Multi-scale modeling of chemical vapor deposition processes for thin film technology

Abstract

Chemical vapor deposition (CVD) process and equipment models aim at relating set macroscopic process conditions (such as the used gases and their respective flowrates, the operating pressure and temperature) to macroscopic and microscopic film properties (such as thickness uniformity, conformality, crystallinity and morphology, and chemical composition and purity). The first analytical models, aiming at the prediction of growth rate and uniformity only, were published in the 1970s. Since then, multi-scale, multi-physics models have been developed, based on advanced computer simulations and aiming at the comprehensive description of all relevant physical and chemical phenomena occurring at length scales from sub-micrometer to meters. CVD process and equipment simulation have been applied successfully not only to optimize hydrodynamic reactor designs with respect to deposition rate and uniformity, but also to predict and control formation and transport of particles, to scale-up existing reactors to larger wafer diameters, to optimize deposition conformality, to predict doping rates, to evaluate loading effects, and to study selectivity loss. Today, CVD simulation models are being used routinely by process engineers and equipment manufacturers, and many modeling features needed for comprehensive CVD simulation have been included in commercial software codes.

In this review paper, the developments—over the past decades—in comprehensive multi-scale simulation of CVD processes and equipment are described and illustrated with examples from the work performed in the authors’ group.

Introduction

Chemical vapor deposition (CVD) constitutes an important technology for the manufacturing of thin solid films, e.g. in semiconductors and solar cells, as anti-reflection and spectrally selective coatings on optical components, and as anti-corrosion and anti-wear layers on mechanical tools and equipment [1], [2], [3], [4], [5], [6]. But CVD is nowadays also widely applied in for instance the manufacture of MEMS [7], [8], [9] and nano-structures [10], [11], [12], [13].

Compared to other deposition techniques, such as sputtering, sublimation and evaporation, CVD is very versatile and offers good control of film structure and composition, excellent uniformity and sufficiently high growth rates. Perhaps the most important advantage of CVD over other deposition techniques is its capability of conformal deposition, i.e. the capability of depositing films of uniform thickness on highly irregularly shaped and patterned surfaces. This makes CVD techniques particularly suitable for the manufacturing, coating and modification of small (nano- to sub-micron scale) structures (see Fig. 1).

The ultimate goal of a CVD process is to produce thin films or structures that possess the particular optical, mechanical, electrical or chemical properties as required for the specific application. These properties result from the chemical composition and physical structure (e.g. thickness, crystallinity, morphology) of the deposited material. In order to realize the desired physico-chemical properties of the material, a suitable substrate surface must be exposed to a suitable growth environment, in terms of its gas temperature, pressure and chemical composition. Next to substrate selection and surface pre-treatment, the main task of a CVD process designer is therefore to create the required conditions in the gas phase close to the substrate surface.

In CVD literature, it is often assumed that a CVD process and the resulting film properties are fully determined by the choice of precursor gases and carrier gas, their respective flow rates, the total pressure in the reactor, and the substrate temperature. An underlying implicit assumption is that these process parameters fully determine the gas phase conditions near the substrate. A majority of CVD literature is therefore devoted to the selection and development of suitable precursors, and to reporting film growth rates and physical and chemical properties of deposited materials as a function of gas flow rates and pressure and temperatures in the reactor. However, from the early days on, CVD literature has been full of observations that cannot be explained from such a view on CVD processes: deposition rates, uniformities and film properties are found to change when, for instance, one inert carrier gas is replaced by another [14], a substrate of different size or a different reactor loading is applied [15], or an identical process is operated in a different reactor setup; non-symmetric film thicknesses are deposited on wafers in axi-symmetric reactors [16].

The explanation of the mentioned phenomena is in the—strongly reactor dependent—gas flow and heat and mass transfer that take place in a CVD reactor, and their complicated interaction with gas and surface chemistry. Attempts have been made to circumvent these complications, particularly in experimental setups used for the determination of intrinsic CVD kinetics, by striving for ‘gradient-less’ or ‘perfectly stirred’ CVD reactor environments [17], [18]. However, the driving forces needed to create a flux of reactant gases towards the substrate lead to the presence of gradients even in such specially designed research tools. This is even stronger the case for (large scale) industrial production reactors. Even under the so-called kinetically limited growth conditions, temperature and concentration gradients within such reactors make the relation between the set process conditions and the actual process conditions near the substrate surface ambiguous.

The experimental consequences are large: kinetic data obtained in one reactor cannot be reproduced in another reactor; processes developed in the laboratory are difficult to scale-up to industrial scale; process conditions reported to lead to good film quality in one reactor fail to do so in another reactor; reactor designs claimed to produce uniform films on $200\mathrm{mm}$ wafers are found to be inadequate for $300\mathrm{mm}$ wafers, etc. Also, in extracting kinetic rate constants from experimental data one should be aware of the fact that many kinetic experiments have suffered from gas transport influences.

It has been recognized in an early stage [1] that simulation models based on computational fluid dynamics, accounting for the interaction between gas flow, heat and mass transfer and chemical reactions, can be of great help in the optimization of CVD equipment and processes. Besides, CVD simulation models may provide fundamental insights into the underlying physico-chemical processes, and can be used in the interpretation of experimental data and in relating local operating conditions to film properties.

In this paper, the developments in computational modeling of CVD processes and equipment in the past three decades will be reviewed, and current developments and challenges will be described. Developments and challenges will be illustrated by examples taken from the work of the author and his co-workers.