Summary
Bonded single crystal thin film multi-functional materials (electro-optic, acousto-electric, acousto-optic, magneto-optic or multi-ferroic materials) are vital for diverse sensing and communications technologies (integrated quantum, photonic, terahertz [THz], radio frequency [RF], and actuator platforms).
Wafer bonding onto compatible substrates is the critical step for integrating single crystal thin films into multi-functional devices and systems. There is currently no method to analytically investigate wafer bonding processes.
Brute force experimentation is required for initial process development, and subsequent changes in materials or device parameters necessitate significant additional trial and error, incurring huge costs, prolonging timeframes, thus limiting exploration of novel material systems. Because there are no generalizable approaches to model wafer bonding between thin film crystals and substrates, wafer bonding process development remains highly empirical.
As a result, the wafer bonding process is highly specific to select materials and device parameters and is limited by experimental knowledge to a small subset of known materials and device parameters. As an example, although lithium niobate on insulator (LNOI) exhibits promising piezo-electric and electro-optic properties, the lack of predictive wafer bonding models impedes development of novel applications and limits scalable integration.
The ability to predictively model wafer bonding of thin film crystals would rapidly accelerate the research and development of multi-functional materials, and their fabrication and scalable integration in diverse applications.
CRYSTAL explores the question: To accelerate development and integration of multi-functional materials, how do we create generalizable models to explore thin film crystal bonding onto suitable substrates under diverse real-world process conditions and parameters?
Performers | ||||
| Machine Learning-Enhanced Molecular Dynamics for Predictive LiNbO₃/SiO₂ Wafer Bonding | ||||
| Institution: | University of Delaware | |||
| PI/Team: | Nuwan Dewapriya, Prof. John W. Gillespie Jr. | |||
| Abstract: | This project developed a machine-learning-enhanced multiscale workflow to predict fracture at bonded interfaces in heterogeneous photonic and electronic devices. Density functional theory provided reference data for training a machine-learned interatomic potential, enabling molecular dynamics simulations of LiNbO₃/SiO₂ interfaces under normal, shear, and mixed-mode loading across multiple temperatures. The atomistic responses were reduced to cohesive-zone laws suitable for device-scale finite-element models. The study quantified thermal weakening of the bonded interphase and applied the workflow to a chemically distinct SiC/SiO₂ interface. This capability turns interfacial strength and fracture resistance into computable design inputs for heterogeneous integration, integrated photonics, and advanced packaging. | |||
| Multiscale Architecture for Study, Engineering, and Replication of Bonding Lithium Niobate Thin Films (MAStER BoLT) | ||||
| Institution: | Northrop Grumman Corporation | |||
| PI/Team: | Kevin Galiano, Iman Javaheri, Vivian Ryan, Nicholas Simonson, Thomas Knight | |||
| Abstract: | NGC brings DoD-knowledgeable simulation capabilities to MAStER BoLT, a novel multi-scale, multi-physics wafer bonding model for photonics and microelectronics technologies. Methodologies include Molecular Dynamics and Density Functional Theory atomistic simulations to extract optical, electro-optical, phonon, and adhesion properties, greatly accelerated by Machine-Learned Force Fields. Implantation simulations can account for the exfoliation of thin films while Finite Element Method simulations address structural mechanical responses to temperature and force over time. The framework is adaptable to rapidly model other systems with different layer materials, interfaces, and orientations, while enabling one to refine bonding parameters (temperature, force, time) to help guide experimental designs. | |||
| Heterogeneous Integration on Patterned Substrates (HIPS) | ||||
| Institution: | University of California, San Diego | |||
| PI/Team: | Prof. Shayan Mookherjea | |||
| Abstract: | The goal of this project is to develop a modeling framework for direct bonding (DB) of thin-film lithium niobate (LN) and lithium tantalate (LT) on patterned substrates which include photonic integrated circuits (PIC) in silicon or silicon nitride. These structures are not only useful for practical photonic integrated circuits (PICs) in the context of heterogeneous integration, but sensitively study the bonding quality over large areas. The hybrid structures, first realized at our university, are now reliably made in collaboration with an external foundry and serve a sizable community. Guided by input from them, we have tried to understand potential debonding mechanisms following a (mostly successful) initial bonding step. We used cohesive-zone models with finite-element analysis methods to study the onset of inelastic damage and aim to make predictions about which materials and patterns can lead to superior DB outcomes. | |||
| Generalized high-throughput wafer-bonding simulation engine | ||||
| Institution: | Matnex | |||
| PI/Team: | Ayush Suhane, Cristian Rebolledo, Riccardo Bassiri, Jonathan Bean | |||
| Abstract: | MatNex is developing a generalizable, physics-informed platform to predict thin-film crystal-substrate bonding under realistic process conditions. In the CRYSTAL project, we developed workflows to generate computationally efficient synthetic datasets to fine-tune charge-aware machine-learning interatomic potentials. The transferability of these models is validated for crystal/substrate interfaces, and the finetuned MLIP is used with molecular-dynamics simulations to predict work of adhesion, interfacial reconstruction and charge redistribution across temperatures and crystal orientations. This workflow enables high-throughput screening of crystal-substrate combinations with strong bonding potential under experimental processing conditions. MatNex welcomes academic and industrial partners to validate, extend and deploy the capability through collaborative research and technology-development programs. | |||
| Framework for Unified Simulation in Epitaxial/Crystal Bonding (FUSE) | ||||
| Institution: | University of Florida | |||
| PI/Team: | Quazi Hossai, Katherine Loske, Prof. Travis Anderson, Prof. Dennis S. Kim | |||
| Abstract: | Heterogeneous integration of multifunctional single-crystal thin films is critical for next-generation photonics, quantum computing, and advanced sensing. Wafer bonding, combined with layer-transfer methods such as ion-cut, enables the assembly of complex multi-material systems, yet process development remains largely empirical, costly, and difficult to generalize to emerging materials. The FUSE program (Framework for Unified Simulation in Epitaxial/Crystal Bonding) develops a predictive, physics-based framework that bridges the quantum-mechanical description of the chemical bond and wafer-scale thermomechanics. It couples density functional theory (DFT), machine-learned interatomic potentials (MLIPs), crystal orbital Hamilton population (COHP) bonding analysis, and finite-element multiphysics to describe ion implantation, amorphization, temperature- and pressure-dependent thermal expansion, and interfacial bonding in a lithium niobate on silica model system, with in-house wafer-bonding experiments providing validation. In recent work, we constructed and stabilized crystalline and amorphous LiNbO3 interfaces from first principles and computed temperature- and pressure-dependent bonding (COHP) across the interface. Initial results indicate that the covalent Nb-O network is the primary load-bearing interaction and a likely control point for adhesion and ion-cut splitting. Together, DFT, QM/MM, and MLIPs form a pipeline that lowers cost and enables larger, transferable simulations required for predictive wafer-bonding design. | |||
| Atomistic Simulation of Surface-Activated Bonding Using Machine Learned Interatomic Potentials | ||||
| Institution: | University of Missouri - Kansas City (UMKC) | |||
| PI/Team: | Prof. Paul Rulis, Prakash Khanal, Prof. Alexander Khanal | |||
| Abstract: | UMKC’s CRYSTAL effort develops an atomistic engine for surface-activation and bonding simulations, combining wafer-bonding process insight with Density Function Theory (DFT), machine-learned interatomic potential (MLIP) generation, and large-scale molecular dynamics (MD). Our plan centers on a four-stage workflow: generate and curate ab initio training data; train MLIPs for plasma/ion activation and bonding; simulate activated-surface contact and post-bonding evolution; and quantify interface quality. We construct targeted training sets spanning bulk, cleaved and strained crystals, amorphous and misoriented surfaces, finite-temperature snapshots, ion-bombarded configurations, and heterogeneous interfaces for LiNbO3, SiO2/Si, GaN, and Ar environments. These data support DeePMD/SNAP-based potentials. Application of these MLIPs to LAMMPS simulations will track activation-induced disorder, surface roughness, pressure, temperature, and equilibration history affects bond formation. Targeted outputs include work of adhesion/surface energy, coordination and bond statistics, relative density, void indicators, roughness evolution, residual stress, and tensile response. Our near-term objective is LiNbO3-on-SiO2 validation followed by GaN-on-LiNbO3 assessment. We are targeting transferable atomistic descriptors and common outputs for cross-PI comparison within ARC CRYSTAL’s broader modeling ecosystem and future program-scale continuum-model coupling efforts. | |||
| Multiscale and multiphysics modeling of LiNbO₃ bonding onto an insulator | ||||
| Institution: | University of Texas at Dallas | |||
| PI/Team: | Prof. Kyeongjae Cho, Prof. Xinfang Jin | |||
| Abstract: | Heterogeneous integrations of multi-functional device materials are required for diverse defense electronics applications including sensing and communication. Wafer bonding onto compatible substrates is a critical step in the fabrication of multifunctional materials. For Lithium Niobate on Insulator (LNOI), bonding to an Si or another insulator substrate can be achieved using various techniques, including direct wafer bonding, adhesive bonding, and the smart-cut process. This research specifically focuses on direct wafer bonding and smart-cut technique. For direct wafer bonding, the simulation will capture both the initial bonding stage, induced by applied external force, and the final bonding achieved through high-temperature annealing. The Multiscale and Multiphysics approaches apply Molecular dynamics (MD) simulations alongside micro- and macro-scale simulations to predict the overall interface strength based on parameters related to materials, structures and processes. In this research, multiscale modeling will integrate atomic scale modeling for interatomic bond formation mechanisms under diverse thermal and mechanical conditions, and micro- and macro-scale modeling for the wafer scale thin film morphology evolution during the bonding processes. The outcome of the proposed research will enable predictive design on single crystal thin film water bonding on diverse insulator substrates. | |||
| Physics-Based Computational Models for Predicting Direct Crystalline Bonding Process Outcomes | ||||
| Institution: | University of Pennsylvania | |||
| PI/Team: | Prof. Kevin Turner, Sage Fulco, Michael Chiou | |||
| Abstract: | The development of direct wafer bonding processes for heterogeneous materials integration remains difficult and highly empirical, as the process window is narrow and depends on the interplay of surface chemistry, roughness, wafer geometry, and thermal treatment. We have developed a multi-scale, physics-based simulation tool to quantitatively predict bonding outcomes by combining three models:
Together, these form an integrated simulation tool for process design and serve as the basis of a digital twin of the process. We have applied these models to processes for lithium tantalate-on-insulator (LTOI) and lithium niobate-on-insulator (LNOI) substrates, but the models can easily be extended to other material systems. | |||
Active topic
The solicitation for this program closed on June 16, 2025. This topic is within its period of performance.
Resources
- DARPA-EA-25-02-04
Solicitation - DARPA-EA-25-02
Exploration Announcement
