Numerical simulation of chemical vapor deposition processes under variable and constant property approximations

Wilson K.S. Chiu, Yogesh Jaluria, Nick G. Glumac

Research output: Contribution to journalArticlepeer-review

Abstract

The chemical vapor deposition (CVD) process for silicon, using silane (SiH4) with hydrogen (H2) as the carrier gas, is modeled numerically using constant properties evaluated at various reference temperatures T r e f. Results are compared with those from a numerical model based on variable transport properties. When the susceptor is isothermally heated, deposition rates predicted by the simplified model agree very well (5% error) with the variable property solution. A susceptor heated by means of a uniform heat flux input has a large temperature variation across the susceptor surface, yielding considerable error from the constant property model. However, a carefully chosen T r e f for cases with large heat flux input, which gives rise to diffusion-controlled deposition (surface Damkohler number Das ≫ 1), is able to capture property variation effects and predict the deposition rate with reasonable accuracy. A variable property model is necessary at low heating rates, since reaction-controlled deposition (Das ≪lt; 1) has a strong dependence arising from exponential temperature dependence of the chemical reactions and the properties. The study shows that the constant property model may be used to obtain solutions with satisfactory accuracy for a variety of operating conditions. The results and observations may be used as guidelines for future CVD reactor design and choice of appropriate operating conditions.

Original languageEnglish (US)
Pages (from-to)113-132
Number of pages20
JournalNumerical Heat Transfer; Part A: Applications
Volume37
Issue number2
DOIs
StatePublished - Feb 11 2000
Externally publishedYes

ASJC Scopus subject areas

  • Numerical Analysis
  • Condensed Matter Physics

Fingerprint Dive into the research topics of 'Numerical simulation of chemical vapor deposition processes under variable and constant property approximations'. Together they form a unique fingerprint.

Cite this