TL;DR
This paper introduces a refined empirical tight-binding model that accurately captures the effects of [110] uniaxial strain in silicon. By separately accounting for Löwdin orbital changes and nearest-neighbor potential displacements, the model achieves better predictions for electron mobility and device performance, crucial for next-generation transistors.
What Problem Does This Paper Address?
Modern transistors rely on strain engineering to boost electron and hole mobility, with [110] uniaxial strain being particularly important for increasing electron mobility. While empirical tight-binding (ETB) methods are excellent for simulating large nanostructures, they often struggle to accurately predict the behavior of devices under this specific type of strain. This inaccuracy stems from neglecting "same-atom, different-orbital" matrix elements that arise due to the strain. Existing ETB models either simplify these effects or lump them together, making it difficult to understand the physical origins and set realistic parameter limits. This paper aims to address this by developing an ETB model that explicitly separates and quantifies the contributions of Löwdin orbital changes and nearest-neighbor potential shifts to [110] uniaxial strain.
Methods
The core of this paper is the development of an improved empirical tight-binding model. The authors build upon existing methods by:
- Separating Strain Effects: They explicitly distinguish between two key mechanisms induced by [110] uniaxial strain:
- Löwdin Orbital Changes: This accounts for how the atomic orbitals themselves are modified and become orthogonal under strain.
- Nearest-Neighbor Potential Displacements: This addresses how the electrostatic potential experienced by an atom changes due to the strained positions of its neighbors.
- Generalizing Löwdin Procedure: They extend the Löwdin procedure to include off-diagonal, same-atom Hamiltonian matrix elements, which are crucial for accurately describing strain.
- Parametrization: They optimize a set of strain parameters for silicon (Si) by fitting to experimental data and ab initio calculations. This optimization process uses a genetic algorithm (GA) to efficiently explore the parameter space.
- Effective Slater-Koster Parameters: The model utilizes effective Slater-Koster parameters derived from the nearest-neighbor potential calculations, allowing for a more transparent connection to physical quantities.
The novelty lies in the explicit separation of these strain-induced effects, allowing for better physical insight and more constrained parameter fitting, unlike previous models that combined these effects.
Datasets
The paper uses the following for parameter optimization and validation:
- Ab initio calculations: Specifically, VASP (Vienna Ab initio Simulation Package) calculations are used to obtain effective masses under strain.
- Model solid theory: van de Walle's model solid theory is used to fit band-edge energies.
- Experimental data: While not explicitly listed as a "dataset," the authors mention comparing their results to experimental measurements of strain-induced current changes in transistors.
No new benchmark datasets are introduced in this paper.
Key Insights and Ideas
- [110] Uniaxial Strain is Peculiar: This specific strain orientation introduces "same-atom, different-orbital" matrix elements that are often neglected, leading to inaccuracies in standard ETB models.
- Separation is Key: Explicitly separating the contributions of Löwdin orbital changes and nearest-neighbor potential shifts provides better physical understanding and allows for more reliable parameter determination.
- Physical Constraints Matter: By separating effects, the authors can place more reasonable physical limits on the strain parameters, leading to more robust and trustworthy simulations.
- Improved Mobility Prediction: The refined model accurately reproduces the splitting of Si's X-valley transverse masses under [110] strain, which is critical for predicting electron mobility.
- Device Performance Impact: The model's improved accuracy in capturing strain effects translates to more realistic predictions of device performance, such as drain current in strained transistors.
Key Results
- Accurate Band Structure: The model successfully reproduces the band-edge energies and effective masses of silicon under both [001] and [110] uniaxial strain, showing good agreement with ab initio calculations and model solid theory.
- Transverse Mass Splitting: The model accurately captures the splitting of the X₂ valley transverse masses under [110] uniaxial strain, a key feature for electron mobility enhancement.
- Device Simulation: When applied to a strained silicon double-gate, ultrathin-body MOSFET, the model predicts a significantly larger change in drain current under [110] strain compared to previous models. This predicted change is in reasonable agreement with experimental observations.
- Parameter Insight: The separation of strain effects allows for a clearer understanding of the physical meaning and reasonable ranges for the strain parameters.
Limitations and Future Work
- Approximations: The model still relies on approximations, such as retaining only the lowest-order terms in multipole expansions for nearest-neighbor potentials. While these approximations are justified by their impact on the results, they do introduce some limitations.
- Parameter Optimization Intensity: The optimization process using a genetic algorithm is computationally intensive, requiring millions of trial candidates.
- Future Work: The authors suggest that this model should be useful for multiband Si device calculations, implying its applicability to more complex device structures and materials. The explicit separation of strain effects also opens avenues for applying this methodology to other materials and strain configurations.
Summary updated