across industries.


For more information, please contact us at: info@numat-tech.com


Return to News