Tonio Buonassisi

Abstract Bayesian optimization (BO) holds promise for accelerating materials science research; however, it faces challenges with high-dimensional inputs and experimental noise in real-world problems. This study addresses these issues by benchmarking batch BO on two synthetic six-variable optimization tasks at varying noise levels: a needle-in-a-haystack task (Ackley function) representing rare materials properties, and a smooth landscape (Hartmann function) simulating process optimization. We evaluate key BO strategies, including acquisition functions, batch-picking methods, and exploration hyperparameter tuning, while presenting a framework for tracking high-dimensional optimization progress. Results show optimization outcomes are highly sensitive to noise levels and landscape shapes. This information enables the design of robust materials optimization campaigns with pre-planned experimental budgets that account for real-world uncertainties. Our methodology facilitates greater BO utilization in experimental materials research, particularly for multi-variable optimization problems, by providing practical guidance for configuring BO campaigns in challenging scientific applications. Graphical abstract

Scientific hypothesis generation is central to materials discovery, yet current approaches often emphasize either conceptual (idea-to-data) reasoning or data-driven (data-to-idea) analysis, rarely achieving an effective integration of both. Here, we present a generalizable active learning workflow that integrates top-down, theory-driven hypothesis generation, guided by a large language model. This is complemented by bottom-up, data-driven hypothesis testing through a root-cause association study. We demonstrate this approach through the design of equimolar quinary-cation two-dimensional perovskite, a chemically complex system with over 850,000 possible cation combinations. In the top-down component, the large language model drives closed-loop optimization by proposing candidates that are likely to achieve phase purity, leveraging domain knowledge and chain-of-thought reasoning. With each iteration, the model identifies an increasing number of near phase-pure compositions, sampling less than 0.004% of the design space. In parallel, the bottom-up association study identifies molecular features with statistically significant influences on phase purity. The integration of these approaches enables the convergence of conceptual and statistical hypotheses, leading to generalizable and rational design rules for phase-pure quinary-cation two-dimensional perovskites. As a proof of concept, we applied the optimized phase-pure quinary-cation two-dimensional perovskite film as a surface capping layer in perovskite solar cells, achieving good performance and stability. Our framework enables the development of interpretable and generalizable design rules that are applicable to a wide range of optimization processes within complex design spaces, providing a foundational strategy for rational, scalable, and efficient materials discovery.

Abstract To realize the full promise of high-throughput experimental workflows, the rate of sample synthesis must be matched by that of characterization. Of growing interest are contactless optical techniques that can rapidly measure material homogeneity and properties. Here, we present a hyperspectral imaging method to measure local optical bandgap distributions within samples, utilizing spatially-resolved reflectance spectra coupled with automated data analysis. We collect approximately one million optical bandgap data across the compositional space of Cs 3 (Bi x Sb 1- x ) 2 (Br y I 1- y ) 9 perovskite-inspired materials. Our results show non-monotonic bandgap variations (i.e., bandgap bowing) along six composition gradient sequences, in addition to identifying samples with multiple bandgaps in statistics. High-throughput transient absorption spectroscopy reveals that within these compositions, the depletion of the ground state carriers to excited states occurred at discrete energy levels with independent carrier dynamics, consistent with the bandgap observation and indicative of phase separation. This work demonstrates the potential for rapid optical measurements to assess material quality and homogeneity in a high-throughput experimental setting, supporting screening and recipe optimization of optoelectronic material candidates with desired carrier dynamics and optical properties.

In the evolving landscape of scientific research, the complexity of global challenges demands innovative approaches to experimental planning and execution. Self-Driving Laboratories (SDLs) automate experimental tasks in chemical and materials sciences and the design and selection of experiments to optimize research processes and reduce material usage. This perspective explores improving access to SDLs via centralized facilities and distributed networks. We discuss the technical and collaborative challenges in realizing SDLs’ potential to enhance human–machine and human–human collaboration, ultimately fostering a more inclusive research community and facilitating previously untenable research projects. Collaborative self-diving research is crucial to research acceleration amidst ever more complex problems. Here, authors identify the key challenges to the dual cultivation of centralised self-driving user facilities and networks of self-driving labs.

Recent advancements in machine learning (ML) for materials have demonstrated that "simple" materials representations (e.g., the chemical formula alone without structural information) can sometimes achieve competitive property prediction performance in common-tasks. Our physics-based intuition would suggest that such representations are "incomplete", which indicates a gap in our understanding. This work proposes a tomographic interpretation of structure-property relations of materials to bridge that gap by defining what is a material representation, material properties, the material and the relationships between these three concepts using ideas from information theory. We verify this framework performing an exhaustive comparison of property-augmented representations on a range of material's property prediction objectives, providing insight into how different properties can encode complementary information.

Archerfish is a low-cost, high-throughput tool for combinatorial materials research. Retrofitted with in situ mixing, Archerfish prints 250 unique compositions per min—a 100× acceleration factor—for aqueous, nanoparticle, and crystalline materials.

The maturation of 3D printing technology has enabled low-cost, rapid prototyping capabilities for mainstreaming accelerated product design. The materials research community has recognized this need, but no universally accepted rapid prototyping technique currently exists for material design. Toward this end, we develop Archerfish, a 3D printer retrofitted to dispense liquid with in-situ mixing capabilities for performing high-throughput combinatorial printing (HTCP) of material compositions. Using this HTCP design, we demonstrate continuous printing throughputs of up to 250 unique compositions per minute, 100x faster than similar tools such as OpenTrons that utilize stepwise printing with ex-situ mixing. We validate the formation of these combinatorial "prototype" material gradients using hyperspectral image analysis and energy-dispersive X-ray spectroscopy. Furthermore, we describe hardware challenges to realizing reproducible, accurate, and precise composition gradients with continuous printing, including those related to precursor dispensing, mixing, and deposition. Despite these limitations, the continuous printing and low-cost design of Archerfish demonstrate promising accelerated materials screening results across a range of materials systems from nanoparticles to perovskites.

We propose to address the challenge of the vast parameter space associated with perovskite manufacturing optimization, by developing a machine learning (ML)-assisted optimization framework for a scalable perovskite PV manufacturing tool. This framework will be interpretable, sequential, and rapidly adaptable to upgraded systems (e.g., via transfer learning). The tool is an open-air rapid spray plasma process (RSPP) of perovskite films, which has already been established at Stanford and is a unique platform to test and deploy the proposed ML-guided framework because the RSPP technique is able to conduct optimization experiments with a high throughput, and easily adjust a wide range of process variables.

The maturation of 3D printing technology has enabled low-cost, rapid prototyping capabilities for mainstreaming accelerated product design. The materials research community has recognized this need, but no universally accepted rapid prototyping technique currently exists for material design. Toward this end, we develop Archerfish, a 3D printer retrofitted to dispense liquid with in-situ mixing capabilities for performing high-throughput combinatorial printing (HTCP) of material compositions. Using this HTCP design, we demonstrate continuous printing throughputs of up to 250 unique compositions per minute, 100x faster than similar tools such as OpenTrons that utilize stepwise printing with ex-situ mixing. We validate the formation of these combinatorial "prototype" material gradients using hyperspectral image analysis and energy-dispersive X-ray spectroscopy. Furthermore, we describe hardware challenges to realizing reproducible, accurate, and precise composition gradients with continuous printing, including those related to precursor dispensing, mixing, and deposition. Despite these limitations, the continuous printing and low-cost design of Archerfish demonstrate promising accelerated materials screening results across a range of materials systems from nanoparticles to perovskites.

Recent advances in machine learning techniques have made it possible to use high-throughput screening to identify novel materials with specific properties. However, the large number of potential candidates produced by these techniques can make it difficult to select the most promising ones. In this study, we develop the oxidation state probability (OSP) method which evaluates ternary compounds based on the probability (the OSP metric) of each element to adopt the required oxidation states for fulfilling charge neutrality. We compare this model with Roost and the Fourier-transformed crystal properties (FTCP)-based synthesizability score. Among the top 1000 systems with the most database entries in Materials Project (MP), more than 500 systems exhibit an attested compound among the top 3 compositions when ranked by the OSP metric. We find that the OSP method shows promising results for certain classes of ternary systems, especially those containing nonmetals, s-block, or transition metals. When applied to the Cu-In-Te ternary system, an interesting system for thermoelectric applications, the OSP method predicted the synthesizability of CuIn$_3$Te$_5$ without prior knowledge, and we have successfully synthesized CuIn$_3$Te$_5$ in experiment. Our method has the potential to accelerate the discovery of novel compounds by providing a guide for experimentalists to easily select the most synthesizable candidates from an arbitrarily large set of possible chemical compositions.

Abstract Globally, 1.6 billion people in rural regions face water scarcity. Expanding freshwater access via brackish groundwater desalination can provide additional resources to address this challenge. In this study, we have developed a time-variant electrodialysis reversal (EDR) technology that flexibly uses available solar energy for desalination. Our proposed photovoltaic-powered desalination system can vary pumping and EDR power to match the availability of intermittent solar power, maximizing the desalination rate. Our results show improved system performance with the direct use of 77% of available solar energy—91% more than in conventional systems—and a 92% reduction in battery reliance. In a village-scale desalination case study in India, these system improvements lead to a 22% reduction in water cost, making the technology competitive with the currently used on-grid, village-scale reverse osmosis systems that are mainly powered by fossil fuels. Future advances could further reduce costs, providing an improved, sustainable solution to water scarcity in remote areas.

Hardware-based startups risk having longer times-to-market, deterring investment in the clean technologies that are critical to a sustainable future. We interviewed 55 leaders at hardware startups, 20 of which are cleantech, mapped their development timelines, and found prototyping to be the longest development step (median of 19 weeks per prototype) regardless of prototype complexity or iteration. Qualitative interview analysis reveals the prototyping team’s choice of development style is a major factor affecting timeline. We define two development styles: natural and structured, typified by free-form exploration and rule-based execution, respectively. On average, natural development takes 35% less time than structured, and is thus preferred for early iterations, but adopting structure at strategic points is needed for timely commercialization. Critical points of transition to a structured style include adding new team members or engaging external partners, which demand clear communication and expectations. When pivoting to a new product or market, returning to a natural style is beneficial.

Correction for ‘Tackling data scarcity with transfer learning: a case study of thickness characterization from optical spectra of perovskite thin films’ by Siyu Isaac Parker Tian et al. , Digital Discovery , 2023, 2 , 1334–1346, https://doi.org/10.1039/D2DD00149G.

Fusion energy is at an important inflection point in its development: multiple government agencies and private companies are now planning fusion pilot plants to deliver electricity to the grid in the next decade. However, realizing fusion as a technically and economically viable energy source depends on developing and qualifying materials that can withstand the extreme environment inside a fusion power plant. This Perspective seeks to engage the broader materials science community in this long-term effort. We first outline the principal materials challenges and research opportunities for fusion. Next, we argue that fusion is distinct from other energy applications with respect to materials, not just in the magnitude and complexity of the technical challenges but also in the present level of uncertainty in materials design requirements. To address this, we finally propose a research framework based on an iterative co-evolution of materials science and fusion power plant design requirements.

Abstract Solar cells are a critical component of renewable energy systems and are becoming increasingly important as society moves towards decarbonization. To assess their economic viability and sustainability, developing solar technologies with desired performance levels is an important consideration. Among various material parameters, carrier lifetime is one of the key material properties which influences the performance metrics of solar cells significantly. Measurement of carrier lifetime using experimental methods is a complex and tedious procedure, hence machine learning (ML) methods can be used to accelerate the prediction of carrier lifetime from the Current density-voltage (J-V) characteristics. However, ML methods generally require a large amount of data to achieve accurate predictions. To this end, looking for an efficient solution to learn transferable knowledge for the carrier lifetime prediction of different solar cell materials with less data is of utmost significance. In this paper, we propose a transfer learning (TL) framework to learn common knowledge between Silicon (Si), Gallium arsenide (GaAs), and perovskite solar cells (PSC) for carrier lifetime prediction. Our TL framework uses deep neural network (DNN) architecture with two methods, i) pre-training and fine-tuning entire layers, and ii) pre-training and fine-tuning regressor layer alone. Hence, these methods facilitate data-efficient and parameter-efficient learning. In this way, one can improve the accuracy of carrier lifetime prediction by learning with the source material data and fine-tuning with less data for the material of interest (target material data). The experimental results on simulated data and experimental data indicate that TL has autonomously identified nontrivial transferability across different materials. This leads to faster convergence and more robustness while training a DNN model and higher accuracy for carrier lifetime prediction.

<title>Abstract</title> High-throughput materials synthesis methods have risen in popularity due to their potential to accelerate the design and discovery of novel functional materials, such as solution-processed semiconductors. After synthesis, key material properties must be measured and characterized to validate discovery and provide feedback to optimization cycles. However, with the boom in development of high-throughput synthesis tools that champion production rates up to 10⁴ samples per hour with flexible form factors, most sample characterization methods are either slow (conventional rates of 10¹ samples per hour, approximately 1000x slower) or rigid (<italic>e.g.</italic>, designed for standard-size microplates), resulting in a bottleneck that impedes the materials-design process. To overcome this challenge, we propose a set of automated material property characterization (autocharacterization) tools that leverage the adaptive, parallelizable, and scalable nature of computer vision to accelerate the throughput of characterization by 85x compared to the non-automated workflow. We demonstrate a generalizable composition mapping tool for high-throughput synthesized binary material systems as well as two scalable autocharacterization algorithms that (1) autonomously compute the band gap of 200 unique compositions in 6 minutes and (2) autonomously compute the degree of degradation in 200 unique compositions in 20 minutes, generating ultra-high compositional resolution trends of band gap and stability. We demonstrate that the developed band gap and degradation detection autocharacterization methods achieve 98.5% accuracy and 96.9% accuracy, respectively, on the FA_1-xMA_xPbI₃, 0 ≤ x ≤ 1 perovskite semiconductor system.

Exceptional molecules and materials with one or more extraordinary properties are both technologically valuable and fundamentally interesting, because they often involve new physical phenomena or new compositions that defy expectations. Historically, exceptionality has been achieved through serendipity, but recently, machine learning (ML) and automated experimentation have been widely proposed to accelerate target identification and synthesis planning. In this Perspective, we argue that the data-driven methods commonly used today are well-suited for optimization but not for the realization of new exceptional materials or molecules. Finding such outliers should be possible using ML, but only by shifting away from using traditional ML approaches that tweak the composition, crystal structure, or reaction pathway. We highlight case studies of high- T c oxide superconductors and superhard materials to demonstrate the challenges of ML-guided discovery and discuss the limitations of automation for this task. We then provide six recommendations for the development of ML methods capable of exceptional materials discovery: (i) Avoid the tyranny of the middle and focus on extrema; (ii) When data are limited, qualitative predictions that provide direction are more valuable than interpolative accuracy; (iii) Sample what can be made and how to make it and defer optimization; (iv) Create room (and look) for the unexpected while pursuing your goal; (v) Try to fill-in-the-blanks of input and output space; (vi) Do not confuse human understanding with model interpretability. We conclude with a description of how these recommendations can be integrated into automated discovery workflows, which should enable the discovery of exceptional molecules and materials.

Exceptional molecules and materials with one or more extraordinary properties are both technologically valuable and fundamentally interesting because they often involve new physical phenomena or new compositions that defy expectations. Historically, exceptionality has been achieved through serendipity, but recently, machine learning (ML) and automated experimentation have been widely proposed to accelerate target identification and synthesis planning. In this Perspective, we argue that the data-driven methods commonly used today are well-suited for optimization but not for realizing new exceptional materials or molecules. Finding such outliers should be possible using ML, but only by shifting away from using traditional ML approaches that tweak the composition, crystal structure, or reaction pathway. We highlight case studies of high-Tc oxide superconductors and superhard materials to demonstrate the challenges of ML-guided discovery and discuss the limitations of automation for this task. We then provide six recommendations for the development of ML methods capable of exceptional materials discovery: (i) Avoid the tyranny of the middle and focus on extrema; (ii) When data is limited, qualitative predictions that provide direction are more valuable than interpolative accuracy; (iii) Sample what can be made and how to make it, and defer optimization; (iv) Create room (and look) for the unexpected while pursuing your goal; (v) Try to fill-in-the-blanks of input and output space; (vi) Do not confuse human understanding with model interpretability. We conclude with a description of how these recommendations can be integrated into automated discovery workflows that should enable the discovery of exceptional molecules and materials.

We present a new dataset for learning to solve, explain, and generate university-level STEM questions from 27 courses across a dozen departments in seven universities. We scale up previous approaches to questions from courses in the departments of Mechanical Engineering, Materials Science and Engineering, Chemistry, Electrical Engineering, Computer Science, Physics, Earth Atmospheric and Planetary Sciences, Economics, Mathematics, Biological Engineering, Data Systems, and Society, and Statistics. We visualize similarities and differences between questions across courses. We demonstrate that a large foundation model is able to generate questions that are as appropriate and at the same difficulty level as human-written questions.

We curate a comprehensive dataset of 4,550 questions and solutions from problem sets, midterm exams, and final exams across all MIT Mathematics and Electrical Engineering and Computer Science (EECS) courses required for obtaining a degree. We evaluate the ability of large language models to fulfill the graduation requirements for any MIT major in Mathematics and EECS. Our results demonstrate that GPT-3.5 successfully solves a third of the entire MIT curriculum, while GPT-4, with prompt engineering, achieves a perfect solve rate on a test set excluding questions based on images. We fine-tune an open-source large language model on this dataset. We employ GPT-4 to automatically grade model responses, providing a detailed performance breakdown by course, question, and answer type. By embedding questions in a low-dimensional space, we explore the relationships between questions, topics, and classes and discover which questions and classes are required for solving other questions and classes through few-shot learning. Our analysis offers valuable insights into course prerequisites and curriculum design, highlighting language models' potential for learning and improving Mathematics and EECS education.

Exceptional molecules and materials with one (or more) extraordinary properties are both technologically valuable and fundamentally interesting because they often involve new physical phenomena or new compositions that defy expectations. Historically, exceptionality has been achieved through serendipity, but recently, machine learning (ML) and automated experimentation have been widely proposed to accelerate target identification and synthesis planning. In this Perspective, we argue that the data-driven methods commonly used today are well-suited for optimization but not for realizing new exceptional materials or molecules. Finding such outliers should be possible using ML, but only by shifting away from using traditional ML approaches that tweak the composition, crystal structure, or reaction pathway. We highlight case studies of high-Tc superconductors and superhard materials to demonstrate the challenges of ML-guided discovery and discuss the limitations of automation for this task. We then provide six recommendations for the development of ML methods capable of exceptional materials discovery: (i) Avoid the tyranny of the middle and focus on extrema; (ii) When data is limited, qualitative predictions that provide direction are more valuable than interpolative accuracy; (iii) Sample what can be made and how to make it, and defer optimization; (iv) Create room (and look) for the unexpected while pursuing your goal; (v) Try to fill-in-the-blanks of input and output space; (vi) Do not confuse human understanding with model interpretability. We conclude with a description of how these recommendations can be integrated into automated discovery workflows that should enable the discovery of exceptional molecules and materials.

High-throughput materials synthesis methods have risen in popularity due to their potential to accelerate the design and discovery of novel functional materials, such as solution-processed semiconductors. After synthesis, key material properties must be measured and characterized to validate discovery and provide feedback to optimization cycles. However, with the boom in development of high-throughput synthesis tools that champion production rates up to $10^4$ samples per hour with flexible form factors, most sample characterization methods are either slow (conventional rates of $10^1$ samples per hour, approximately 1000x slower) or rigid (e.g., designed for standard-size microplates), resulting in a bottleneck that impedes the materials-design process. To overcome this challenge, we propose a set of automated material property characterization (autocharacterization) tools that leverage the adaptive, parallelizable, and scalable nature of computer vision to accelerate the throughput of characterization by 85x compared to the non-automated workflow. We demonstrate a generalizable composition mapping tool for high-throughput synthesized binary material systems as well as two scalable autocharacterization algorithms that (1) autonomously compute the band gap of 200 unique compositions in 6 minutes and (2) autonomously compute the degree of degradation in 200 unique compositions in 20 minutes, generating ultra-high compositional resolution trends of band gap and stability. We demonstrate that the developed band gap and degradation detection autocharacterization methods achieve 98.5% accuracy and 96.9% accuracy, respectively, on the FA$_{1-x}$MA$_{x}$PbI$_3$, $0\leq x \leq 1$ perovskite semiconductor system.

Defining the metric for synthesizability and predicting new compounds that can be experimentally realized in the realm of data-driven research is a pressing problem in contemporary materials science. The increasing computational power and advancements in machine learning (ML) algorithms provide a new avenue to solve the synthesizability challenge. In this work, using the Inorganic Crystal Structure Database (ICSD) and the Materials Project (MP) database, we represent crystal structures in Fourier-transformed crystal properties (FTCP) representation and use a deep learning model to predict synthesizability in the form of a synthesizability score (SC). Such an SC model, as a synthesizability filter for new materials, enables an efficient and accurate classification to identify promising material candidates. The SC prediction model achieved 82.6/80.6% (precision/recall) overall accuracy in predicting ternary crystal materials. We also trained the SC model by only considering compounds uploaded on the MP before 2015 as the training set and testing on multiple sets of materials uploaded after 2015. In the post-2019 test set, we obtain a high 88.60% true positive rate accuracy, coupled with 9.81% precision, indicating that newly added materials remain unexplored and have high synthesis potential. Further, we provide a list of 100 materials predicted to be synthesizable from this post-2019 dataset (highest SC) for future studies, and our SC model, as a validation filter, is beneficial for future material screening and discovery.

ADVERTISEMENT RETURN TO ISSUEPREVCommentsNEXTReply to Comment on "Environmental Stability of Crystals: A Greedy Screening"Nicholas M. TwymanNicholas M. TwymanDepartment of Materials, Imperial College London, London SW7 2AZ, United KingdomPhotovaltaic Research Laboratory, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, United StatesMore by Nicholas M. Twyman, Aron Walsh*Aron WalshDepartment of Materials, Imperial College London, London SW7 2AZ, United Kingdom*Email: [email protected]More by Aron Walshhttps://orcid.org/0000-0001-5460-7033, and Tonio Buonassisi*Tonio BuonassisiPhotovaltaic Research Laboratory, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, United States*Email: [email protected]More by Tonio Buonassisihttps://orcid.org/0000-0001-8345-4937Cite this: Chem. Mater. 2023, 35, 2, 804Publication Date (Web):January 11, 2023Publication History Received5 December 2022Published online11 January 2023Published inissue 24 January 2023https://pubs.acs.org/doi/10.1021/acs.chemmater.2c03627https://doi.org/10.1021/acs.chemmater.2c03627article-commentaryACS PublicationsCopyright © 2023 American Chemical Society. This publication is available under these Terms of Use. Request reuse permissions This publication is free to access through this site. Learn MoreArticle Views933Altmetric-Citations-LEARN ABOUT THESE METRICSArticle Views are the COUNTER-compliant sum of full text article downloads since November 2008 (both PDF and HTML) across all institutions and individuals. These metrics are regularly updated to reflect usage leading up to the last few days.Citations are the number of other articles citing this article, calculated by Crossref and updated daily. Find more information about Crossref citation counts.The Altmetric Attention Score is a quantitative measure of the attention that a research article has received online. Clicking on the donut icon will load a page at altmetric.com with additional details about the score and the social media presence for the given article. Find more information on the Altmetric Attention Score and how the score is calculated. Share Add toView InAdd Full Text with ReferenceAdd Description ExportRISCitationCitation and abstractCitation and referencesMore Options Share onFacebookTwitterWechatLinked InRedditEmail PDF (837 KB) Get e-AlertscloseSUBJECTS:Algorithms,Biological databases,Materials,Organic reactions,Stability Get e-Alerts

thicknessML predicts film thickness from reflection and transmission spectra. Transfer learning enables thickness prediction of different materials with good performance. Transfer learning also bridges the gap between simulation and experiment.

Designs for an open source environmental chamber for stability testing of metal halide perovskites and other materials using optical degradation fingerprints. The design suite can accommodate bulk samples, thin films or full photovoltaic devices.

Datasets used in the study named <em>Tackling Data Scarcity with Transfer Learning: A Case Study of Thickness Characterization from Optical Spectra of Perovskite Thin Films.</em>