Inverse material design, which seeks microstructures yielding desired macroscopic properties, is inherently ill-posed due to non-uniqueness and the existence of unattainable targets. While data-driven generative models offer powerful empirical solutions, they often lack rigorous theoretical guarantees. This paper establishes a formal probabilistic learning framework to address these challenges. We first prove that the forward homogenization map, which predicts effective properties from a microstructure descriptor field, is Frechet differentiable and locally Lipschitz continuous under physically reasonable assumptions. This foundational result justifies the use of gradient-based methods and sensitivity analysis. Building upon this, we demonstrate the probabilistic well-posedness of the Bayesian inverse problem: the posterior distribution over microstructures is well-defined and depends continuously on the target property data. Furthermore, we prove Bayesian consistency, showing that as observational uncertainty vanishes, the posterior measure concentrates on the true set of solutions. This theoretical foundation validates advanced probabilistic inference techniques (e.g., MCMC, variational inference) for robustly exploring the solution manifold and quantifying uncertainty, thereby enabling a more complete and practical exploration of the material design space.

Model-based disturbance-conscious design space exploration for drug substance flow synthesis using the Grignard reaction

Computational thermo-mechanical modelling and design-space exploration of cryogenic hydrogen tanks for aviation

Purpose Computational design synthesis (CDS) provides a systematic means to explore the design space of complex systems. However, the scope of exploration in many CDS studies is biased by limited parametrization, where component parameters remain fixed or arbitrarily assigned. This paper investigates the influence of configuration and parametric diversity on the effectiveness of design-space exploration in vehicle powertrain synthesis. Design/methodology/approach A simulation-driven CDS framework is developed with an autonomous control tuning mechanism integrated to ensure consistent evaluation of the synthesized topologies. Powertrain topologies are synthesized by randomly interconnecting port-compatible components and assigning parameters from predefined ranges. Two topology sets are generated, one emphasizing configuration diversity and another emphasizing parametric diversity, to analyze their impact on exploration. Findings The results indicate that parameter initialization strongly influences perceived performance. A topology’s parameter assignment defines its effective position in design space, and poor initialization can bias evaluation outcomes even with structural diversity. Research limitations/implications The framework is limited to sequential topologies, which makes it easier to explore the entire design space but also makes the results harder to generalize. Practical implications The groundwork established here impacts the development of generative topology synthesis frameworks, enabling autonomous generation, control tuning and simulation of systems. Originality/value The novelty is the autonomous control tuning integrated in a CDS workflow, alongside the investigation into the influence of configuration and parametric diversity on the effectiveness of design-space exploration.

As the size of a circuit increases, previous reinforcement learning (RL) approaches struggle to effectively explore the logic optimization sequences of large-scale Boolean networks due to the long runtime overhead with poor optimization results. This article proposes LOFMPL: an open-source logic optimization framework with Maximum Fanout-Free Cone (MFFC) based hypergraph partitioning and reinforcement learning. The novel two-stage MFFC-based hypergraph partitioning can divide the circuit into highly independent subnetworks, which can be explored by an enhanced parallel RL-based design space exploration engine with an improved objective function. The experiment is conducted based on more than 150 benchmarks with logic optimization and ASIC technology mapping tasks and compared with other ML-based and greedy methods. The different partitioning algorithms are also compared for the subsequent logic optimization. Experimental results demonstrate that the proposed partitioning algorithm significantly enhances optimization quality without greatly increasing partitioning time, outperforming the KaHypar algorithm. Additionally, for the logic optimization task, the proposed method achieves a node-level-product improvement of 13% over the RLG synthesis exploration technique, 3% over the ESE reinforcement learning framework, 14% over the Boils synthesis method, and 7% over the DRiLLS synthesis method, while delivering greater reductions in node count compared with the Bulls-Eye optimization technique. For the ASIC technology mapping task, the proposed method achieves an area-delay-product improvement of 23% over the LSOracle framework, 9% over the Boils synthesis method, and 5% over the DRiLLS synthesis method. Hence, LOFMPL can achieve better results within the same runtime constraints compared with state-of-the-art works.

The theoretical maximum efficiency of heterogeneous multi-core processors (HMPs) is fundamentally determined by their hardware designs. Nonetheless, the achieved efficiency of HMPs is significantly influenced by the strategy employed in resource management (RM). To alleviate the time-consuming nature of comprehensive design space exploration for HMPs, existing work typically follows a route that independently explores HMP designs and RM strategies. However, this route tends to restrict the overall exploration space and efficiency. In this work, we introduce HetDSE, a comprehensive design space exploration framework for HMP design that coordinately explores both HMP designs and RM strategies. We first identify requirements that enable comprehensive exploration, including high-quality design space, fast evaluation, and high-accuracy exploration. To meet these requirements, we propose representative-based design space generation, unified prediction-based evaluation, and iterative refinement exploration methods to achieve comprehensive exploration within a reasonable timeframe. Experimental results demonstrate that HetDSE reduces energy delay square product by 88% compared to the state-of-the-art framework.

Halide perovskites have attracted significant interest due to their potential in optoelectronic devices. However, challenges related to complex compositional spaces, environmental sensitivity, and stability limitations continue to constrain their systematic development and application. Machine learning (ML) has emerged as an effective tool to address these challenges by enabling the prediction of material properties, the identification of promising compositions, and optimization of processing conditions, while reducing reliance on conventional trial‐and‐error methods. By capturing complex, nonlinear relationships among compositional, structural, and processing parameters, ML enables the exploration of broad design spaces that are essential for advancing perovskite research. Additionally, ML accelerates the discovery and optimization of perovskite materials through data‐driven approaches, including high‐throughput screening and inverse design, enabling rapid identification of optimal compositions and processing conditions for enhanced device performance and stability. This review provides an overview of recent efforts to integrate ML into halide perovskite studies, discussing workflows, implementation strategies, and notable progress in device‐level development. This article highlights how ML enables systematic materials discovery and optimization, supporting the advancement of stable and efficient perovskite optoelectronic devices.

The rapid advancement of machine learning (ML) has ushered in a new era for materials science, particularly in the design and understanding of high-entropy alloys (HEAs). As a class of compositionally complex materials, HEAs have greatly benefited from the predictive power and computational efficiency of ML techniques. Recent years have witnessed remarkable expansion in the scope and sophistication of ML applications to HEAs, spanning from phase formation prediction to property and microstructure modeling. These developments have significantly accelerated the discovery and optimization of novel HEA systems. This review provides a comprehensive overview of the current progress and emerging trends in applying ML to HEA research. We first discuss phase prediction methodologies, encompassing both pure ML frameworks and hybrid physics-informed models. Subsequently, we summarize advances in ML-driven prediction of HEA properties and microstructural features. Further sections highlight the role of ML in exploring vast compositional spaces, guiding the design of high-performance HEAs, and optimizing existing alloys through data-driven algorithms. Finally, the challenges and limitations of current approaches are critically examined, and future directions are proposed toward interpretable models, mechanistic understanding, and efficient exploration of the HEA design space.

Multi-agent workflows have become an effective strategy for tackling complicated tasks by decomposing them into multiple sub-tasks and assigning them to specialized agents. However, designing optimal workflows remains challenging due to the vast and intricate design space. Current practices rely heavily on the intuition and expertise of practitioners, often resulting in design fixation or an unstructured, time-consuming exploration of trial-and-error. To address these challenges, this work introduces FLOWFORGE, an interactive visualization tool to facilitate the creation of multi-agent workflow through i) a structured visual exploration of the design space and ii) in-situ guidance informed by established design patterns. Based on formative studies and literature review, FLOWFORGE organizes the workflow design process into three hierarchical levels (i.e., task planning, agent assignment, and agent optimization), ranging from abstract to concrete. This structured visual exploration enables users to seamlessly move from high-level planning to detailed design decisions and implementations, while comparing alternative solutions across multiple performance metrics. Additionally, drawing from established workflow design patterns, FLOWFORGE provides context-aware, in-situ suggestions at each level as users navigate the design space, enhancing the workflow creation process with practical guidance. Use cases and user studies demonstrate the usability and effectiveness of FLOWFORGE, while also yielding valuable insights into how practitioners explore design spaces and leverage guidance during workflow development.

Cool-3D: An End-to-End Thermal-Aware Framework for Early-Phase Design Space Exploration of Microfluidic-Cooled 3DICs

Purpose This study aims to propose a systematic reinforcement strategy in large, thin-walled panels with an image-based optimization framework that maximizes stiffness gains while minimizing added mass. By leveraging image-based inputs and generative modeling techniques, specifically variational autoencoders (VAEs), the approach enables efficient exploration and optimization of high-dimensional design spaces. The goal is to identify the optimal placement and configuration of local directed energy deposition (DED) reinforcements, fulfilling industrial demands for lightweight, yet rigid, structures through an automated design pipeline. Design/methodology/approach A path-optimization framework integrating a VAE with pretrained convolutional neural networks (CNNs) was developed to enhance thin-walled structure stiffness in large-sized parts. Latent variables of VAEs serve as continuous design parameters; the decoder generates candidate reinforcement-path images; and CNNs evaluate to compute stiffness and mass objectives. An optimization loop adjusts the latent codes to minimize mass under stiffness constraints, enabling nonparametric discovery of optimal local reinforcement geometries. Findings The VAE-based method achieved a 43.48% increase in panel stiffness with only a 2.81% increase in mass. Beyond straight-line reinforcements, the VAE generated curved paths, internal voids, and mid-panel start/end points, novel geometries not present in the training data, superior material utilization and structural performance. These results demonstrate that VAEs are highly effective at modeling complex design spaces and hold strong potential for generative design applications. Originality/value This work pioneers combining generative deep learning with an additive manufacturing process freedom to optimize structural reinforcements. Unlike traditional parametric or rule-based methods, the VAE uncovers creative reinforcement designs tailored to complex constraints, leveraging DED’s local deposition capabilities. Additionally, representing the design space through the continuous latent dimensions offered by VAEs enables the exploration of a significantly larger and more diverse design space compared to previous studies. The result is a flexible, efficient design tool for automotive, aerospace and defense applications seeking innovative, lightweight, high-performance, thin-walled components.

Next-generation mobility systems—ranging from electric vertical takeoff and landing (eVTOL) vehicles to hypersonic aircraft and high-efficiency autonomous drones—require advanced aerothermal and structural optimization techniques to meet demands for safety, performance, and sustainability. Traditional computational approaches, such as finite element analysis, computational fluid dynamics (CFD), and multi-physics simulation, face increasing computational burdens due to the nonlinear interactions among thermal loads, aerodynamic behavior, material deformation, and structural integrity. This study investigates the role of quantum computing in accelerating and improving aerothermal and structural optimization workflows. Leveraging quantum algorithms, including quantum annealing, variational quantum eigensolvers (VQE), quantum approximate optimization algorithms (QAOA), and quantum-inspired multi-objective solvers, the proposed framework enhances the exploration of high-dimensional design spaces and improves the computational efficiency of complex optimization tasks. Early experiments on benchmark aerothermal–structural models demonstrate improved convergence rates, superior Pareto-optimal solutions, and enhanced predictive accuracy compared to classical methods. The findings show that quantum computing has the potential to fundamentally transform mobility system design by enabling faster iteration cycles, better thermal–structural co-optimization, and more energy-efficient architectures suitable for sustainable transportation futures.

The integration of renewable energy into power grids is challenged by intermittent energy production, necessitating efficient energy storage systems. Vanadium Redox Flow Batteries (VRFBs) offer significant advantages such as scalability, long lifespan, and independent scaling of power and energy capacities. Nevertheless, technical hurdles like vanadium crossover, electrolyte instability, and costly electrode materials hinder their widespread commercialization. Optimizing the porous electrode structure, as the core reactor components in flow batteries, is crucial to address some of the abovementioned challenges, as these structures greatly influence battery performance and cost-effectiveness.Previous research has explored material optimization to enhance energy and power density by engineering flow-field architecture and mass transport properties in porous electrodes. However, traditional approaches frequently face limitations such as restricted design spaces and tedious manual iterations. Recent progress in computational capabilities and machine learning (ML) techniques presents novel opportunities for inverse microstructural design. Specifically, ML approaches like deep neural networks (DNNs) and generative models (e.g., GANs, VAEs) enable rapid exploration of extensive and complex design spaces, revealing innovative electrode architectures often overlooked by conventional methods. Such inverse design approaches can be integrated into constantly evolving additive manufacturing techniques to fully leverage the high level of freedom and control they offer to fabricate high-performance electrodes with nearly arbitrarily complex porous structures.This study introduces a modular, data-driven computational framework integrating generative ML models with adjoint-based optimization to minimize power losses in VRFB porous electrodes. Our approach employs advanced generative models to develop optimized microscale unit cells tailored to local macroscale property distributions identified by the adjoint optimizer. Preliminary computational results show improved power efficiency compared to conventional uniform electrode designs, underscoring the potential of ML-assisted inverse design methodologies. This framework provides a flexible foundation for future research, potentially accelerating the design of high-performance, economically viable electrode structures for enhanced renewable energy storage.

Supercapacitors are at the forefront of energy storage innovation, excelling in power density, rapid charge-discharge cycles, and operational lifespan. Despite these advantages, their relatively low energy density and storage capacity remain significant barriers to widespread adoption and replacement of conventional energy storage systems. Addressing this challenge requires an efficient approach to explore and optimize complex electrode and electrolyte designs – an area where computational modeling proves indispensable. Computation methods enable systematic exploration of design spaces that are experimentally infeasible or costly to investigate, providing a deeper understanding of the interplay between material properties and electrochemical performance. This study introduces a diffuse-interface model, which offers a continuous representation of phase boundaries to simplify simulations of multiphase electrochemical systems. The model incorporates essential phenomena such as faradaic reactions and double-layer capacitance by employing indicator functions that delineate distinct phases within the system. By integrating key electrochemical equations such as Butler-Volmer kinetics, the model simulates charge distribution and electrostatic potential, with validation achieved through cyclic voltammetry (CV) curves.Building on this foundation, the model's capabilities were extended to optimize porous electrode microstructures using the phase-field framework. The Cahn–Hilliard equation was employed to simulate phase separation, enabling the generation of porous electrode structures. Supercapacitors with optimized microstructures demonstrated improved charge storage and power density, validating the proposed methodology's effectiveness in linking microstructural properties to electrochemical behaviour. This comprehensive analysis underscores the importance of tailoring electrode designs to achieve balanced performance metrics, including high energy density and power density. In addition to enhancing supercapacitor performance, this research provides a robust computational framework that can be adapted to study a wide range of energy storage devices. By incorporating the generation of electrode and electrolyte microstructures within the phase-field model, this approach facilitates a deeper understanding of the interplay between morphology and electrochemical processes. The diffuse-interface model offers significant adaptability, enabling its application in designing advanced supercapacitors and extending its utility to other battery technologies. The insights gained through this research lay the groundwork for the development of next-generation energy storage devices with unprecedented performance, contributing to a sustainable energy future.

With the rapid advancement of 2.5D FPGA technology, the integration of multiple FPGA dies enables larger design capacity and higher computing power. This progress provides a high-speed hardware platform well-suited for neural network acceleration. In this paper, we present a high-performance accelerator design for large-scale neural networks on 2.5D FPGAs. First, we propose a layer pipeline architecture that utilizes multiple accelerator cores, each equipped with individual high-bandwidth DDR memory. To address inter-die data dependencies, we introduce a block convolution mechanism that enables independent and efficient computation across dies. Furthermore, we propose a design space exploration scheme to optimize computational efficiency under resource constraints. Experimental results demonstrate that our proposed accelerator achieves 4860.87 GOPS when running VGG-16 on the Alveo U250 board, significantly outperforming existing layer pipeline designs on the same platform.

Abstract This paper proposes a structural optimization framework referred to as the ternary‐quantized gradient (TQG) method. Departing from the prevailing assumption of a fixed design variable dimension during the search, it performs integrated optimization of size, shape, and topology without pre‐specifying the dimension. The proposed method combines a single‐agent search scheme, zeroth‐order optimization, a Leaky ReLU‐based penalty function, and an additional exploration strategy to enable efficient and automated design space exploration through implicit topology control. The proposed method was validated through optimization of a truss cantilever and truss girder, representing stiffness‐ and strength‐governed structures, respectively. In both cases, TQG method successfully determined the optimal panel count while simultaneously optimizing size and shape, producing results comparable to those obtained by well‐known metaheuristic algorithms under predefined topology settings. The proposed method was applied to the early‐stage decision‐making process of high‐rise building design, optimizing panel count and configuration to efficiently resist lateral loads while satisfying serviceability constraints. These results demonstrate that the proposed TQG method can optimize the number of design variables through implicit topology control while achieving integrated optimization of size, shape, and topology in a single run, offering a practical and efficient approach for early‐stage structural design.

Unmanned vertical takeoff and landing (VTOL) aircraft are increasingly deployed for logistics, surveillance, and urban air mobility (UAM) applications. However, the limitations of full-electric (FE) and internal combustion engine (ICE) systems in meeting diverse mission requirements have motivated the development of hybrid-electric (HE) propulsion systems. The design of HE powertrains remains challenging due to configuration flexibility and the lack of unified criteria for performance trade-offs among FE, ICE-powered, and HE configurations. This study proposes an integrated propulsion co-design framework coupling power allocation, energy management, and component capacity constraints through parametric system modeling. These interdependencies are represented by three key matching parameters: the power ratio (Φ), energy ratio (Ω), and maximum continuous discharge rate (rc). Through Pareto-optimal design space exploration, trade-off analysis results and optimization principles are derived for diverse mission scenarios such as UAM, remote sensing, and military surveillance. Different technological conditions are considered to guide component-level technological advancements. The method was applied to the power system retrofit of the Great White eVTOL. Subsystem steady-state tests provided accurate design inputs, and a simulation model was developed to reproduce the full flight mission. By comparing the simulation with flight-test measurements, mean absolute percentage errors of 8.91% for instantaneous fuel consumption and 0.26% for battery voltage were obtained. Based on these error magnitudes, a dynamic design margin was defined and then incorporated into a subsequent re-optimization, which achieved the 1.5 h endurance target with a 10.49% increase in cost per ton-kilometer relative to the initial design. These results demonstrate that the proposed co-design methodology offers a scalable, data-driven foundation for early-stage hybrid-electric VTOL powertrain design, enabling iterative performance correction and supporting system optimization in subsequent design stages.

The convergence of quantum computing and artificial intelligence (AI) is redefining the boundaries of computational engineering, offering an unprecedented capacity for optimization, simulation, and design automation. This research introduces Quantum-Driven Computer-Aided Engineering (Q-CAE), a hybrid AI–quantum framework developed to accelerate design space exploration and enhance the precision of optimization in complex engineering systems. Q-CAE combines the analytical rigor of quantum algorithms with the adaptability of AI-driven optimization models to address the challenges of high-dimensional, nonlinear, and interdependent design parameters that traditionally strain classical computing resources. Within this framework, quantum subroutines such as Quantum Approximate Optimization Algorithms (QAOA) and Variational Quantum Circuits (VQC) are employed to efficiently search large solution spaces, while AI models leverage reinforcement learning and neural networks to predict optimal configurations and refine performance iteratively. This hybrid approach enables dynamic learning from multi-objective data sets, optimizing energy efficiency, material properties, and system reliability. Participants engaged with Q-CAE applications gained first-hand experience in algorithmic modeling, quantum heuristics, and AI-driven engineering design, cultivating interdisciplinary skills essential for future computational innovation. Results indicate that hybrid Q-CAE systems can significantly reduce computational time and enhance accuracy compared to traditional CAE workflows. The study demonstrates how integrating quantum principles into engineering design pipelines can transform modern computational methodologies, setting the foundation for autonomous, intelligent, and quantum-enhanced engineering systems.

The aerodynamic performance of racing motorcycles plays a crucial role in improving speed, stability, and rider control under dynamic conditions. While most existing studies focus on front-mounted winglets and fairing extensions, the aerodynamic role of rear fairing appendages remains comparatively unexplored despite their potential influence on drag, downforce distribution, and wake behaviour. In this work, three alternative rear winglet configurations were parametrically designed in Siemens NX and systematically evaluated within a validated CFD framework based on Simcenter STAR-CCM+, with the aim of assessing how geometric variations influence aerodynamic performance and achieve a favourable trade-off between reduced aerodynamic resistance and enhanced rear downforce. The numerical setup employed has been previously validated against wind-tunnel measurements in similar aerodynamic applications, ensuring the reliability and accuracy of the predicted flow fields. A Design Space Exploration (DSE) was performed through an automated multi-software workflow, enabling systematic variation in key geometric parameters and real-time assessment of their aerodynamic effects. The study revealed distinct influences of the different configurations on drag and lift coefficients, as well as on wake structure and flow detachment, highlighting the critical aerodynamic mechanisms governing rear stability and flow closure. Through iterative design and simulation, the workflow identified the most effective configuration, achieving a balance between reduced aerodynamic resistance and increased downforce, both essential for competitive racing performance. The results demonstrate the potential of integrating parametric modelling, automated CFD simulation, and DSE optimization in the aerodynamic design phase. This methodology not only offers new insights into the scarcely studied rear aerodynamic region of racing motorcycles but also establishes a replicable framework for future developments involving advanced optimization algorithms, experimental validation, and wake-interaction analyses between leading and trailing riders.

This paper presents a computationally efficient approach for the design of acoustic duct filters composed of axisymmetric step discontinuities. A rigorous approach based on the mode-matching technique and the building block method is employed to model sound propagation and compute transmission loss across a cascade of circular duct segments. Unlike existing studies relying on empirical approximations or simulations, the proposed method retains wave-theoretical accuracy while enabling fast evaluations suitable for iterative optimization. In optimization, some global metaheuristic algorithms are coupled with a local search phase to ensure both broad design space exploration and fine-tuned convergence near local optima. The methodology is validated through the design of some filters for frequency bands of practical importance.