EE527 Project - University of Washington



EE539A: Physics and Modeling of VLSI Fabrication

Susan Soggs and Rebecca Powell

June 13, 2003

Modeling and Simulation Feasibility Study of the Bosch DRIE (Deep Reactive Ion Etch) Process as performed in the WTC (Washington Technology Center)

Overview

DRIE (deep reactive ion etch) processes such as the Bosch process, are necessary for the development of many MEMS devices. The physics behind these processes are not yet well established or analyzed. In fact, most of the existing literature on the Bosch process, in particular, was found to be quantitative rather than qualitative. Due to the complexity of this process, including interaction of the process steps, full analytical modeling would be complex. In this report, the main processes are defined, simplified models developed, and initial simulation results using the Taurus Topography software suite are presented.

Introduction

Micro-electrical-mechanical system (MEMS) devices are created from Si and polysilicon using mature manufacturing materials and processes initially developed for the Semiconductor IC industry. This is possible due to the inherent material strengths of Si for small-scale systems(1,2,3). However, the special needs of MEMS devices require certain additional processes not required in IC manufacturing. Since MEMS is an emerging industry, these processes are less mature and well-understood. An example of this is DRIE.

Standard RIE, reactive ion etching, of Si is widely used in the IC industry. It utilizes a synergistic balance of chemical etching and physical etching to produce Si trenches and structures with aspect ratios on the order of 5:1(4,5,6,7). However, MEMS devices may require extremely tall structures or extremely deep trenches, often on the order of 30:1. Standard RIE is incapable of meeting this requirement. The group of processes used to create these high aspect ratio structures are called DRIE(1,2).

Process simulation predicts the feasibility of process changes and integration of new devices without the time and expense of initial prototyping. Thus, a predictive model of the Bosch DRIE process used in the WTC would be useful for the growing MEMS development community at the University of Washington. However, the majority of existing literature on the Bosch process is quantitative, e.g. results of process recipe changes on etch profiles, rather than qualitative models. Models may exist internal to MEMS design houses such as Bosch, but if so they are unlikely to be shared with the University. Therefore, this project explores the feasibility of building a useful simulation tool for the Bosch process performed in the WTC as an aid to active MEMS development at the University of Washington.

The Bosch Process

The Bosch process is patented by Franz Larner and Andrea Schilp of Bosch GmbH(12). The patent sets forth the concept of multiple cycles of alternating etches and depositions, advancing the trenches formed in small increments until the desired aspect ratio is reached. The deposition serves to protect the sidewalls during each etch step, which causes the trench bottom to be selectively etched to extend the aspect ratio. This concept is illustrated in Figure 1.

[pic]

Figure 1: Process steps of Bosch DRIE

The Bosch process thus has an overall deep anisotropic profile, although each individual etch step may be more or less isotropic depending on the process conditions. The resulting profile has a characteristic “scalloping” of the sidewalls, as the SEM image in Figure 2 illustrates. This may make the original process developed at Bosch unsuitable for some applications, and sidewall optimization efforts are ongoing in various laboratories and foundries(7,8,13). The resulting “Bosch-like” processes vary widely.

[pic]

Figure 2: Characteristic Bosch profile “scallops”.

Each scallop represents one dep/etch cycle (SEM courtesy Kerwin Wang)

The patented dep/etch cycle concept is licensed through each equipment manufacturer and is part of the equipment purchase price (3,8). According to literature, there are at least three equipment manufacturers licensing the Bosch process, all HDP (high density plasma) systems in different configurations(7,8,12,14). Each new equipment configuration, e.g. gas delivery and plasma uniformity, can have different process results. The varied equipment configurations as well as process conditions cause development of an overall model of the “Bosch Process” to be challenging. Therefore, the focus in this project will be on modeling the process as done in the WTC at the University of Washington.

The Bosch etch process at the WTC is performed in an Oxford Instruments ICP 280 system, an inductively coupled HDP etch tool. The system is capable of two different DRIE processes: the Bosch process and the Cryogenic (sometimes called the “Black Si”) process(8,9). Although both DRIE processes are useful for different MEMS applications, the most widely used process for high aspect ratio MEMS devices at the University of Washington is the Bosch process(3). Figure 3 shows an example of structures processed with the Bosch etch in the WTC.

[pic]

Figure 3: High Aspect Ratio Results of the Bosch Process

(SEM courtesy Kerwin Wang)

The WTC process uses SF6 for the etch step, and plasma- polymerized c-C4F8 for the deposition step, at 25oC. The process parameters are varied to obtain the desired results; It may take several repetitions to find the optimal profile depending on the purpose of the structure. One particular issue is that the process is known to be sensitive to open areas of Si, i.e it exhibits pronounced loading effects(4,8,9). Thus each new mask must be optimized before creating usable structures.

The deposition and etch steps are performed separately. In the deposition step, plasma breaks apart the strained cyclic hydrocarbon c-C4F8 into highly excited fragments. The individual fragments react one with another on the exposed surface and build up a more or less strongly cross-linked layer of polymer (15). This is called plasma polymerization. Although many neutral and ionic species are produced, the highest fluxes of species at the surface during deposition have been measured to be CF2 and C2F4. The general chemical reactions are:

[pic]

The deposited polymer (CF)x is essentially Teflon. Note that other species created in the plasma are etching species, which serve to erode the polymer surface. This is an example of simultaneous dep/etch, and the balance of conditions determines whether the mechanism tips toward etching or deposition.

The etch step selectively attacks Si while the deposited polymer protects the sidewalls. The main chemical reactions in the gas phase of the SF6 etch are:

[pic]

At the wafer surface, Si is etched by the following reaction:

[pic]

Ion fluxes at the wafer surface are relatively low compared to the F flux, therefore SiF6 is a primarily chemical etch producing an isotropic profile. Continuity equations at the surface can provide us with an analytical expression for the etch rate, as follows.

[pic]

There are many other reactions in the plasma, but it has been reported that tthe one producing SF3 accounts for about 2/3 and that producing SF2 accounts for 1/3 of the released etching species F(12). Solving for gas phase concentration of SF6 and F:

[pic][pic]

[pic][pic]

Therefore, the etch rate at the surface is:

[pic][pic]

In spite of the rounded profiles, F chemistry is preferential to other etch chemistries (e.g. Cl) due to the high volatility of SiF4, the Si etch product(1,2,13). This is necessary for deep features, as the presence of less volatile etch products in small feature sizes can inhibit etch rate(11).

The preferential etching of the bottom of the trench compared to its sides can be modeled several ways. Some literature suggests that the polymer is selectively deposited on the vertical sidewalls rather than on horizontal features(7,8,13) as part of the dep/etch process balance. This would create a very thin polymer film on the bottom of the trench as compared to the sidewalls. Other sources indicate that polymer deposition is uniform, and that there is a third “breakthrough” etch step through the polymer from the bottom of the trench exposing Si to subsequent processing by the mostly chemical SiF6 etch mechanism. This third step could be included as part of a two-part etch step, and at least one laboratory is developing a process with three separately controllable process steps(14).

Process modeling can be broken into two pieces. The first is reactor modeling, i.e. fluid dynamics/ thermodynamics of the gas-phase. This would include gas delivery, reactor volume, and temperature. Reactor modeling helps describe larger scale effects such as across-wafer uniformity, and is useful to optimize equipment design. The second aspect of process modeling focuses on localized effects, such as the evolution of a surface due to incident material fluxes. An inter-related model that describes the relationship of equipment level parameters and the localized model is an ideal; such a model would enable virtual prototyping to optimize processes and design new devices.

SiF6 and the c-C4F8 plasma reactions by themselves have been modeled fairly extensively(15,16). Modeling of the Bosch process would involve the integration of the two models for deposition and etch, and include a mechanism(s) to account for the removal of the polymer from the bottom of the trench during each cycle. In addition, overall etch profile results, including features such as footing at the bottom corners of free-standing structures, are very likely to involve non-linear interactions between the separate steps.

Surface Flux Model

The physical processes of deposition and etch are a function of the material fluxes of reactants at the wafer’s surface. A simplified view of in a plasma-enhanced reactor, neglecting the gas boundary layer near the surface, is shown in Figure 4.

Figure 4: Material fluxes at the surface

Equation 1 describes the sum of all fluxes acting at the surface:

[pic]

A diffusion flux term may also be added in the case of higher temperature processes such as furnace CVD. This can be neglected here since plasma processing is performed at relatively low temperatures.

Essentially the same description of fluxes can be used for deposition as for etching; the differences lie in the balance between the fluxes and the behavior of chemical species. Simulation requires that each flux be described mathematically and balanced to properly reflect the system being modeled.

The first two terms in Equation 1 describe fluxes of species arriving directly from the reactor to the wafer surface. These can be calculated from the concentration and velocity of gas at the surface (4,6). There are several models for describing the direct flux arrival(4). A simple method with good results for most cases is the cosine distribution of flux model, shown in Figure 5.

[pic]

Figure 5: Cosine Distribution model

Surface modification rate at an arbitrary point S on the wafer surface depends on the normal component, or the cosine of the angle, of incoming flux to the surface. Near the surface, the distribution of the arrival flux follows a cosine distribution. If Fo is the flux toward the surface, and Fs is the flux normal to the surface at point S, then

[pic]

Therefore, point S sees a maximum incoming flux at θ’0o. As θ increases, flux decreases until a minimum is seen at θ’90o. In plasma systems, the incident flux is more directional than a simple cosine distribution. A cosnθ distribution is used, and the generalized flux expression becomes (4).

(3) [pic]

The direct neutral flux is the normal component of incoming chemically reactive flux from the gas phase. The direct ion flux is the normal component of the incoming physically reactive flux. The parameter n describes the result of system pressure and thus mean-free path of the incident flux toward the surface. As n increases, the directionality also increases. This concept is illustrated in Figure 6(1,4).

[pic]

Figure 6: Distribution of arrival fluxes (left) cos distribution (right) cosn distribution

The third term in Equation 1, the emitted flux term, considers the fact that not all molecules stick where they arrive at the surface, and those which do not stick are re-emitted. The probability that an incident flux will remain is described by the sticking coefficient Sc. The emitted flux term is then:

[pic]

Sc is defined as the ratio of the number of incident atoms that actually stay or “stick” on the surface relative to the total number of incident atoms(1,3,4). Related to the material properties of the reactant gas(es) and to some extent equipment configuration, Sc is here modeled as a constant for a particular system, which is often an good approximation. More advanced models take temperature and local area effects into account. Figure 7 illustrates the Sc concept.

[pic]

Figure 7: Sticking coefficient effects on profile evolution:

(left) high Sc, (right) low Sc; (top) deposition, (bottom) etch

On the left, incident species with a high sticking coefficient react where they strike. On the right, species with a low sticking coefficient may not “stick” where they first strike but may be re-emitted to re-deposit elsewhere. Emission/re-deposition can occur multiple times.

As can be seen, deposition species with a high Sc produce less conformal films, those with a low Sc result in more conformal films. Etching species with a high Sc produce more anisotropic etch, while those with a low Sc result in a more isotropic etch. Generally, ions are assumed to stick (SC=1) and chemically reactive neutrals are assumed to have an Sc ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download