A Kinetic Phase-Field Model of Diffusion Bonding: A Nonlocal Approach to Interface Coalescence

1. Problem Statement

Conventional phase-field models (typically based on Allen-Cahn or Cahn-Hilliard equations) rely on free energy minimization to drive interface evolution. A fundamental limitation of these models in the context of solid-state diffusion bonding is their inherent tendency to cause interfaces to coalesce (merge) whenever they are in close proximity.

• The Issue: In diffusion bonding processes (e.g., joining ceramics or dissimilar metals), specific thermodynamic conditions may require interfaces or thin interlayers to remain distinct and stable rather than merging. Conventional models often fail to preserve these distinct boundaries, leading to unphysical predictions where thin interlayers disappear or interfaces broaden uncontrollably.
• Current Limitations: Previous attempts to control this behavior often rely on fine-tuning numerical mobilities or creating complex, concentration-dependent energy landscapes. These approaches are often fragile, requiring precise energetic balancing that may not hold under varying kinetic parameters or thermodynamic conditions.

2. Methodology

The authors propose a Kinetic Phase-Field Model that decouples interface motion (kinetics) from interface creation (nucleation) and introduces a geometric control mechanism to arrest interface coalescence.

A. Evolution Equation

Instead of the standard gradient descent evolution ($\partial_t \phi \propto -\delta_\phi P$), the model adopts an evolution law derived from a geometric conservation law (Agrawal and Dayal, 2015):
$$\frac{\partial \phi}{\partial t} = |\nabla \phi| v_n^\phi + G$$

• $|\nabla \phi|$: Ensures kinetic evolution occurs only at existing interfaces.
• $G$: A source term for nucleation (controlled separately).
• $v_n^\phi$: The interface velocity, which is modulated by a coalescence kinetic function, $g$.

B. The Coalescence Kinetic Function ($g$)

The core innovation is the function $g(\phi, \nabla \phi, \nabla^2 \phi, \dots)$, which acts as a "brake" on interface motion.

• Mechanism: The interface velocity is defined as $v_n^\phi = \kappa(1-g)F$, where $F$ is the thermodynamic driving force.
• Activation Criteria: The function $g$ approaches 1 (arresting motion) only when specific nonlocal geometric criteria are met. These criteria detect regions representing a persistent, thin material layer (a local spatial minimum of the phase field).
  • 1D: Requires $\phi$ near a target value, low gradient ($|\phi_x| < \gamma$), and positive curvature ($\phi_{xx} > \beta$).
  • 2D/3D: Uses principal curvatures (eigenvalues of the Hessian matrix) or Sylvester's criterion (leading principal minors) to ensure the region is a strict local minimum (convexity) rather than a saddle point or flat region.
• Thermodynamic Consistency: The model satisfies the Second Law of Thermodynamics because $g \in [0, 1]$, ensuring the rate of energy dissipation remains non-negative.

C. Numerical Implementation

• Mixed Finite Element Method (FEM): Due to the presence of second-order derivatives ($\nabla^2 \phi$) in the geometric criteria, the authors use a mixed formulation. They introduce an auxiliary vector field $\psi = \nabla \phi$ to handle higher-order derivatives within a $C^0$ continuous framework.
• Software: Implemented using FEniCS.
• Coupling: The model is coupled with mechanical equilibrium equations to simulate stress effects during bonding.

3. Key Contributions

1. Geometric Control of Coalescence: Introduces a physically motivated, nonlocal geometric criterion to explicitly stop interface merging, addressing a critical gap in conventional phase-field modeling for diffusion bonding.
2. Decoupling Kinetics and Nucleation: Utilizes a geometric conservation law to separate interface motion from nucleation, allowing for arbitrary complex kinetic laws that are not strictly tied to free energy gradients.
3. Temperature-Dependent Thresholds: The geometric thresholds ($\alpha, \beta, \gamma$) can be linked to macroscopic variables like temperature, allowing the model to predict critical interlayer thicknesses without explicitly solving complex advection-diffusion equations for solute concentration.
4. Robustness: Demonstrates that the model can preserve distinct interfaces (e.g., thin Ti interlayers between ZrC blocks) under conditions where standard models would predict complete homogenization.

4. Results

The paper validates the model through 1D, 2D, and 3D simulations, as well as a coupled mechanics application.

• 1D Simulations:
  • Standard Model: Two interfaces approach and merge, eliminating the central region.
  • Proposed Model: As the central region narrows and meets geometric criteria, $g \to 1$, arresting motion and preserving a stable gap.
• 2D Simulations:
  • Successfully preserved parallel vertical interfaces, diagonal interfaces, and circular inclusions (elliptical shapes) that would otherwise shrink and vanish due to curvature-driven evolution.
  • Parametric Sensitivity: The model is robust to mesh refinement and time-step variations. The geometric thresholds ($\alpha, \beta, \gamma$) directly control the final interlayer width, offering a calibration tool for experimental data.
• Ti-ZrC Diffusion Bonding (Coupled with Mechanics):
  • Simulated the bonding of ZrC blocks with a Titanium (Ti) interlayer.
  • Critical Thickness: The model successfully predicted a critical interlayer thickness below which the interface arrests, matching experimental observations of critical thickness in Ti-ZrC systems.
  • Mechanical Effects: Showed that while eigenstrain (transformation strain) influences local stress accumulation, the geometric control function $g$ dominates the macroscopic morphological outcome, preserving the interlayer structure.

5. Significance

This work provides a computationally efficient alternative to traditional multi-field phase-field models (like Cahn-Hilliard coupled models) for diffusion bonding.

• Efficiency: By avoiding the explicit resolution of solute concentration fields, the model reduces the number of coupled partial differential equations, making it faster while still capturing the macroscopic outcome (interface stability).
• Design Capability: It offers a direct mechanism to design processes where specific residual interlayers are desired (e.g., for thermal barrier properties or stress mitigation) by tuning geometric thresholds rather than complex chemical potentials.
• Broad Applicability: The geometric approach is applicable beyond diffusion bonding to other problems involving interface pinning, such as grain boundary stabilization, crack healing, and the behavior of functionally graded materials.

In summary, the paper presents a paradigm shift from energy-minimization-only models to kinetically controlled models that use local geometric invariants to enforce physical constraints on interface coalescence, significantly improving the predictive capability of phase-field simulations for advanced manufacturing processes.