Estimation of exciton diffusion lengths of organic semiconductors in random domains

Jingrun Chen, Ling Lin, Zhiwen Zhang, Xiang Zhou

Abstract

Exciton diffusion length plays a vital role in the function of opto-electronic devices. Oftentimes, the domain occupied by an organic semiconductor is subject to surface measurement error. In many experiments, photoluminescence over the domain is measured and used as the observation data to estimate this length parameter in an inverse manner based on the least square method. However, the result is sometimes found to be sensitive to the surface geometry of the domain. In this paper, we employ a random function representation for the uncertain surface of the domain. After non-dimensionalization, the forward model becomes a diffusion-type equation over the domain whose geometric boundary is subject to small random perturbations. We propose an asymptotic-based method as an approximate forward solver whose accuracy is justified both theoretically and numerically. It only requires solving several deterministic problems over a fixed domain. Therefore, for the same accuracy requirements we tested here, the running time of our approach is more than one order of magnitude smaller than that of directly solving the original stochastic boundary-value problem by the stochastic collocation method. In addition, from numerical results, we find that the correlation length of randomness is important to determine whether a 1D reduced model is a good surrogate for the 2D model.

Introduction

In the development of organic electronics—such as OLED displays and flexible solar cells—excitons (bound electron-hole pairs) serve as the primary energy carriers. The efficiency of these devices is heavily dependent on the exciton diffusion length, which defines how far an exciton can travel before recombining. Accurately measuring this parameter is critical for optimizing device architecture.

Typically, researchers estimate this length by performing photoluminescence (PL) experiments and applying mathematical models to "back out" the diffusion length from the observed data. However, a significant challenge arises from the fact that material interfaces are rarely perfectly flat. Nanoscale roughness and measurement errors introduce geometric uncertainties that standard 1D models often ignore. These small perturbations can lead to significant errors in diffusion length estimation. While Uncertainty Quantification (UQ) methods can address this, they are traditionally computationally expensive and slow.

Mathematical Modeling and Methodology

To address the computational bottleneck of UQ, this research proposes an asymptotic-based method to solve the forward diffusion model. The core idea is to transform a complex stochastic boundary problem into a series of simpler deterministic equations.

Schematic of the donor domain with a random interface x=h(z) and incident solar light.

1. Random Domain Representation

The uncertain interface is modeled as a random function $x=h(z)$, often represented as a sine series with random coefficients. This accounts for the "jaggedness" or roughness observed in real-world organic semiconductor films.

2. Asymptotic Expansion

Rather than employing a Monte Carlo approach—which would require solving the diffusion equation thousands of times for different surface realizations—the researchers assume the roughness is small relative to the film thickness ($\epsilon \ll 1$). The solution is expanded into a Taylor-like series: $$w = w_0 + \epsilon w_1 + \epsilon^2 w_2 + \dots$$

3. Deterministic Reduction

The strength of this method lies in its ability to solve each term in the expansion using standard deterministic Partial Differential Equations (PDEs) on a fixed, perfectly flat domain. This "non-intrusive" approach avoids the need to sample the random space directly, which is the primary bottleneck in methods like Stochastic Collocation (SC).

4. Inverse Solver

Finally, Newton’s method is used to iteratively find the diffusion length that best fits the experimental PL data, utilizing the fast asymptotic solver at each iteration to maintain high efficiency.

Numerical Experiments and Results

The model was validated using synthetic datasets designed to mimic real-world bilayer devices. The researchers simulated photoluminescence data using a high-fidelity 2D model with known parameters, varying film thickness ($d$) between 10 nm and 100 nm.

Key Findings

  • Efficiency Gains: The proposed asymptotic method achieved an 18x speedup over the Stochastic Collocation method while maintaining comparable accuracy.
  • Correlation Length Importance: A critical insight from the numerical results is the role of the "correlation length" of surface randomness. In materials with high crystalline order, the interface tends to be "smoothly" rough (long correlation length), allowing 1D models to perform reasonably well.
  • Failure of 1D Models: For materials with low crystalline order and highly jagged interfaces (short correlation length), 1D models fail significantly. In these scenarios, the 2D asymptotic approach is necessary for accurate parameter estimation.
  • Convergence and Robustness: The inverse estimation of the diffusion length converged reliably across various levels of surface uncertainty ($\epsilon$ ranging from 0.01 to 0.32). Furthermore, the optimization algorithm converged faster for larger diffusion lengths, making it particularly suitable for high-performance materials.

Conclusion and Future Work

This research provides a highly efficient framework for estimating exciton diffusion lengths in the presence of geometric uncertainty. By reducing stochastic problems to deterministic ones, the method offers the precision of 2D modeling with a fraction of the computational cost.

However, some limitations remain. The method relies on the small perturbation limit; if the surface roughness is comparable to the film thickness (e.g., in ultra-thin films), the asymptotic approximation may lose accuracy. Future work will focus on exploring non-Gaussian random fields to represent specific manufacturing defects and validating the framework against raw Atomic Force Microscopy (AFM) data from physical laboratory samples.

Summary updated

Abstract

Exciton diffusion length plays a vital role in the function of opto-electronic devices. Oftentimes, the domain occupied by an organic semiconductor is subject to surface measurement error. In many experiments, photoluminescence over the domain is measured and used as the observation data to estimate this length parameter in an inverse manner based on the least square method. However, the result is sometimes found to be sensitive to the surface geometry of the domain. In this paper, we employ a random function representation for the uncertain surface of the domain. After non-dimensionalization, the forward model becomes a diffusion-type equation over the domain whose geometric boundary is subject to small random perturbations. We propose an asymptotic-based method as an approximate forward solver whose accuracy is justified both theoretically and numerically. It only requires solving several deterministic problems over a fixed domain. Therefore, for the same accuracy requirements we tested here, the running time of our approach is more than one order of magnitude smaller than that of directly solving the original stochastic boundary-value problem by the stochastic collocation method. In addition, from numerical results, we find that the correlation length of randomness is important to determine whether a 1D reduced model is a good surrogate for the 2D model.
BibTeX
@article{chen2019estimation,
  author = {Jingrun Chen and Ling Lin and Zhiwen Zhang and Xiang Zhou},
  title = {{Estimation of exciton diffusion lengths of organic semiconductors in random domains}},
  journal = {Journal of Computational Physics},
  year = {2019},
  doi = {10.1016/j.jcp.2018.10.008},
}

Loading related content...

Figure