Stockpile modernization, enhanced defense systems, rapid threat response, advanced sensing and detection, and predictable energy and supply chain security—all critical components of Lawrence Livermore’s mission—rely on the advancement of materials research. In the commercial sphere, the Laboratory’s work to develop novel materials accelerates additive manufacturing innovations, which have led to technology transfer and industrial partnerships. To meet ever-changing material development needs, researchers must consider more variables, such as a material’s composition, performance properties, and lifetime predictability in different circumstances and environments. However, with each added variable, the sheer number of combinations makes predicting how a new material might perform increasingly difficult and testing too complex with traditional, iterative experiments.
To outpace these challenges, Livermore scientists and engineers across a spectrum of disciplines are integrating their expertise to accelerate the pace of materials discovery. Called the SPOC (studying polymers on a chip) project, the research aims to automate the process of materials development—from selecting the best materials for a given purpose to mixing and depositing them to testing them on a printed circuit board (PCB). Though still in its early stages, SPOC shows promise to improve efficiency and predictability of materials research, reduce the time needed for testing, and speed the overall process of research. By leveraging Livermore’s additive manufacturing technologies and high-performance computing (HPC) capabilities, SPOC is poised to significantly reduce the time from materials design to prototyping.
Live Long and Automate
The SPOC platform enables high-throughput screening (HTS), a characterization method that simultaneously evaluates many chemical or material combinations for different properties or activities. One method by which the SPOC team prepares and tests the samples involves placing different polymer samples, made from different material compositions of viscous fluids, into discrete measurement wells on a PCB. Once the deposition is complete, the samples are cured with heat or light, solidifying the polymer and ensuring consistent sample properties for measurement. For applications such as polymer electrolytes for battery materials, each sample well on the PCB is electrically connected via a potentiostat—an instrument for electrochemical measurements—that applies a small voltage across the samples and measures the electrical current. These measurements reveal impedance data and can be performed across the entire PCB array, providing insight on the ionic conductivity of polymer electrolytes to create next-generation solid-state batteries. Every stage of this process is automated and controlled via Python computing scripts. Depending on the research requirements, the SPOC team can devise other preparation and testing protocols as needed. For example, the team can also do mechanical measurements, compositional measurements, and imaging to understand inherent properties within the deposited polymers.
Previously, the process of mixing and testing new materials was laborious and time-consuming. “Before SPOC, whenever we developed formulations, we identified the properties the material needed, mixed the materials together by hand, printed the formulation to the desired geometry, and measured the printed structure in another machine to verify its properties,” says Livermore polymer chemist and project lead Johanna Schwartz. “At that point, we often discovered that the sample did not meet the desired performance, so we had to start over.”
SPOC’s automation capabilities in mixing have streamlined this process significantly. The SPOC platform uses a special print head that incorporates multiple computer-controlled feedstocks and a mechanically robust mixer to deliver precise material combinations to specific locations on the PCB, eliminating the need to mix the materials in advance. Using active-mixing direct ink writing (DIW), the materials are added to separate reservoirs and blended in a controlled manner at the printhead. The automation enables the SPOC team to vary composition on-machine and study many different materials at once that previously would have required hand mixing. Conventional pipetting-based HTS systems, for example, have a maximum viscosity limit of 1,000 to 10,000 centipoise (cP), similar to the viscosity of honey. SPOC’s automated mixing capabilities can extrude resins with viscosities as low as water (1 cP) and as h igh as 1 million cP, a consistency closer to window putty or toothpaste.
In addition, the researchers’ ability to conduct tests on multiple samples directly on the PCB allows them to gain a first-pass understanding of properties they might want to study in the future. “What we’re doing is similar to using a very expensive pancake mixer where we control the input of our recipe and are reading out directly what makes a good pancake. We want to know if it has the right fluffiness and if it does what we want it to do, for example, absorb the maple syrup,” says Schwartz. “Everyone can imagine the best pancake they’ve tasted and identify what properties made it the best. Similarly for polymers, certain properties exist that we need to achieve. Knowing these properties lays the groundwork for using SPOC to find advanced materials for any application.”
Insufficient Data Invites Opportunities
HPC, machine learning (ML), and AI are also key components of improving, automating, and accelerating materials development. Led by Tuan Anh Pham, a scientist in Livermore’s Quantum Simulations group, SPOC is leveraging Livermore’s resources and expertise in quantum simulations to improve understanding of polymer materials in real time and enhance the predictability of how different formulations might behave and which might be the most effective for a given application.
Data collection and analysis are a critical component of the SPOC project, but the process of structuring data management was challenging. “Since our data set was small, we had to incorporate historical data, and making historical data usable and AI-ready is a messy, resource-intensive process,” says Livermore scientist Jayvic Jimenez, who leads SPOC’s databasing efforts. “We standardized the data, established solid data practices moving forward, and centralized our databases, which enabled us to automate data analysis, troubleshoot, and eventually apply AI tools.”
Focusing on this critical element in the experimental pipeline has helped the team see their vision take form as they collect new data. “SPOC offers the ability to intake the data in the same way across different materials or samples and control what’s changing across the variables,” says Schwartz. Standardization enables the SPOC team to analyze “bad samples,” which can both inform and define a good sample and better train AI and ML models for use in future efforts toward meeting national security and commercial needs. “AI and machine learning will be some of the most significant tools for us to leverage, but we need a data repository,” says Livermore engineer Brian Au, who leads SPOC’s hardware and robotics advancements. “With the Laboratory’s research in materials science and so much data from different programs available, the SPOC project offers a useful stepping-stone into the future of materials science and manufacturing at Livermore and across the country.”
Logic Is Only the Beginning
The year 2026 is the “year of integration” at the Laboratory, with several institutional efforts converging around similar themes such as connecting systems, people, data, and capabilities so that Livermore can deliver on larger and more complex missions. SPOC’s collaborative structure offers an excellent example of this goal. With so many scientific fields targeted, SPOC’s home at the Laboratory is particularly fitting. “SPOC is highly interdisciplinary and involves experts from many areas of research,” says Schwartz. “We need an understanding of material composition and performance properties, which requires characterization. However, we also need to understand the significant amounts of data we’re generating, so we need a database to train computer models for the future.” Pham says, “From the beginning, this project has focused on integration. We cannot solve a problem just by conducting experiments or running simulations alone, even with the best tools in the world. We’re integrating simulations, experiments, data science, and machine learning to solve challenging problems.”
SPOC’s research team is already applying the platform to battery research. SPOC was invented and built to improve the ionic conductivity of solid polymer electrolytes in both lithium-and sodium-based batteries. Polymer electrolytes are safer than liquid electrolyte systems, as they are not flammable or prone to thermal runaway, but the ionic conductivities of solid polymer electrolytes are too low for industrial relevance. Using SPOC, the team screened more than 860 formulations of solid polymer electrolytes within one year. During this initial effort, the team also validated the SPOC approach compared to traditional ionic conductivity testing methods using coin cells. Michell Marufu, a Livermore staff scientist and polymer chemist, says, “Of the materials we’re working with, particularly the best-performing ones, we wanted to validate trends and ensure that even if the results were not exactly the same, they still correlated between the traditional and SPOC conductivity tests."
In the study, the SPOC platform’s trends in ionic conductivity closely matched those observed in traditional testing, validating the ability to use SPOC to quickly home in on desirable formulations for further characterization. SPOC enabled the rapid extension of this screening to sodium-ion battery systems, revealing similar trends and identifying optimal formulations in a fraction of the time. Compared to traditional testing, SPOC demonstrated a significant increase in research throughput, with 3.9 times greater unique formulations, 3.1 times more discrete samples, and 3.7 times more measurements produced.
Another aspect of this study included integrating SPOC data with Bayesian active learning models. These efforts suggested that AI-guided screening could identify top-performing materials roughly twice as fast as conventional approaches. SPOC’s automated workflow projected a sixfold increase in annual sample throughput, yet required only a tenth of the manual ef fort and offered a transformative leap in the rapid identification of materials and properties essential for scientific innovation.
This foundational database laid the groundwork for industry collaboration with California-based innovative battery start-up DarmokTech and selection for funding through the California Energy Commission EPIC (Electric Program Investment Change) Program. Livermore will screen and identify polymer electrolytes for safe, stable, and recyclable sodium solid-state batteries for grid storage applications to create a sodium battery that can match or exceed performance of more commonly used lithium-ion batteries. In addition to sodium’s higher abundance and safety compared to lithium, sodium-ion batteries offer better opportunities for recyclability, which can minimize hazardous waste and enable circular economy approaches.
The Future Frontier
As the SPOC platform continues to mature, its applicability to many different materials and industries shows great promise. At a more localized level, the results from current research efforts with SPOC are informing how the team can further modify and optimize the tools at their disposal. “As the SPOC platform grows and we’re able to test a wider variety of materials, we’ll also need to ensure the platform is modular enough to incorporate those different materials and application spaces,” says Au. “The work we’re doing with SPOC is enabling us to reframe how we think about manufacturing at the Laboratory and more broadly.”
From a data management perspective, SPOC is an example of how experimental data can meet AI integration and highlights Livermore’s approach to tackling different challenges. Jimenez says, “Experimental data sets can be expensive to generate, so they’re often sparse in volume, which makes applying AI tools more difficult. Our approach, leveraging medium-sized but high-fidelity data sets, demonstrates that closing the loop of AI and experimental data is applicable in other research areas.”
SPOC’s automation capabilities offer researchers more room to flex their creative problem-solving skills in addressing some of the thorniest and most critical research areas for national security and commercial uses. “Ten years from now, I’d like SPOC to be a truly autonomous screening platform where it can iterate on what it learns and report out hypotheses for testing with a human in the loop,” says Schwartz. “In a new system like this, we won’t have to mix a million different formulations by hand, but we would still have millions of formulations of information and data. We can train the system on this growing course of data, so the platform will constantly be expanding and identifying other problems to solve.”
—Sheridan Hyland
For further information contact Johanna Schwartz (925) 422-4273 (schwartz28 [at] llnl.gov (schwartz28[at]llnl[dot]gov))
