Download PDF Fall Bridge on the Materials Genome Initiative September 29, 2025 Volume 55 Issue 3 The Fall 2025 issue explores the Materials Genome Initiative’s progress and future outlook, showcasing advances in autonomous experimentation, sustainable polymers, next-generation batteries, and the broader role of AI in engineering. Autonomous Experimentation and Self-Driving Labs for Materials Synthesis Using Deposition Techniques Tuesday, September 30, 2025 Author: Benji Maruyama, Ichiro Takeuchi, and Jason Hattrick-Simpers Self-driving labs are the innovation leading to accelerated discovery. Introduction Autonomous Experimentation (AE), also known as Self-Driving Labs (SDLs), promises to speed materials synthesis research and development (R&D) by orders of magnitude, revolutionizing the research process. AE uses artificial intelligence (AI) and robotics to design, execute, and analyze experiments in rapid, iterative fashion, combining the results from the experiments with modeling and simulation to design the next best experiment to do. The MGI (Materials Genome Initiative) Autonomous Materials Innovation Infrastructure (AMII) report (Boswell-Koller et al. 2024) captures demonstrations and advances for materials synthesis that have been made in chemical vapor deposition (CVD), physical vapor deposition (PVD), and electrochemical deposition. By dynamically searching over synthesis parameters, AE can optimize the process quality and speed, resulting in improved and advanced materials at a fraction of the time and labor compared to conventional, human-driven laboratory processes. SDLs can generate and test scientific hypotheses faster and more effectively than human researchers alone. More importantly, SDLs can generate and test scientific hypotheses faster and more effectively than human researchers alone. AE experimental campaigns therefore produce deeper scientific understanding of materials phenomena, enabling rational investigations, extrapolation, and exploitation beyond naïve machine learning–only approaches. It is worth noting that the term “Self-Driving Labs” is more common in the chemistry community, whereas “Autonomous Experimentation” is preferred in the materials community. It is also helpful to distinguish AE/SDL from high-throughput or combinatorial methods, which focus on performing many experiments in parallel or rapidly but not autonomously or iteratively by design. Here, we focus on fully autonomous experimentation, where iterative experiments occur in a closed loop without human intervention. Automated—but not autonomous—research is increasingly being pursued as a stepping stone toward full autonomy (Nikolaev et al. 2014). AE can thus be thought of as “human on the loop” rather than “human in the loop.” Here we discuss the future value proposition for AE/SDLs, assess current progress, and discuss future directions and findings from the MGI AMII report (Boswell-Koller et al. 2024). We also consider the implications of AE/SDLs for materials R&D in the future. Chemical Vapor Deposition CVD is an important materials synthesis technique for thin film materials, 2-D materials, and nanomaterials (Choy 2019). CVD starts with an input stream of precursor gases that thermally decompose onto a substrate to form the target material. Carbon nanotubes (CNTs) can be synthesized using a metal nanoparticle to catalytically decompose the precursor. Here the CNT is templated by the catalyst, driving the cylindrical structure of the carbon nanotube (Rao et al. 2018). CVD in general, and CNT synthesis in particular, presents a significant challenge in understanding and optimizing growth. Key control variables include the gas mixtures—typically a hydrocarbon such as ethylene, reducing gases like hydrogen, and oxidants such as water vapor or CO2 (Bulmer et al. 2023)—as well as experimental parameters like system temperature, temperature ramp rates, and gas flow rates. Interestingly, factors not traditionally considered control variables can also play an important role, such as laboratory humidity, the number of times the furnace tube has been used, and the age of chemical precursors. These factors become especially important when attempting to capture growth phenomena beyond the obvious variables and to explain spurious or intermittent results. ARES is a CVD AE system developed by the Air Force Research Laboratory that was the first fully autonomous system for materials synthesis (Nikolaev et al. 2016). The system, depicted schematically in Figure 1 (Left), consists of a cold-wall CVD system where growth gases are introduced into a chamber. Small silicon pillars, which are essentially microreactors, are seeded with CNT catalysts. A high-power laser heats a single pillar to the target growth temperature, growing carbon nanotubes. The growth is characterized in real time by analyzing scattered laser light with Raman spectroscopy as the nanotube forms. After the experiment is completed and the results analyzed, an AI planner selects the growth conditions for the next experimental iteration, guided by goals defined by the user (Figure 1 [Right]). When designing an AE campaign, the first step is to define its objective. For example, one objective might aim to maximize CNT growth rate while minimizing diameter variation (Waelder et al. 2025). Another possible objective is hypothesis testing, as in a 2024 study where we proposed that the CNT catalyst would be most active under synthesis conditions in which the metal catalyst was in equilibrium with its oxide (Waelder et al. 2024). To test this, we systematically varied the growth environment from more oxidizing (e.g., higher water vapor or CO2 content, lower temperature) to more reducing (e.g., greater hydrocarbon partial pressure, higher temperature), thereby probing catalyst activity for CNT synthesis (Figure 2) as a function of the reducing potential. In this case, the objective was to confirm or refute the reduction hypothesis rather than to maximize CNT growth rate. With the reduction hypothesis as the physical phenomenon under investigation, the next step in campaign design is selecting the planner decision method, also known as the acquisition function (Stach et al. 2021). The acquisition function determines the experimental input conditions expected to advance the campaign objective most effectively. Several strategies exist for acquisition functions, including minimizing overall uncertainty, maximizing a particular feature, or combining both. This highlights an important principle of iterative experimental design: if a campaign identifies a peak, subsequent experiments may test nearby conditions to determine whether even better outcomes exist, an approach known as exploitation (Stach et al. 2021). Alternatively, experiments may probe unexplored regions to search for other, potentially superior peaks, an approach known as exploration. Acquisition functions typically balance exploration and exploitation within the experimental budget to best meet campaign objectives. Compared with traditional methods such as full factorial or one-variable-at-a-time approaches, iterative optimal experimental design achieves progress much more rapidly (Stach et al. 2021). We confirmed our hypothesis that the catalyst exhibits its highest activity when the catalyst metal is in equilibrium with its oxide. Using the ARES AE system, we were able to probe the oxidizing and reducing nature of the growth environment across an exceptionally broad range of conditions covering a 500°C temperature window and oxidizing-to-reducing gas partial pressure ratios spanning 8–10 orders of magnitude. Thus, for AE, a campaign can be framed as a Blackbox or naïve optimization of conditions to maximize a target property. However, the more powerful objective is hypothesis testing with SDLs, as the resulting scientific insights can be generalized for broader impact and applied to related material syntheses and reactor scale-up. While there are currently limited examples of fully autonomous CVD systems, we expect a significant increase in their number and capabilities as automation and in situ/in-line characterization advances. Physical Vapor Deposition PVD techniques such as magnetron sputtering, molecular beam epitaxy (MBE), and electron-beam evaporation form the backbone of the electronics industry by enabling crucial thin-film synthesis and device fabrication. PVD tools can produce uniform coatings of a wide range of materials, from nanometer- to micron-scale thicknesses, making them well-suited for materials exploration. Modern advanced PVD systems are often equipped with automated operation capabilities (e.g., sequential wafer or chip transfer, or executing programmed deposition recipes), which serve as key prerequisites for AE workflows. The concept of designing materials through controlled deposition and (multi)layering of ultrathin (sometimes atomically thin) films has long provided the materials science community with a powerful approach to discovering and realizing properties not found in nature. To accelerate the discovery of new materials and their properties using PVD, high-throughput experimentation provides a particularly powerful platform. In this approach, combinatorial library wafers or chips contain arrays of samples with varying compositions, enabling sequential experiments to be conducted in a straightforward and efficient manner. The ability to synthesize and characterize material arrays sequentially—and even in closed loops—greatly enhances the effectiveness of AE. In some cases, the exercise of self-driving combinatorial experimentation can be singularly focused on characterization. While fabrication of libraries such as thin-film composition spreads can often be carried out quickly and reliably using PVD techniques (Green et al. 2017), quantitative evaluation of physical properties of interest can be time- and resource-intensive for each individual sample. In such situations, Gaussian process models can effectively guide the measurement sequence across the library. For example, Kusne and colleagues identified the composition of a phase-change memory material with the largest bandgap contrast between amorphous and crystalline phases from a prefabricated ternary thin-film composition spread after measuring only a fraction of the full compositional range (Kusne et al. 2020). The newly discovered phase-change memory material, Ge4Sb6Te7, lies at a structural phase boundary between a host matrix and a secondary phase on the Ge–Sb–Te compositional phase diagram. This composition exhibits unusually large contrast between on and off states and, in recent scaled-up device comparisons, was found to significantly outperform the widely used Ge2Sb2Te5 (Khan et al. 2023; Wu et al. 2024). Importantly, the functionality of this material arises from a coherently formed nanocomposite state at the phase boundary. Thus, in this case, autonomous exploration not only led to the discovery of a new material but also revealed a novel strategy for designing high-performance phase-change memory systems. In another example of AE on a combinatorial library fabricated by a PVD technique, Liang et al. (2024) demonstrated real-time, self-driving, cyclical interaction between experiments and computational predictions for materials exploration. Specifically, they performed rapid mapping of a temperature–composition phase diagram, a fundamental task in the search and discovery of new materials. Thermal processing and experimental determination of compositional phase boundaries in thin films were autonomously integrated with real-time updates of phase diagram predictions through Gibbs free energy minimization. Using this workflow, the eutectic phase diagram of the Sn–Bi binary thin-film system was accurately determined on the fly from a self-guided campaign that sampled only a small fraction of the entire composition–temperature space, achieving a six-fold reduction in the number of required experiments. This study provided the first demonstration of real-time, autonomous, and iterative integration of experiment and theory carried out entirely without human intervention (Liang et al. 2024). To establish fully autonomous PVD-based closed-loop cycles that integrate synthesis and characterization at each step, Shimizu and colleagues (2020) demonstrated the successful operation of a robot-controlled multi-chamber vacuum system. By transferring each deposited thin film sample from a sputtering chamber to a materials characterization chamber (for resistance measurements) at each iteration, the robot-based system was able to find optimized materials synthesis conditions within a small number of deposition runs (Shimizu et al. 2020). While autonomous science systems incorporating sample-handling robots are gaining popularity across various areas of materials science, they can be costly and complex to operate and maintain. In contrast, in situ monitoring of thin-film processes has long been a central practice in the thin-film community. Because modern PVD chambers often accommodate modular in situ characterization tools, any technique that can provide direct feedback on thin-film quality can serve as an effective foundation for closed-loop workflows. Using this approach, Lippmaa and colleagues (2002) have demonstrated autonomous control of epitaxial unit cell-level growth of oxide thin films implemented in a combinatorial pulsed laser deposition (PLD) system (Figure 3). This PLD system is notable for its ability to fabricate multiple distinct thin-film samples in a single pump-down on one wafer or chip through combinatorial masking, while simultaneously employing a reflection high-energy electron diffraction (RHEED) system for in situ monitoring of the surface nanostructure at different positions on the library substrate. Building on this, the development of computer vision–based automated quantitative analysis of live RHEED images (Liang et al. 2022) enabled autonomous navigation of multi-dimensional deposition parameter space (temperature, partial pressure, and laser pulse rate), rapidly identifying optimal growth conditions for targeted material phases (Price et al. 2025). Closed-loop PLD has also been demonstrated using Raman spectroscopy as the characterization feedback mechanism (Harris et al. 2024). In this way, provided that sequential synthesis or deposition and in situ monitoring of relevant material properties are feasible, virtually any PVD system can, in principle, be converted into an autonomous platform. This concept extends beyond PVD, and many additional agile demonstrations of autonomous thin-film—and even bulk—materials synthesis tools are expected in the near future. Electrochemical Deposition Electrochemical deposition, or plating, is a widely used technique for producing conformal coatings on surfaces. The first demonstration of the technology was by Luigi Brugnatelli in 1805, when he gold-plated silver medallions (Reid et al. 1975). Copper plating is a common example in introductory chemistry classes, typically performed using a copper sulfate solution and a copper anode. Modern industrial applications include Zn–Ni coatings for automotive components (DeCost et al. 2022) and through-silicon vias for connecting stacked integrated circuit layers (Kim et al. 2022). At its core, basic level plating is relatively straightforward: a solution containing the target cation is prepared and a sufficiently reducing bias is applied to drive its reduction onto the cathode. Increasing the applied potential (or, more commonly, the applied current) raises the deposition rate, which can influence the morphology of the resulting film. As long as the surface is free of oil, debris, and oxides, electrochemical reduction proceeds without obstruction. To create an alloy, one can simply use a solution containing multiple cations and identify the appropriate redox potential to plate each element. In principle, adjusting the relative concentrations of the cations in solution allows fine-tuning of the deposited composition. In practice, however, electroplating is much more complex. The choice of metal salt used to supply the cation plays an important role in the deposit, with cyanides generally producing smooth and uniform coatings while modern baths more often use sulfates or chlorides. Similar to physical vapor deposition, processing parameters such as pH, temperature, and agitation are commonly adjusted to control deposition rate and quality. In addition, deposition potential and overall deposition time are frequently used to influence the final coating. For alloys, considering differences in electrochemical redox potential alone does not account for behaviors such as anomalous co-deposition, which can lead to unexpected oversaturation of one (typically less noble) metal in the coating. Unlike physical vapor deposition, the most challenging part of the plating search space is not the mixing of metals, but rather the identification of the organic and inorganic additives such as brighteners, levelers, surfactants, complexing agents, and pH control agents that facilitate deposition. For instance, during electroplating, the pH often increases as hydronium ions are consumed and hydroxyl ions are produced, leading to the precipitation of metal hydroxides. A buffering agent, such as boric acid, is typically added to maintain the pH and suppress precipitation. Conversely, complexing agents, such as cyanide, bind specifically to the metal cation to form stable complexes in solution, releasing the metal only near the cathode in response to local changes in pH or intermediates. None of these parameters are independent, and changes to one strongly influence the optimal deposition conditions. There are limited examples of SDLs for electrodeposition reported in the technical literature. Joress and colleagues attempted to mix metal cation salts to facilitate the autonomous deposition of Ni–Co alloys via a scanning droplet cell system (Joress et al. 2022). In this study, a series of alloys were deposited using mixtures of cobalt sulfate and nickel sulfate exposed to different reduction potentials. Post-deposition analysis revealed that a simple rule of mixtures linearly superimposing the individual metal deposition currents weighted by their relative concentrations was insufficient to predict the composition of the coating. This is consistent with literature reports of weak anomalous co-deposition in the Ni–Co system, although in this case the deviation from solution stoichiometry was only a few percent. AE will allow the discovery and optimization of materials and processes that are not possible or practical with current approaches. More recently, Quinn and colleagues (2024) developed a low-cost CNC gantry-based tool to investigate the electrodeposition of poly(3,4-ethylenedioxythiophene):poly(styrene sulfonate) (PEDOT:PSS) films. They aimed to identify correlations between monomer concentration, deposition time, and deposition voltage with the electrochromic color change of the coating and the total charge passed during film reduction. Using this approach, the investigators were able to rapidly converge on a range of optimal deposition times and voltages for color-changing films. The lack of demonstrated electroplating depositions, coupled with the broad importance of the synthesis method, highlights an opportunity for the AE community. Since most work uses a solution carrying cations and additives, it is readily compatible with Scanning Droplet Cell or pipetting platforms. One challenge is keeping the cations in solution; for example, sudden changes in orifice diameter in a tube adapter can cause a saturated salt solution to precipitate and plug the line. Other considerations include identifying the appropriate figure of merit to guide the optimization. For single-element coatings, simple metrics such as reflectivity, roughness, coating uniformity in color, and linear polarization resistance sweeps can suffice to identify high-quality coatings. However, for multi-element coatings, it is also important to understand both the average composition and its depth dependence, which is challenging to achieve within the loop. In such systems, common electrochemical measurements can also conflate a desired property (e.g., oxygen evolution reaction overpotential) with an undesired property (e.g., dissolution of the coating). For these reasons, autonomous electrodeposition represents a low-hanging fruit for the community to pursue. There are significant opportunities to use AE to optimize and monitor plating conditions for existing solutions, improving both solution efficiency and coating quality. Looking ahead, substantial gains could be achieved by using such platforms to identify novel plating solution chemistries—including cation concentrations and additives—to produce high-performance coatings. Additionally, AE offers the potential to generate high-quality datasets that can help resolve existing disputes regarding mechanisms and uncover new electrochemical phenomena. Gaps, Outlook, and Future Directions The outlook for AE in materials development and synthesis is very bright. Pathfinders in physical and chemical vapor deposition and in electrochemistry research have demonstrated the value of the technology in accelerating research to improve both processes and fundamental understanding. Importantly, AE impacts both science and technological advancement. Moreover, it has the potential to influence domestic and global challenges related to critical minerals, economic competitiveness, human welfare, energy, and the environment. In the summer of 2024, the MGI Autonomous Materials Innovation Infrastructure Interagency Working Group (AMII-IWG) published a workshop report in support of the MGI 2021 Strategic Plan (Boswell-Koller et al. 2024). The report captured the current state of the AMII based on input from workshop participants across industry, academia, and federal government agencies. The report’s findings highlight substantial progress in the United States and summarize existing capabilities. However, significant infrastructure gaps remain, including hardware, software, decision tools, and workforce development, which are necessary to enable AE to be more broadly accessible to researchers. The required infrastructure investment is substantial and indeed comparable to the automotive industry’s transition from manual labor to automated assembly-line robots. This involves modernization of R&D, large capital investments in automation, and workforce retraining for automated systems. Digitization of advanced materials R&D also requires major investments in software and software engineering. The digital transformation of research and development will continue to evolve, emphasizing data generation, exploitation, and AI-driven reasoning as foundational tools for scientific and technological advancement. Associated with the AMII were the 2024 MGI Challenges (MGI 2024). The five MGI Challenges, which reflect priorities from multiple federal agencies, are summarized in Table 1. The intent of the Challenges was to “utilize challenges to help unify and promote adoption of the Materials Innovation Infrastructure—through the expansion and integration of capabilities including autonomy, artificial intelligence, and robotics—to realize solutions to challenges of national interest.” Autonomous Experimentation Future of Synthesis Workforce is a critical issue for AE. First, our STEM workforce is insufficient, and demographic forecasts predict a reduced supply and rapid loss of expertise as the current workforce retires. Second, AE requires a modernized workforce capable of conducting materials R&D while leveraging advances in AI, autonomy, and digital transformation. Addressing this challenge will require retooling university curricula and significant retraining for the current workforce. Often overlooked is the importance of early preparation of K–12 students for future work in AI and autonomy. Developing this future workforce will demand substantial investment in teacher training, curriculum development, and hands-on, experiential, or project-based learning. Finally, public–private partnerships were identified as essential to advancing the AMII. We expect the near future of AE to focus on building infrastructure and advancing decision methods to accelerate research and reduce the repetitive, mindless tasks faced by bench researchers, allowing humans to concentrate on higher-level research goals and understanding. This represents a shift in the fundamental role of human researchers away from tedium, ironically creating more room for humans (especially students) to exercise greater autonomy in their research. We caution that safety and security considerations must remain central to AE, learning from the autonomy community that delegation of decision authority to a human or to a research robot does not constitute abdication of responsibility for research outcomes. In the near term, we observe several trends in AE/SDLs. As the technology is still in its early stages, much of the focus will be on expanding laboratory automation, including operando, in situ, and in-line characterization, as well as synthesis, sample transfer, and sample loading. It is encouraging that materials research is a focus of recent advances in AI large language models and foundational models (Ball 2025). We also see progress in AI decision methods that incorporate materials prior knowledge (e.g., phase diagrams) and in-line modeling and simulation into the AE iterative loop. More researchers are integrating explicit hypothesis testing, generation, and regression (Noack and Ushizima 2023). Finally, there are efforts to improve access and reduce barriers to AE. Cloud labs like the Emerald Cloud Lab aim to provide automated and autonomous research fulfillment that is accessible remotely online (Canty et al. 2025). Additional initiatives focus on affordability, including LEGOLAS autonomous chemistry robots from the University of Maryland and the National Institute of Standards and Technology (Saar et al. 2022) and Athena Educational Autonomous Experimentation 3D printers from the Air Force Research Laboratory (Lo et al. 2024), among others. As AE systems mature, we note a current and natural bias toward replicating human laboratory workflows in AE, rather than taking advantage of de novo AE workflows that can be faster and more efficient, as demonstrated in the PVD community (see above) through the use of automation and in situ diagnostic tools. Actions or constraints that might be obvious to human researchers are often overlooked when programming an AE robot to conduct research. For example, in an early ARES campaign, we allowed the planner to choose any temperature, and it selected negative absolute temperatures that were clearly unphysical to human researchers. Finally, the phenomenon of serendipitous discovery is often cited as a potential loss with AE. However, AE/SDLs can be programmed to identify unexpected results and flag them for further investigation and attention by human researchers. In the longer term, we see a revolution in how research is conducted, with ever-increasing speed and progress reducing R&D time from decades to years. We see AE as a force multiplier, enabling one researcher to do the work of a hundred. AE will allow the discovery and optimization of materials and processes that are not possible or practical with current approaches. By enabling exponentially more and better experimentation, AE shifts the balance of risk and reward in deciding what research to pursue. That is, if a researcher can conduct many more experimental iterations, they can attempt high-risk, high-reward AE campaigns that might otherwise not be worth risking progress toward a successful dissertation or research grant. The impact of AE/SDLs is potentially revolutionary. However, there are concerns about the changing roles of human researchers as AE expands. What will humans do? Will robots take away jobs? How will human researchers work with AE systems? As with any revolution, these seismic shifts can be concerning and disruptive. We see AE/SDLs as an augmentation to human efforts in the same way that electronic computers freed the human computers of the 1940s to focus on more interesting work. We see human researchers using AE systems to multiply their efforts as farmers use tractors. Because technology advances exponentially (Mokyr 1992), we see this multiplication of effort and speed as necessary to maintain the rate of progress such that AE systems will augment human researchers rather than displace them. The impact of AE/SDLs is potentially revolutionary. First, there is the prospect of an exponential explosion in materials science advancement, akin to a “Moore’s Law” for the speed of research (Seffers 2017). Next, we expect to narrow the “Valley of Death” that prevents many technological advancements from transitioning to use. In addition to accelerating materials R&D from decades to years, we expect materials advances to respond to new needs more rapidly, solving problems in the near term instead of the far term. Finally, we hope to see barriers to AE drop so significantly that citizen scientists can participate in research, unleashing the power of many minds on the vast endeavor of materials R&D. Broadly, we expect AE to impact economic prosperity, national security, and human health and welfare. References Ball DW. 2025. The Future of Materials Science: AI, Automation, and Policy Strategies. Mercatus Center, George Mason University. Online at https://www.mercatus.org/research/policy-briefs/future- materials-science-ai-automation-and-policy-strategies. Boswell-Koller C, Warren J, Maruyama B, Yang C. 2024. Accelerated Materials Experimentation Enabled by the Autonomous Materials Innovation Infrastructure (AMII): A Workshop Report; Subcommittee on the MGI. Online at https://www.mgi.gov/sites/default/files/documents/MGI_ Autonomous_Materials_Innovation_Infrastructure_Workshop_ Report.pdf. Bulmer JS, Sloan AWN, Glerum M, Carpena-Núñez J, Waelder R, Humes J, Boies AM, Pasquali M, Rao R, Maruyama B. 2023. Forecasting carbon nanotube diameter in floating catalyst chemical vapor deposition. Carbon 201(5 January):719–33. Canty RB, Bennett JA, Brown KA, Buonassisi T, Kalinin SV, Kitchin JR, Maruyama B, Moore RG, Schrier J, Seifrid M, Sun S, Vegge T, Abolhasani, M. 2025. Science acceleration and accessibility with self-driving labs. Nature Communications 16(1):3856. Choy K-L, ed. 2019. Chemical Vapour Deposition (CVD): Advances, Technology and Applications. CRC Press. DeCost B, Joress H, Sarker S, Mehta A, Hattrick-Simpers J. 2022. Towards automated design of corrosion resistant alloy coatings with an autonomous scanning droplet cell. JOM 74(8):2941–50. Green ML, Choi CL, Hattrick-Simpers JR, Joshi AM, Takeuchi I, Barron SC, Campo E, Chiang T, Empedocles S, Gregoire JM, and 7 others. 2017. Fulfilling the promise of the Materials Genome Initiative with high-throughput experimental methodologies. Applied Physics Reviews 4(1). Harris SB, Biswas A, Yun SJ, Roccapriore KM, Rouleau CM, Puretzky AA, Vasudevan RK, Geohegan DB, Xiao K. 2024. Autonomous synthesis of thin film materials with pulsed laser deposition enabled by in situ spectroscopy and automation. Small Methods 8(9). Joress H, DeCost B, Hassan N, Braun TM, Gorham JM, Hattrick-Simpers J. 2022. Development of an automated millifluidic platform and data-analysis pipeline for rapid electrochemical corrosion measurements: A pH study on Zn-Ni. Electrochimica Acta 428(October):140866. Khan AI, Yu H, Zhang H, Goggin JR, Kwon H, Wu X, Perez C, Neilson KM, Asheghi M, Goodson KE, Vora PM, Davydov A, Takeuchi I, Pope E. 2023. Energy efficient neuro‐inspired phase–change memory based on Ge4Sb6Te7 as a novel epitaxial nanocomposite. Advanced Materials 35(30):2300107. Kim S-H, Braun TM, Lee H-J, Moffat TP, Josell D. 2022. Microstructure and texture in copper filled millimeter scale through silicon vias. Journal of The Electrochemical Society 169(3):032508. Kusne AG, Yu HS, Wu CM, Zhang HR, Hattrick-Simpers J, DeCost B, Sarker S, Oses C, Toher C, Curtarolo S, and 6 others. 2020. On-the-fly closed-loop materials discovery via Bayesian active learning. Nature Communications 11. Liang H, Stanev V, Kusne AG, Tsukahara Y, Ito K, Takahashi R, Lippmaa M, Takeuchi I. 2022. Application of machine learning to reflection high-energy electron diffraction images for automated structural phase mapping. Physical Review Materials 6. Liang H, Wang C, Yu H, Kirsch D, Pant R, McDannald A, Kusne AG, Zhao JC, Takeuchi I. 2024. Real-time experiment-theory closed-loop interaction for autonomous materials science. arXiv preprint. arXiv:2410.17430. Lippmaa M, Koida T, Minami H, Jin Z, Kawasaki M, Koinuma H. 2002. Design of compact pulsed laser deposition chambers for the growth of combinatorial oxide thin film libraries. Applied Surface Science 189(3–4):205-9. Lo S, Baird SG, Schrier J, Blaiszik B, Carson N, Foster I, Aguilar-Granda A, Kalinin SV, Maruyama B, Politi M, Tran H. 2024. Review of low-cost self-driving laboratories in chemistry and materials science: The “frugal twin” concept. Digital Discovery 3(5):842–68. MGI [Materials Genome Initiative]. 2024. The 2024 Materials Genome Initiative (MGI) challenges. Online at https://www.mgi.gov/2024-materials-genome-initiative-mgi- challenges. Mokyr J. 1992. The Lever of Riches: Technological Creativity and Economic Progress. Oxford University Press. Nikolaev P, Hooper D, Perea-Lopez N, Terrones M, Maruyama B. 2014. Discovery of wall-selective carbon nanotube growth conditions via automated experimentation. ACS Nano 8(10):10214–22. Nikolaev P, Hooper D, Webber F, Rao R, Decker K, Krein M, Poleski J, Barto R, Maruyama B. 2016. Autonomy in materials research: A case study in carbon nanotube growth. npj Computational Materials 2:16031. Noack M, Ushizima D., eds. 2023. Methods and Applications of Autonomous Experimentation. CRC Press. Price C, Takeuchi I, Rondinelli J, Chen W, Hinkle C. 2025. AI-accelerated electronic materials discovery and development. Computer 58(2):98–104. Quinn H, Robben GA, Zheng Z, Gardner AL, Werner JG, Brown KA. 2024. PANDA: A self-driving lab for studying electrodeposited polymer films. Materials Horizons 11(21):5331–40. Rao R, Pint CL, Islam AE, Weatherup RS, Hofmann S, Meshot ER, Wu FQ, Zhou CW, Dee N, Amama PB, and 32 others. 2018. Carbon nanotubes and related nanomaterials: Critical advances and challenges for synthesis toward mainstream commercial applications. ACS Nano 12(12):11756–84. Reid FH, Goldie W, Brooman EW. 1975. Gold plating technology. Journal of The Electrochemical Society 122(6). Saar L, Liang H, Wang A, McDannald A, Rodriguez E, Takeuchi I, Kusne AG. 2022. The LEGOLAS Kit: A low-cost robot science kit for education with symbolic regression for hypothesis discovery and validation. MRS Bulletin 47:881–85. Seffers GI. 2017. Scientists pick AI for lab partner. Signal. Online at https://www.afcea.org/signal-media/scientists-pick-ai- lab-partner. Shimizu R, Kobayashi S, Watanabe Y, Ando Y, Hitosugi T. 2020. Autonomous materials synthesis by machine learning and robotics. APL Materials 8:111110. Stach E, DeCost B, Kusne AG, Hattrick-Simpers J, Brown KA, Reyes KG, Schrier J, Billinge S, Buonassisi T, Foster I, and 11 others. 2021. Autonomous experimentation systems for materials development: A community perspective. Matter 4(9):2702–26. Waelder R, Park C, Sloan A, Carpena-Núñez J, Yoho J, Gorsse S, Rao R, Maruyama B. 2024. Improved understanding of carbon nanotube growth via autonomous jump regression targeting of catalyst activity. Carbon 228:119356. Waelder R, Kim W, Pitt MA, Myung JI, Maruyama B. 2025. Multi-objective Bayesian optimization of carbon nanotube yield and diameter control at synthesis. APL Machine Learning 3:026114. Wu X, Khan AI, Lee H, Hsu C-F, Zhang H, Yu H, Roy N, Davydov AV, Takeuchi I, Bao X. 2024. Novel nanocomposite-superlattices for low energy and high stability nanoscale phase-change memory. Nature Communications 15:13. About the Author:Benji Maruyama is autonomous materials lead, Materials & Manufacturing Directorate, Air Force Research Laboratory. Ichiro Takeuchi is professor and chair, Materials Science and Engineering, University of Maryland. Jason Hattrick-Simpers is professor in the Department of Materials Science and Engineering at the University of Toronto.