Access Official Technical PDF

Share Summary:LinkedInEmail

As copper interconnects shrink, their microstructure becomes a critical factor in electrical performance and reliability. Grain boundaries and material interfaces affect both electron transport and the movement of atoms and vacancies during electromigration (EM). For engineers using electromigration modeling to evaluate advanced interconnects, representing these features accurately without making simulations impractical is a central challenge.

A spatial-parameter method offers an alternative to explicitly defining and finely meshing every boundary. Instead, it uses each location’s proximity to a grain boundary or interface to assign local material properties. This approach can help capture microstructure-dependent behavior while reducing the mesh and computational burden.

Why copper microstructure matters in electromigration

Reducing interconnect dimensions increases the relative influence of surfaces, interfaces and grain boundaries. These features affect copper resistivity through electron scattering, while the smaller conductor cross-section also means that less material needs to be displaced before a void can disrupt the line. Together, these effects make interconnect performance and lifetime more sensitive to local structure.

Electromigration occurs when an electric current drives the movement of atoms. The resulting redistribution of atoms and vacancies can create regions of accumulation and depletion. Over time, vacancy behavior contributes to stress development and may lead to void formation, increased resistance and, ultimately, an open-circuit failure.

Microstructure complicates this process because the grain bulk, grain boundaries and material interfaces do not share the same transport properties. Boundaries and interfaces can act as faster diffusion paths and as sites where vacancies are generated or annihilated. Grain-boundary orientation also matters: depending on its direction relative to the electron wind, a boundary may facilitate transport or impede it.

Electromigration modeling without explicitly meshing every boundary

One common way to account for microstructure is to use an effective diffusivity that combines bulk and grain-boundary behavior. This can represent an overall material response, but it does not resolve how individual boundaries influence local vacancy movement or stress.

Explicitly representing each grain and its thin boundaries can provide more detail, but the boundary thickness is very small relative to the overall interconnect. Resolving that difference with a conventional mesh can require many elements and substantial memory and computation, limiting the size or complexity of structures that can be simulated.

The spatial-parameter approach instead distinguishes the grain bulk, grain boundaries and material interfaces through properties assigned across the simulation domain. A geometry-generation step creates the grain structure, after which a parameter-assignment tool maps locations such as distance to the nearest boundary onto a grid. Those parameters are then used in a finite-element electromigration simulation.

From grain geometry to local material properties

At each grid location, the method identifies the nearest grain edge and determines whether it represents a grain boundary or a material interface. The location is assigned the corresponding local properties, or bulk properties when it is sufficiently far from either feature. The approach can represent properties including electrical conductivity, vacancy-diffusion parameters and the effective valence used in the electromigration model.

Grain-boundary orientation can also be included. In the described framework, the grain-boundary diffusion contribution is adjusted according to its angle relative to the electric-field direction. This allows the model to distinguish boundaries that may impede transport from those that can promote it. Material interfaces are treated separately, rather than being assigned the same properties as grain boundaries.

Because the properties are assigned spatially, the method does not require every boundary to be represented as a separate material region with a complex, tightly resolved mesh. Different interface properties can also be assigned to different copper surfaces when the surrounding materials have different effects.

What the simulations show

In the reported test case, the method was applied to a narrow copper line with a bamboo-like grain structure. Simulations showed that vacancies redistribute in response to the different properties of the grain bulk, boundaries and interfaces. Grain boundaries affected the resulting vacancy distribution, consistent with their role as potential transport paths and blocking sites.

The study also examined how mesh spacing affected results. Vacancy concentrations were reasonably reproduced with comparatively coarse grids, including cases where the grid spacing was larger than the assumed boundary thickness. The reported simulations maintained useful agreement at grid spacings up to 2 nm, illustrating the potential to represent microstructure without resolving each thin feature with multiple mesh elements.

Stress calculations showed elevated stress at grain boundaries and material interfaces, with notable build-up at triple points where a grain boundary meets an interface. However, the calculated maximum stress was sensitive to mesh resolution. This distinction is important: the method’s ability to model vacancy distribution and stress at specific microstructural locations does not mean that every stress metric is automatically mesh-independent. Engineers should evaluate convergence for the outputs that matter to their reliability assessment.

Practical implications for interconnect analysis

For semiconductor process and reliability teams, the method provides a way to include more realistic copper microstructure in TCAD and finite-element workflows without relying on extremely fine meshes throughout the structure. It may support analysis of how grain size, boundary orientation and interface behavior influence local vacancy transport and stress development.

A typical workflow involves:

  • Generating the grain geometry for the interconnect dimensions and target average grain size.
  • Assigning spatial properties based on distance to boundaries and interfaces, including their distinct transport behavior.
  • Solving the coupled model for electrical conditions, vacancy dynamics and mechanical stress.
  • Checking mesh sensitivity for the specific results used to inform reliability decisions.

The approach does not remove the need for appropriate material parameters or validation against relevant measurements. Its value is in making microstructure-aware simulations more manageable while retaining local distinctions that a single effective material value cannot show.

Conclusion

As copper interconnects scale, grain boundaries and material interfaces increasingly shape both conductivity and electromigration behavior. A spatial-parameter method can represent these features without explicitly constructing a fine mesh around every boundary, enabling more practical analysis of vacancy distributions and stress at microstructural locations such as triple points. For engineering teams, the key is to pair this modeling efficiency with careful parameter selection and mesh-convergence checks, especially when interpreting peak stress and other reliability-critical outputs.

Access Official Technical PDF

About The Author