Diffusion and desorption of<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:mrow><mml:mi mathvariant="normal">Si</mml:mi><mml:msub><mml:mi mathvariant="normal">H</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:math>on hydrogenated<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:mrow><mml:mi mathvariant="normal">H</mml:mi><mml:mo>:</mml:mo><mml:mi mathvariant="normal">Si</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mn>100</mml:mn><mml:mo>)</mml:mo></mml:mrow><mml:mtext>−</mml:mtext><mml:mrow><mml:mo>(</mml:mo><mml:mn>2</mml:mn><mml:mo>×</mml:mo><mml:mn>1</mml:mn><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:math>from first principles

Michele Ceriotti, Marco BernasconiPhysical 2007 · DOI · arXiv

Abstract

We have studied diffusion pathways of a silyl radical adsorbed on the hydrogenated $\text{Si}(100)\text{-(2}\times\text{1)}$ surface by density-functional theory. The process is of interest for the growth of crystalline silicon by plasma-enhanced chemical vapor deposition (PECVD). Preliminary searches for migration mechanisms have been performed using metadynamics simulations. Local minima and transition states have been further refined by using the nudged-elastic-band method. Barriers for diffusion from plausible adsorption sites as low as 0.2 eV have been found, but trap states have also been spotted, leading to a more stable configuration, with escape barriers of 0.7 eV. Diffusion among weakly bound physisorbed states is also possible with very low activation barriers (<50 meV). However, desorption mechanisms from physisorbed or more strongly bound adsorption configurations turn out to have activation energies similar to diffusion barriers. Kinetic Monte Carlo simulations show that the silyl radical diffuses at most by a few lattice spacing before desorbing at temperatures in the range 300–1000 K.

Introduction

In the manufacturing of solar cells and microelectronics, plasma-enhanced chemical vapor deposition (PECVD) is a workhorse technique for growing smooth thin films of silicon. The "magic" of PECVD is its ability to produce high-quality films even at relatively low temperatures. For years, the prevailing theory was that the primary growth precursor—the silyl radical ($\text{SiH}_3$)—lands on a hydrogen-saturated silicon surface and diffuses rapidly until it finds a reactive "dangling bond" site to stick to.

However, previous theoretical models often relied on "educated guesses" about how these radicals move. This paper aims to remove the guesswork by using unbiased search algorithms to find every possible path the radical might take and then calculating the exact energy barriers for those paths using first-principles calculations.

Methods and Computational Setup

The authors employed a multi-scale computational approach to bridge the gap between picosecond-scale atomistic movements and macroscopic growth timescales:

  • Density Functional Theory (DFT): Used to calculate the electronic structure. Crucially, the researchers used spin-unrestricted calculations to correctly model the unpaired electron of the silyl radical.
  • Metadynamics: An accelerated simulation technique used to explore the energy landscape and find unexpected transition states.
  • Nudged Elastic Band (NEB): Used to pinpoint the exact activation barriers for the paths discovered via metadynamics.
  • Kinetic Monte Carlo (KMC): Used to simulate the long-term fate of the radicals based on the calculated energy barriers.
Top view of the silyl radical adsorbed on the hydrogenated Si(100)-(2x1) surface.

The study utilized a slab model representing the $\text{H:Si}(100)-(2\times1)$ surface, consisting of six silicon layers and a vacuum gap. The DFT setup was validated against known gas-phase reaction enthalpies for disilane ($\text{Si}_2\text{H}_6$) decomposition.

Key Insights and Results

The Spin Discrepancy

A major finding was that neglecting spin polarization in calculations—a common practice in earlier studies—can lead to errors as large as 1.0 eV. This discrepancy made radicals appear much more strongly bound to the surface than they actually are, fundamentally altering the predicted dynamics.

Diffusion vs. Desorption

The metadynamics search revealed that while radicals can move easily between some sites with barriers as low as 0.35 eV, they frequently encounter "trap states." Escaping these traps requires overcoming a much higher barrier of approximately 0.80 eV.

Energy profiles along different diffusion pathways for the silyl radical on the surface.

Crucially, the energy required for a radical to move to a new site is very similar to the energy required for it to desorb back into the gas phase. This competition is the "bottleneck" of the process.

Challenging the "Fast Diffusion" Myth

KMC simulations across a wide temperature range (300–1000 K) showed that a silyl radical travels at most a few lattice spacings before desorbing. This suggests that the high smoothness of PECVD films cannot be explained by long-range radical diffusion as previously assumed.

Conclusion

This study demonstrates that silyl radicals are far less mobile on hydrogenated silicon surfaces than previously thought. The findings suggest that the smoothness of PECVD films might instead be driven by "etching" processes, where incoming radicals or hydrogen atoms selectively remove poorly placed silicon atoms, effectively polishing the surface during growth. Future work modeling multi-particle interactions will be essential to confirm this new perspective on silicon film growth.

Summary updated

Abstract

We have studied diffusion pathways of a silyl radical adsorbed on the hydrogenated $\mathrm{Si}\phantom{\rule{0.3em}{0ex}}(100)\text{\ensuremath{-}}(2\ifmmode\times\else\texttimes\fi{}1)$ surface by density-functional theory. The process is of interest for the growth of crystalline silicon by plasma-enhanced chemical vapor deposition. Preliminary searches for migration mechanisms have been performed using metadynamics simulations. Local minima and transition states have been further refined by using the nudged-elastic-band method. Barriers for diffusion from plausible adsorption sites as low as $0.2\phantom{\rule{0.3em}{0ex}}\mathrm{eV}$ have been found, but trap states have also been spotted, leading to a more stable configuration, with escape barriers of $0.7\phantom{\rule{0.3em}{0ex}}\mathrm{eV}$. Diffusion among weakly bound physisorbed states is also possible with very low activation barriers $(&lt;50\phantom{\rule{0.3em}{0ex}}\mathrm{meV})$. However, desorption mechanisms (either as $\mathrm{Si}{\mathrm{H}}_{3}$ or as $\mathrm{Si}{\mathrm{H}}_{4}$) from physisorbed or more strongly bound adsorption configurations turn out to have activation energies similar to diffusion barriers. Kinetic Monte Carlo simulations based on ab initio activation energies show that the silyl radical diffuses at most by a few lattice spacing before desorbing at temperatures in the range $300--1000\phantom{\rule{0.3em}{0ex}}\mathrm{K}$.
BibTeX
@inproceedings{ceriotti2007diffusion,
  author = {Michele Ceriotti and Marco Bernasconi},
  title = {{Diffusion and desorption of<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:mrow><mml:mi mathvariant="normal">Si</mml:mi><mml:msub><mml:mi mathvariant="normal">H</mml:mi><mml:mn>3</mml:mn></mml:msub></mml:mrow></mml:math>on hydrogenated<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:mrow><mml:mi mathvariant="normal">H</mml:mi><mml:mo>:</mml:mo><mml:mi mathvariant="normal">Si</mml:mi><mml:mrow><mml:mo>(</mml:mo><mml:mn>100</mml:mn><mml:mo>)</mml:mo></mml:mrow><mml:mtext>−</mml:mtext><mml:mrow><mml:mo>(</mml:mo><mml:mn>2</mml:mn><mml:mo>×</mml:mo><mml:mn>1</mml:mn><mml:mo>)</mml:mo></mml:mrow></mml:mrow></mml:math>from first principles}},
  booktitle = {Physical},
  year = {2007},
  doi = {10.1103/physrevb.76.245309},
}
Figure