arXiv subjects

Dilip Krishnamurthy

Publications and source records attributed to Dilip Krishnamurthy.

9 recordsLinked to original sources

arXiv2026

Transforming Computational Lithography with AC and AI -- Faster, More Accurate, and Energy-efficient

From climate science to drug discovery, scientific computing demands have surged dramatically in recent years -- driven by larger datasets, more sophisticated models, and higher simulation fidelity. This growth rate far outpaces transistor scaling, leading to unsustainably rising costs, energy consumption, and emissions. Semiconductor manufacturing is no exception. Computational lithography -- involving transferring circuitry to silicon in diffraction-limited conditions -- is the largest workload in semiconductor manufacturing. It has also grown exceptionally complex as miniaturization has advanced in the angstrom-era, requiring more accurate modeling, intricate corrections, and broader solution-space exploration. Accelerated computing (AC) offers a solution by dramatically freeing up the compute and power envelope. AI augments these gains by serving as high-fidelity surrogates for compute-intensive steps. Together, they present a sustainable, next-generation computing platform for scientific workloads. This new paradigm needs a fundamental redesign of the software stack. For computational lithography, NVIDIA cuLitho reinvents the core primitives -- diffractive optics, computational geometry, multi-variant optimization, data processing -- to achieve a transformative 57X end-to-end acceleration. Beyond dramatically faster cycles, this expanded compute envelope enables more rigorous solutions, including curvilinear masks, high-numerical aperture extreme ultraviolet (high-NA EUV) lithography, and subatomic modeling. We reinvest a small fraction of the freed-up compute to include through-focus correction for better process resilience. Silicon experiments at IMEC show significant benefits compared to conventional methods -- 35% better process window and 19% better edge placement error. This is the first quantified chip-scale demonstration of the lithography benefits of AC and AI in silicon.

eess.SP↗

arXiv2021

Validation and parameterization of a novel physics-constrained neural dynamics model applied to turbulent fluid flow

In fluid physics, data-driven models to enhance or accelerate solution methods are becoming increasingly popular for many application domains, such as alternatives to turbulence closures, system surrogates, or for new physics discovery. In the context of reduced order models of high-dimensional time-dependent fluid systems, machine learning methods grant the benefit of automated learning from data, but the burden of a model lies on its reduced-order representation of both the fluid state and physical dynamics. In this work, we build a physics-constrained, data-driven reduced order model for the Navier-Stokes equations to approximate spatio-temporal turbulent fluid dynamics. The model design choices mimic numerical and physical constraints by, for example, implicitly enforcing the incompressibility constraint and utilizing continuous Neural Ordinary Differential Equations for tracking the evolution of the differential equation. We demonstrate this technique on three-dimensional, moderate Reynolds number turbulent fluid flow. In assessing the statistical quality and characteristics of the machine-learned model through rigorous diagnostic tests, we find that our model is capable of reconstructing the dynamics of the flow over large integral timescales, favoring accuracy at the larger length scales. More significantly, comprehensive diagnostics suggest that physically-interpretable model parameters, corresponding to the representations of the fluid state and dynamics, have attributable and quantifiable impact on the quality of the model predictions and computational complexity.

physics.flu-dyn↗

arXiv2021

Differentiable Physics: A Position Piece

Differentiable physics provides a new approach for modeling and understanding the physical systems by pairing the new technology of differentiable programming with classical numerical methods for physical simulation. We survey the rapidly growing literature of differentiable physics techniques and highlight methods for parameter estimation, learning representations, solving differential equations, and developing what we call scientific foundation models using data and inductive priors. We argue that differentiable physics offers a new paradigm for modeling physical phenomena by combining classical analytic solutions with numerical methodology using the bridge of differentiable programming.

cs.LG↗

arXiv2020

Closed-Loop Design of Proton Donors for Lithium-Mediated Ammonia Synthesis with Interpretable Models and Molecular Machine Learning

In this work, we experimentally determined the efficacy of several classes of proton donors for lithium-mediated electrochemical nitrogen reduction in a tetrahydrofuran-based electrolyte, an attractive alternative method for producing ammonia. We then built an interpretable data-driven classification model which identified solvatochromic Kamlet-Taft parameters as important for distinguishing between active and inactive proton donors. After curating a dataset for the Kamlet-Taft parameters, we trained a deep learning model to predict the Kamlet-Taft parameters. The combination of classification model and deep learning model provides a predictive mapping from a given proton donor to the ability to produce ammonia. We demonstrate that this combination of classification model with deep learning is superior to a purely mechanistic or data-driven approach in accuracy and experimental data efficiency.

physics.chem-ph↗

arXiv2019

Engineering Three-Dimensional (3D) Out-of-Plane Graphene Edge Sites for Highly-Selective Two-Electron Oxygen Reduction Electrocatalysis

Selective two-electron oxygen reduction reaction (ORR) offers a promising route for hydrogen peroxide synthesis, and defective sp2 carbon-based materials are attractive, low-cost electrocatalysts for this process. However, due to a wide range of possible defect structures formed during material synthesis, identification and fabrication of precise active sites remain a challenge. Here, we report a graphene edgebased electrocatalyst for two-electron ORR - nanowire-templated three-dimensional fuzzy graphene (NT3DFG). NT-3DFG exhibits excellent efficiency (onset potential of 0.79 $\pm$ 0.01 V versus RHE), selectivity (93 $\pm$ 3 % H2O2), and tunable ORR activity as a function of graphene edge site density. Using spectroscopic surface characterization and density functional theory calculations, we find that NT-3DFG edge sites are readily functionalized by carbonyl (C=O) and hydroxyl (C-OH) groups under alkaline ORR conditions. Our calculations indicate that multiple site configurations at both armchair and zigzag edges may achieve a local coordination environment that allows selective, two-electron ORR. We derive a general geometric descriptor based on the local coordination environment that provides activity predictions of graphene surface sites within ~0.1 V of computed values. Herein we combine synthesis, spectroscopy, and simulations to improve active site characterization and accelerate carbon-based electrocatalyst discovery.

cond-mat.mtrl-sci↗

arXiv2018

Quantifying Confidence in DFT Predicted Surface Pourbaix Diagrams of Transition Metal Electrode-Electrolyte Interfaces

Density Functional Theory (DFT) calculations have been widely used to predict the activity of catalysts based on the free energies of reaction intermediates. The incorporation of the state of the catalyst surface under the electrochemical operating conditions while constructing the free energy diagram is crucial, without which even trends in activity predictions could be imprecisely captured. Surface Pourbaix diagrams indicate the surface state as a function of the pH and the potential. In this work, we utilize error-estimation capabilities within the BEEF-vdW exchange correlation functional as an ensemble approach to propagate the uncertainty associated with the adsorption energetics in the construction of Pourbaix diagrams. Within this approach, surface-transition phase boundaries are no longer sharp and are therefore associated with a finite width. We determine the surface phase diagram for several transition metals under reaction conditions and electrode potentials relevant for the Oxygen Reduction Reaction (ORR). We observe that our surface phase predictions for most predominant species are in good agreement with cyclic voltammetry experiments and prior DFT studies. We use the OH$^*$ intermediate for comparing adsorption characteristics on Pt(111), Pt(100), Pd(111), Ir(111), Rh(111), and Ru(0001) since it has been shown to have a higher prediction efficiency relative to O$^*$, and find the trend Ru>Rh>Ir>Pt>Pd for (111) metal facets, where Ru binds OH$^*$ the strongest. We robustly predict the likely surface phase as a function of reaction conditions by associating c-values to quantifying the confidence in predictions within the Pourbaix diagram. We define a confidence quantifying metric using which certain experimentally observed surface phases and peak assignments can be better rationalized.

cond-mat.mtrl-sci↗

arXiv2018

Quantifying Confidence in DFT Predicted Surface Pourbaix Diagrams and Associated Reaction Pathways for Chlorine Evolution

Density functional theory calculations can be used to identify dominant reaction mechanisms. However, the dominant reaction mechanism is sensitive to choice of the exchange correlation functional. Here, we demonstrate using the example case of chlorine evolution reaction on rutile oxides, which can occur through at least three reaction mechanisms each mediated by different surface intermediates and active sites. We utilize Bayesian error estimation capabilities within the BEEF-vdW exchange correlation (XC) functional to quantify the uncertainty associated with predictions of the operative reaction mechanism by systematically propagating the uncertainty originating from DFT-computed adsorption free energies. We construct surface Pourbaix diagrams based on the calculated adsorption free energies for rutile oxides of Ru, Ir, Ti, Pt, V, Sn and Rh. We utilize confidence-value (c-value) to determine the degree of confidence in the predicted surface phase diagrams. Using the scaling relations between the adsorption energies of intermediates we construct a generalized Pourbaix diagram showing the stable surface composition as a function of potential and the oxygen binding energy on the cus site. This is used to consistently determine activity volcano relationships. We incorporate the uncertainty in linear scaling relations to quantify the confidence in generalized Pourbaix diagram and the associated activity. This allows us to compute the expectation limiting potential as a function of $ΔE_{O^c}$, which provides a more appropriate activity measure incorporating DFT uncertainty. We show that the confidence in the classification problem of identifying the active reaction mechanism is much better than the prediction problem of determining catalytic activity.

cond-mat.mtrl-sci↗

arXiv2017

Maximal predictability approach for identifying the right descriptors for electrocatalytic reactions

Density Functional Theory (DFT) calculations are being routinely used to identify new material candidates that approach activity near fundamental limits imposed by thermodynamics or scaling relations. DFT calculations have finite uncertainty and this raises an issue related to the ability to delineate materials that possess high activity. With the development of error estimation capabilities in DFT, there is an urgent need to propagate uncertainty through activity prediction models. In this work, we demonstrate a rigorous approach to propagate uncertainty within thermodynamic activity models. This maps the calculated activity into a probability distribution, and can be used to calculate the expectation value of the distribution, termed as the expected activity. We prove that the ability to distinguish materials increases with reducing uncertainty. We define a quantity, prediction efficiency, which provides a precise measure of the ability to distinguish the activity of materials for a reaction scheme over an activity range. We demonstrate the framework for 4 important electrochemical reactions, hydrogen evolution, chlorine evolution, oxygen reduction and oxygen evolution. We argue that future studies should utilize the expected activity and prediction efficiency to improve the likelihood of identifying material candidates that can possess high activity.

cond-mat.mtrl-sci↗

arXiv2017

Surface Restructuring of Nickel Sulfide Generates Optimally-Coordinated Active Sites for ORR Catalysis

First-row transition metal oxides and chalcogenides have been found to rival the performance of precious metal-based catalysts for the interconversion of water and O$_2$. The high lability of the first-row transition metal ions leads to surface dynamics under the conditions of catalysis and results in active site structures distinct from those expected by surface termination of the bulk lattice. While these surface transformations have been well-characterized on many metal oxides, the surface dynamics of heavier chalcogenides under electrocatalytic conditions are largely unknown. We recently reported that the heazlewoodite Ni$_3$S$_2$ bulk phase supports efficient ORR catalysis under benign aqueous conditions and exhibits excellent tolerance to electrolyte anions such as phosphate which poison Pt. Herein, we combine electrochemistry, surface spectroscopy and high resolution microscopy to characterize the surface dynamics of Ni$_3$S$_2$ under ORR catalytic conditions. We show that Ni$_3$S$_2$ undergoes self-limiting oxidative surface restructuring to form an approximately 2 nm amorphous surface film conformally coating the Ni$_3$S$_2$ crystallites. The surface film has a nominal NiS stoichiometry and is highly active for ORR catalysis. Using DFT calculations we show that, to a first approximation, the catalytic activity of nickel sulfides is determined by the Ni-S coordination numbers at surface exposed sites through a simple geometric descriptor. In particular, we find that the surface sites formed dynamically on the surface of amorphous NiS during surface restructuring provide an optimal energetic landscape for ORR catalysis. This work provides a systematic framework for characterizing the rich surface chemistry of metal-chalcogenides and provides principles for developing a broader understanding of electrocatalysis mediated by amorphous materials.

cond-mat.mtrl-sci↗