Exploring Rashba effect in layered 2D materials

DDP Project

Presented by: Keshav Bulia

Guide: Professor Amrita Bhattacharya

PhD guide: Dr. Krishna Kundavu

Content

Introduction:

Theoretical background of Spintronics, Rashba materials, tuning methods etc.

​

Background:

Current Research for 2D materials

​

​

Machine Learning Approach:

Whats done up till now for ML driven predictions

​

Results & Discussion:

Dataset construction

Trends in materials Found up to date

Feature constructions & Running ML model

future work

Spintronics

Introduction

Figure 2.6: Modulation of the Rashba parameter αR in 2D Janus TMDs under (a) external electric field and (b) biaxial strain

Figure 2.7: Variation in the Rashba parameter of 2D TMDs in response to applied external electric field.

AB Binary Buckled Monolayers

Includes but not limited to: Phosphorene alloys (P–X: As, Bi, Sb), BiSb, AlBi, h-NbN, h-TaN, MgX, ZnTe, CdTe, PbX (X = S, Se, Te)

​

Trend: Rashba coefficient (αR) increases with atomic number → stronger SOC

Example: P–As (0.13 eV·Å) → P–Bi (1.56 eV·Å)

​

Geometric activation:

Planar PbS = centrosymmetric

Buckled PbS = large αR

Janus Transition Metal Dichalcogenides (TMDs)

Janus TMDs: asymmetric monolayers of form MXY

M = Mo, W, Cr X ≠ Y ∈ {S, Se, Te}

Feature importance analysis revealed that B-site cation atomic number (Bz ) was the

dominant predictor, confirming the central role of spin-orbit coupling strength.

​

Eight of the ten most important features related to BX6 octahedral geometry, including:

​

• Octahedral distortion parameters (Σ, ∆, Λ2, Λ3) measuring bond angle and length

deviations

• B-X bond length statistics (σ(RBX ), max(RBX ))

• Electronegativity difference (χBX Diff), which showed positive correlation with αRD

(PCC = 0.42), suggesting enhanced covalency facilitates the structural distortions

necessary for strong Rashba splitting

\

• Developed two complementary datasets of 2D Rashba materials from literature and computational repositories.

​

• Conducted data analysis to figure out trends (space group, k-path, material family, elemental trends).

​

• Engineered 250+ physics-based features (elemental, structural).

​

• correlation based reduction of features for ML algorithms

​

• Trained multiple regression models (SVM, XGBoost, RF, Ridge).

Top 10 most frequent elements,

highlighting dominance of chalcogens (Te, Se, S) and heavy metals (Bi, W, Mo).

Figure 4.3: Mean Rashba parameter by material family. Janus structures show highestvalues due to intrinsic asymmetry

BiTeCl 7.48, BiTeBr 9.15, BiTeI 2.086,

2D

BiTeCl 1.9, BiTeBr 2.0, BiTeI 3.85

Bulk

Derived from 35 atomic properties × 5 statistics (mean, max, range, ratio, std).

Voronnoi radii of 0.5,1.0,1.5,2.0 * Lattice parameter

​

Statistical values, for each radii, for each atom in the structure

Correlation Filtering with > 0.8 Correlation between values

Reduced dimensions by 77%.

Retained 16 AFS + 14 GRDF + 28 elemental = 58 total features.

1. Descriptor Improvement

Create new features using available orbital and electronic data.

Capture more physics behind Rashba behavior through refined descriptors.

​

2. Dataset Expansion

Add more materials through high-throughput DFT calculations, especially for verification.

​

3. Advanced ML Methods

Use Graph Neural Networks for structure-based learning.

Explore transfer learning and ensemble models for better accuracy and reliability.