Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (15 results)

  • Thesis
  • Peer Reviewed
UC Berkeley Electronic Theses and Dissertations (2024)

The development of new inorganic materials largely depends on manual experiments that are costly and time intensive. While automation has greatly advanced the computational discovery of promising new materials, the rate at which they are experimentally synthesized has fallen behind. Bridging this gap requires an improved approach to materials synthesis and characterization, whereby automation is used to streamline the experimental realization of predicted compounds. In this dissertation, I will summarize my contributions to this area. These include automating the collection and analysis of X-ray diffraction patterns, developing theory-driven decision-making algorithms to guide experimental solid-state synthesis trials, and implementing these methods in a fully autonomous, robotic platform known as the A-Lab.

X-ray diffraction (XRD) is a cornerstone of materials research that is widely used to identify and characterize the structures of distinct crystalline phases. Traditional interpretation of XRD patterns requires manual analysis, which becomes challenging when dealing with multi-phase samples that are often complicated by experimental artifacts such as lattice strain and texture. In Chapter 2, I will describe the development and validation of a machine learning (ML) framework that can automate the identification of crystalline materials from XRD patterns. This framework leverages an ensemble of convolutional neural networks, uniquely trained with physics-informed data augmentation to ensure they are robust against common experimental artifacts. A distribution of predicted phases is generated for each pattern given to these trained models, from which a measure of prediction confidence is evaluated. This method outperforms traditional peak search-match algorithms on a variety of experimental samples without requiring manual intervention, making autonomous phase identification possible.

Because ML models are fast once trained, they can be integrated with experimental measurements to perform analysis in real time. This provides the opportunity to use any information gained from preliminary analysis to control the subsequent measurements, improving the efficiency of data collection. Such an approach can benefit XRD measurements, which typically require 20-30 min of scan time per sample to obtain results that have sufficient quality for post hoc analysis. As outlined in Chapter 3, a much shorter scan time of 5-10 min per sample can be achieved by using in-line ML analysis to steer the diffractometer toward parts of the XRD pattern that matter most for phase identification. This approach is shown to provide more precise detection of impurities and short-lived reaction intermediates that are critical to the study of solid-state synthesis.

In early attempts to synthesize a new compound, XRD often reveals the formation of unwanted byproducts instead of the desired target. Avoiding these byproducts and achieving the target requires careful redesign of the experimental procedure. In solid-state synthesis, the most common approach used to make bulk inorganic materials, redesigning the experiments generally involves choosing alternative precursors or reaction conditions. While conditions like temperature and partial pressures are numerical and can therefore be optimizing using well established methods like Bayesian optimization, precursor selection requires a different approach. In Chapter 4, I will describe an algorithm we developed to optimize the selection of precursors used in solid-state synthesis by actively learning from experimental outcomes. It does so by identifying unfavorable reactions that lead to unwanted byproducts, and then choosing precursors that it expects to avoid these reactions and instead favor the target’s formation. The effectiveness of this approach is showcased on three separate targets, for which optimal synthesis recipes are identified while requiring few experimental iterations.

The automation of data analysis and decision making, combined with robotics that can perform solid-state synthesis experiments, have made autonomous materials development possible. The integration of these tools into a platform known as the A-Lab is discussed in Chapter 5. Given a set of targeted materials screened using ab-initio computations, this lab can devise initial synthesis recipes based on historical data mined from the literature. It tests these recipes using robotics for automated powder handling and high-temperature annealing, followed by characterization with XRD. The resulting patterns are analyzed by ML models, which then feed into automated decision making to improve upon the initial recipes and achieve higher target yield. We demonstrate the capabilities of the A-Lab by using it to synthesize 41 materials in just 17 days of closed-loop experimentation.

The work reported herein demonstrates the feasibility of autonomous materials development while also highlighting areas that require further improvement. Several promising directions for future work are highlighted in Chapter 6. These include the development and integration of automated characterization to new techniques beyond XRD, the extension of robotic platforms to deal with air-sensitive samples and to measure device performance, and the generalization of decision- making algorithms to deal with experimental issues like melting and volatility.

    Cover page: Automating the Synthesis and Characterization of Inorganic Materials
    • Article
    • Peer Reviewed
    UC Berkeley Previously Published Works (2024)
    To bolster the accuracy of existing methods for automated phase identification from X-ray diffraction (XRD) patterns, we introduce a machine learning approach that uses a dual representation whereby XRD patterns are augmented with simulated pair distribution functions (PDFs). A convolutional neural network is trained directly on XRD patterns calculated using physics-informed data augmentation, which accounts for experimental artifacts such as lattice strain and crystallographic texture. A second network is trained on PDFs generated via Fourier transform of the augmented XRD patterns. At inference, these networks classify unknown samples by aggregating their predictions in a confidence-weighted sum. We show that such an integrated approach to phase identification provides enhanced accuracy by leveraging the benefits of each model’s input representation. Whereas networks trained on XRD patterns provide a reciprocal space representation and can effectively distinguish large diffraction peaks in multi-phase samples, networks trained on PDFs provide a real space representation and perform better when peaks with low intensity become important. These findings underscore the importance of using diverse input representations for machine learning models in materials science and point to new avenues for automating multi-modal characterization.
      Cover page: Integrated analysis of X-ray diffraction patterns and pair distribution functions for machine-learned phase identificationCreative Commons 'BY' version 4.0 license
      • Article
      • Peer Reviewed
      UC Berkeley Previously Published Works (2021)
      Autonomous experimentation driven by artificial intelligence (AI) provides an exciting opportunity to revolutionize inorganic materials discovery and development. Herein, we review recent progress in the design of self-driving laboratories, including robotics to automate materials synthesis and characterization, in conjunction with AI to interpret experimental outcomes and propose new experimental procedures. We focus on efforts to automate inorganic synthesis through solution-based routes, solid-state reactions, and thin film deposition. In each case, connections are made to relevant work in organic chemistry, where automation is more common. Characterization techniques are primarily discussed in the context of phase identification, as this task is critical to understand what products have formed during synthesis. The application of deep learning to analyze multivariate characterization data and perform phase identification is examined. To achieve "closed-loop" materials synthesis and design, we further provide a detailed overview of optimization algorithms that use active learning to rationally guide experimental iterations. Finally, we highlight several key opportunities and challenges for the future development of self-driving inorganic materials synthesis platforms.
        Cover page: Toward autonomous design and synthesis of novel inorganic materials
        • Article
        • Peer Reviewed
        UC Berkeley Previously Published Works (2021)
        Autonomous synthesis and characterization of inorganic materials require the automatic and accurate analysis of X-ray diffraction spectra. For this task, we designed a probabilistic deep learning algorithm to identify complex multi-phase mixtures. At the core of this algorithm lies an ensemble convolutional neural network trained on simulated diffraction spectra, which are systematically augmented with physics-informed perturbations to account for artifacts that can arise during experimental sample preparation and synthesis. Larger perturbations associated with off-stoichiometry are also captured by supplementing the training set with hypothetical solid solutions. Spectra containing mixtures of materials are analyzed with a newly developed branching algorithm that utilizes the probabilistic nature of the neural network to explore suspected mixtures and identify the set of phases that maximize confidence in the prediction. Our model is benchmarked on simulated and experimentally measured diffraction spectra, showing exceptional performance with accuracies exceeding those given by previously reported methods based on profile matching and deep learning. We envision that the algorithm presented here may be integrated in experimental workflows to facilitate the high-throughput and autonomous discovery of inorganic materials.
          Cover page: Probabilistic Deep Learning Approach to Automate the Interpretation of Multi-phase Diffraction Spectra
          • Article
          • Peer Reviewed
          UC Berkeley Previously Published Works (2023)
          Machine learning (ML) has become a valuable tool to assist and improve materials characterization, enabling automated interpretation of experimental results with techniques such as X-ray diffraction (XRD) and electron microscopy. Because ML models are fast once trained, there is a key opportunity to bring interpretation in-line with experiments and make on-the-fly decisions to achieve optimal measurement effectiveness, which creates broad opportunities for rapid learning and information extraction from experiments. Here, we demonstrate such a capability with the development of autonomous and adaptive XRD. By coupling an ML algorithm with a physical diffractometer, this method integrates diffraction and analysis such that early experimental information is leveraged to steer measurements toward features that improve the confidence of a model trained to identify crystalline phases. We validate the effectiveness of an adaptive approach by showing that ML-driven XRD can accurately detect trace amounts of materials in multi-phase mixtures with short measurement times. The improved speed of phase detection also enables in situ identification of short-lived intermediate phases formed during solid-state reactions using a standard in-house diffractometer. Our findings showcase the advantages of in-line ML for materials characterization and point to the possibility of more general approaches for adaptive experimentation.
            Cover page: Adaptively driven X-ray diffraction guided by machine learning for autonomous phase identificationCreative Commons 'BY' version 4.0 license
            • Article
            • Peer Reviewed
            UC Berkeley Previously Published Works (2024)
            Electron microscopy is indispensable for examining the morphology and composition of solid materials at the sub-micron scale. To study the powder samples that are widely used in materials development, scanning electron microscopes (SEMs) are increasingly used at the laboratory scale to generate large datasets with hundreds of images. Parsing these images to identify distinct particles and determine their morphology requires careful analysis, and automating this process remains challenging. In this work, we enhance the Mask R-CNN architecture to develop a method for automated segmentation of particles in SEM images. We address several challenges inherent to measurements, such as image blur and particle agglomeration. Moreover, our method accounts for prediction uncertainty when such issues prevent accurate segmentation of a particle. Recognizing that disparate length scales are often present in large datasets, we use this framework to create two models that are separately trained to handle images obtained at low or high magnification. By testing these models on a variety of inorganic samples, our approach to particle segmentation surpasses an established automated segmentation method and yields comparable results to the predictions of three domain experts, revealing comparable accuracy while requiring a fraction of the time. These findings highlight the potential of deep learning in advancing autonomous workflows for materials characterization.
              Cover page: Uncertainty-aware particle segmentation for electron microscopy at varied length scalesCreative Commons 'BY' version 4.0 license
              • Article
              • Peer Reviewed
              UC Berkeley Previously Published Works (2023)
              Solid-state synthesis plays an important role in the development of new materials and technologies. While in situ characterization and ab-initio computations have advanced our understanding of materials synthesis, experiments targeting new compounds often still require many different precursors and conditions to be tested. Here we introduce an algorithm (ARROWS3) designed to automate the selection of optimal precursors for solid-state materials synthesis. This algorithm actively learns from experimental outcomes to determine which precursors lead to unfavorable reactions that form highly stable intermediates, preventing the target material's formation. Based on this information, ARROWS3 proposes new experiments using precursors it predicts to avoid such intermediates, thereby retaining a larger thermodynamic driving force to form the target. We validate this approach on three experimental datasets, containing results from over 200 synthesis procedures. In comparison to black-box optimization, ARROWS3 identifies effective precursor sets for each target while requiring substantially fewer experimental iterations. These findings highlight the importance of domain knowledge in optimization algorithms for materials synthesis, which are critical for the development of fully autonomous research platforms.
                Cover page: Autonomous and dynamic precursor selection for solid-state materials synthesisCreative Commons 'BY' version 4.0 license
                • Article
                • Peer Reviewed
                Materials Sciences (2023)
                The effects of cation (Ti4+) and anion (O2−) substitution on the electrochemical properties of KVPO4F cathodes are investigated in this work. Both forms of substitution lead to smoothing of the associated voltage profile as well as reduced charging time at high voltage (>4.8 V vs. K/K+). These changes effectively suppress electrolyte decomposition, thereby enhancing the capacity retention in the ion-substituted KV(1 −x)TixPO4+yF1−y materials. Ionic substitution also enables the intercalation of excess K ions at a reasonably high voltage (∼1.8 V vs. K/K+) relative to that required in pure KVPO4F (<1.0 V vs. K/K+). Finally, this study reveals that detours in the synthesis reaction pathways can be created by using a pre-reacted precursor, VPO4, rather than the conventional precursors V2O3 and NH4H2PO4 to form stoichiometric (or non-substituted) KVPO4F. These results highlight several design criteria that can be used to optimize KVPO4F and related compositions to prepare K-ion batteries with high capacity, good cyclability, and fast rate capability.
                  Creative Commons 'BY' version 4.0 license
                  • Article
                  • Peer Reviewed
                  UC Berkeley Previously Published Works (2024)
                  With increasing battery demand comes a need for diversified Li sources beyond brines. Among all Li-bearing minerals, spodumene is most often used for its high Li content and natural abundance. However, the traditional approach to process spodumene is costly and energy-intensive, requiring the mineral be transformed from its natural α to β phase at >1000 °C. Acid leaching is then applied, followed by neutralization to precipitate Li2CO3. In this work, we report an alternative method to extract Li directly from α-spodumene, which is performed at lower temperatures and avoids the use of acids. It is shown that Li2CO3 is formed with >90% yield at 750 °C by reacting α-spodumene with Na2CO3 and Al2O3. The addition of Al2O3 is critical to reduce the amount of Li2SiO3 that forms when only Na2CO3 is used, instead providing increased thermodynamic driving force to form NaAlSiO4 and Li2CO3 as the sole products. We find that this reaction is most effective at 4 h, after which volatility limits the yield. Following its extraction, Li2CO3 can be isolated by washing the sample using deionized water. This energy-saving and acid-free route to obtain Li2CO3 directly from spodumene can help meet the growing demand for Li.
                    Cover page: Direct Lithium Extraction from α‑Spodumene through Solid-State Reactions for Sustainable Li2CO3 ProductionCreative Commons 'BY' version 4.0 license
                    • Article
                    • Peer Reviewed
                    UC Berkeley Previously Published Works (2024)
                    The success of solid-state synthesis often hinges on the first intermediate phase that forms, which determines the remaining driving force to produce the desired target material. Recent work suggests that when reaction energies are large, thermodynamics primarily dictates the initial product formed, regardless of reactant stoichiometry. Here, we validate this principle and quantify its constraints by performing in situ characterization on 37 pairs of reactants. These experiments reveal a threshold for thermodynamic control in solid-state reactions, whereby initial product formation can be predicted when its driving force exceeds that of all other competing phases by ≥60 milli-electron volt per atom. In contrast, when multiple phases have a comparable driving force to form, the initial product is more often determined by kinetic factors. Analysis of the Materials Project data shows that 15% of possible reactions fall within the regime of thermodynamic control, highlighting the opportunity to predict synthesis pathways from first principles.
                      Cover page: Quantifying the regime of thermodynamic control for solid-state reactions during ternary metal oxide synthesisCreative Commons 'BY' version 4.0 license
                      Top