The disclosure is generally related to systems and methods for predicting fractures in crystalline microstructures.
Crystalline microstructures include micro-electromechanical systems (MEMS), monolithic integrated electronic and optoelectronic devices (e.g. optical ridge waveguides), solar cells, and other structures that are made from crystalline materials and are small in at least one dimension (e.g. thickness). Throughout the specification, MEMS are used as examples of crystalline microstructures and the definition of “MEMS” is expanded to include structures that do not necessarily include moving parts or electrical actuators, but that are made from crystalline bulk materials. Further, a crystalline microstructure or MEMS device (according to the expanded definition of “MEMS”) has at least one “small” dimension that accentuates the importance of surface properties in determining the structural characteristics of the device. For example, a solar cell may be many centimeters in length and width, but less than a millimeter in thickness. The small ratio of thickness to length and width means that surface properties of the solar cell are important determinants of its mechanical strength, and therefore that it is considered to be a “microstructure”. The most common crystalline microstructures, however, are MEMS devices in the traditional sense of that term.
Micro-electromechanical systems (MEMS) are mechanical devices roughly ranging in size from micrometers to millimeters, although the term “MEMS” is generalized somewhat in the specification. A subset of MEMS, albeit the majority of MEMS, is silicon MEMS, meaning devices that are fabricated from silicon wafers. Examples of silicon MEMS devices include accelerometers used to trigger airbag deployment in cars, and chips containing millions of microscopic mirrors that modulate light in projection displays.
Silicon MEMS began with the realization that silicon can be used as a mechanical material rather than (or in addition to) as an electronic material. The growth of silicon MEMS technology has been aided by the availability of silicon processing tools originally developed for the semiconductor integrated circuit industry. More recently, tools such as deep reactive ion etchers have been built specifically for MEMS fabrication.
Silicon wafers, the starting point for silicon MEMS device fabrication, are essentially perfect crystals in their bulk, but may have varying surface properties. A polished silicon surface has different properties than an etched surface, for example. In particular, different surfaces are characterized by different imperfections or deviations from the perfect bulk crystalline lattice. The surface properties of micro-scale devices are especially critical to performance because the surface to volume ratio is greater for small devices than for large ones.
Silicon and other crystalline materials are brittle. They can be stressed repeatedly to less than their breaking point with no ill effects. But once a silicon structure breaks, it is permanently destroyed. (In the specification “fracture” and “failure” are used interchangeably.) Structures made from brittle materials fail catastrophically when applied stress exceeds the strength of the weakest part of the structure. In silicon MEMS the maximum stress often occurs on a surface rather than in the bulk material and the weakest link is most often a surface flaw.
Currently, the reliability of silicon MEMS devices can be estimated from the results of destructive testing of finished devices. Simply put, one makes devices and then breaks them. This approach is inefficient because it leaves engineers without reliability data until late in the development cycle when a device has been fully designed and built. Furthermore, if problems are discovered during reliability testing, there is no assurance that whatever changes are made will have the desired effect. One must repeat the design-build-break cycle to see if modifications are successful. The duration and expense of this mode of development leads to designs that may be unnecessarily bulky as engineers try to avoid the risk of having to run extra development cycles.
What is needed is a method to predict failures in crystalline microstructures. Ideally, such a method would be sensitive to varying surface properties of crystalline materials, and offer predictive data before complex devices are fabricated.
The drawings are heuristic for clarity.
A method for predicting the failure of crystalline microstructures is described. Using the method, one may predict the probability of failure of a structure as a function of applied load. For example, using the method, one might discover that a certain MEMS device has a 50% probability of failure when it is loaded by a five Newton force applied in a particular direction to a specific part of the device. Predictions may be obtained for different device designs or different loads without having to build an actual device.
The method described herein depends on the fabrication and destructive testing of test samples that are prepared according to the same manufacturing processes (e.g. etches, depositions, anneals, oxidation, epitaxy) as those that are contemplated for building an actual device. The test samples are very simple compared to typical MEMS devices, and are therefore inexpensive and quick to build. Furthermore, the test samples are used to characterize manufacturing processes, not particular devices, so data obtained from test samples is valid for a wide range of device designs made according to a particular process flow.
Once manufacturing processes have been characterized, that information is combined with the results of a numerical simulation to predict failures in devices. An assumption underlying the method described herein is that the bulk material (e.g. a silicon wafer) from which a device is made is an essentially perfect crystal. In other words, all defects that affect device reliability are assumed to be surface defects. Therefore, the failure prediction method takes into particular account differences in surface properties caused by different manufacturing processes. For example, an etched surface is treated differently than a polished surface.
Surfaces prepared by various manufacturing processes, such as etching, polishing, deposition, thermal annealing, etc., can be characterized in terms of statistical distributions. Just as test scores for students in a class can be represented by a bell curve, the fracture strengths of a set of test surfaces can be represented by a statistical function known as a Weibull distribution. A statistical distribution function is used to characterize a manufacturing process applied to a silicon surface because, even in ideal circumstances, the exact shapes and sizes of flaws left on the surface by the manufacturing process differ among identically prepared test surfaces. The Weibull distribution is appropriate for “weakest link” phenomena such as the failure of brittle materials.
Test samples are designed to isolate the effect of surface flaws on failure probability while normalizing other variables. Often, test samples are made in geometries which lend themselves to closed form analysis of stresses and surface effects. Using results from simple test samples, failure probability may be computed per unit area of stressed surface. The normalized data can then be applied to complex surfaces that have been divided into small pieces corresponding to nodes in a numerical simulation. Numerical simulation replaces closed form analysis for structures that have complicated geometry and different types of surfaces.
Step 120 is calculating the failure probability per unit area at each node on each different type of surface of a device. Step 125 is calculating the failure probability for each surface of the device by performing a sum over nodes: Pj=[exp[−ΣΨiΔAi]]j where ΔAi is the area associated with node i. Step 130 is finding the total probability of failure of the device by multiplying together the failure probability of each surface: P=πPj. Each of the steps outlined in
σf=3FL/4wt2
In an experiment, a set of 50 beams like those shown in
Each of the 100 test beams was loaded (i.e. force ‘F’ in
The data shown in
Empty areas in
A numerical model is used to calculate the principal stresses at nodes on the surface of a device under load. Depending on the material from which the device is made, fully anisotropic material properties may be modeled. For silicon, orthotropic material properties are defined for the bulk crystal from which a device is fabricated. However, in cases where solution speed is critical, such as parametric design optimization, isotropic material properties may be used. In silicon, making the assumption that material properties are isotropic leads to errors in the calculated stress distribution of approximately 3% or less.
Conventional finite element modeling practice or other numerical techniques may be used to calculate the stress distribution in a device with a load applied. (In simple cases a closed form solution may be available.) For example, a model could be run separately for loads applied at points “1” and “2” in
Calculations of the tensile stresses at surface nodes are now combined with characteristics of different types of surfaces obtained from test specimens. Let us define ntype as the number of different surface types associated with a particular device to be analyzed. For example, if a device has only polished silicon surfaces and etched silicon surfaces, created by a single type of etch, then ntype=2. The analysis of such a device requires ntype sets of Weibull parameters σ0, m, and A0. These sets of parameters are differentiated by subscript k: σ0k, mk, and Δ0k.
Next, let us define narea as the number of areas on the surface of the device associated with a particular surface type. The surface of the device may be subdivided into as many areas as surface types or into more areas as needed for computational convenience. Finally, the number of nodes on a single surface area is defined as nnode. The value of nnode for a given surface area depends on the size of the area and the density of the numerical mesh.
The scalar probability of failure per unit area, Ψi, at each surface node on a surface of type k is then given by:
The use of an isotropic failure probability function is justified in silicon because the fracture toughness does not vary much with crystal axis orientation. Anisotropic failure probability functions may be required for other materials. The scalar failure probability at each surface node is then given by:
fi=ψiΔAi
where ΔAi associated with a node is illustrated schematically in
Here, integrating the failure probability over the surface of a device is approximated by summing over surface nodes. The total failure probability is given by the product of failure probabilities for each surface of each type. (Note: ea+b=eaeb.) The corresponding probability of survival, or reliability is given by:
R=1−Ps.
The result of these calculations is the probability of failure of a device under a certain load. In order to test the accuracy of this method of estimating failure probability, failure predictions for an actual MEMS device were compared to data obtained by loading many copies of an actual device to failure; i.e. by breaking them.
Silicon micro mirrors were fabricated according to the design shown in
The method for predicting fractures in crystalline microstructures so far described may be applied to any structure in which flaws affecting the structure's strength are located on a surface rather than in the volume of the material from which the structure is made. This condition may apply to any micro-machined structure made from a single crystal material. Although the method has been described in terms of an example application to MEMS structures, it is equally applicable to integrated circuits, semiconductor lasers, solar cells, or other structures made from a crystalline material.
Examples of materials available in single crystal form that are used in MEMS and other microstructures, and that are amenable to the method for predicting failures described herein include: silicon, germanium, and silicon germanium; quartz, silicon carbide, diamond, graphite, and sapphire; gallium arsenide, indium phosphide, indium gallium phosphide, indium gallium arsenide phosphide, and other so called “III-V” compound semiconductors; cadmium selenide, cadmium telluride, and other so called “II-VI” compound semiconductors. Lithium niobate, potassium dihydrogen phosphate and other electro-optic materials are also amenable to the method.
Test specimens used to gather data on probability of failure versus fracture stress may be four-point bend beams, three-point bend beams, cantilevers, rings, pull tabs or other shapes. Maximum stresses in four-point bend beams can be calculated in closed form, but that is not a requirement for test specimens. When closed form solutions for stresses in test beams are not available, numerical methods such as finite element analysis may be used.
Statistical distribution functions used to fit data obtained from a population of test specimens may be two-parameter Weibull distributions, three-parameter distributions, or other statistical distribution functions.
The origin of stresses developed in either test specimens or actual devices, or both, may be structural (i.e. the result of an applied mechanical force), thermal, electrical, magnetic, fluidic, chemical. Thermal stresses are induced by heating or cooling a device while electrical or magnetic stresses are induced by applying electric or magnetic fields to the device. Fluidic stresses result from pressure applied to a structure by a fluid. Chemical stresses result from chemical reactions acting on a device. Furthermore, combinations or superpositions of these and other stresses, such as those derived from coupled or multiphysics processes may be modeled. Forces and stresses may be time varying and dynamic, and include impulsive shocks. Repetitive forces may induce resonances and resonant enhancement of stress in a structure.
The method described herein may be extended by removing isotropy assumptions in calculations of stresses and/or failure probability. Furthermore, while the example described in detail above involved two different manufacturing processes (polishing and etching), the method is applicable to any number of manufacturing processes. Similar etches performed by different etch tools or using different settings of a single tool may be considered different manufacturing processes.
When closed form solutions for stresses are not available, calculations in the method described herein may be implemented as modules (e.g. libraries, programs, scripts) in numerical analysis packages (such as ANSYS, ABAQUS, NASTRAN/PATRAN, COMSOL, FEAP, ALGOR, etc.) or as modules written in general programming languages (such as C, Verilog, SystemVerilog, FORTRAN, MATLAB, Mathematica, etc.).
Modules can also be designed to import or export fracture stress distribution data that describe the surface condition resulting from use of a particular process recipe (e.g. control parameters), on a specific processing tool, to fabricate the microstructure. For example, for deep reactive ion etch, data corresponding to recipes for process tools made by STS, Alcatel, Plasmatherm, Centura, Lam or other manufacturers may be included. These modules can be used to add microstructure failure prediction to the capabilities of CAD programs such as those sold by Tanner EDA, Coventor, Intellisense, Cadence, Mentor Graphics and others.
Calculations in the method herein described may be implemented as a computer program that offers output in many forms. Described so far are plots of probability of failure versus applied load. Other forms of output include: plots (optionally including confidence intervals) of safe maximum service stress for a prescribed survival rate; contour plots over the surface of a device of the reliability estimate for a given applied load; contour plots combining stress and reliability data; device reliability as a function of input design parameters; and device reliability as a function of input loading. Examples of output generated by the method are given in
For example,
Furthermore, software may include the ability to automatically detect which surfaces in a MEMS device are prepared by which manufacturing processes (e.g. specific etch processes) and to suggest design changes to improve device reliability. All computer programs or software may be supplied on computer readable media.
As one skilled in the art will readily appreciate from the disclosure of the embodiments herein, processes, machines, manufacture, means, methods, or steps, presently existing or later to be developed that perform substantially the same function or achieve substantially the same result as the corresponding embodiments described herein may be utilized according to the present invention. Accordingly, the appended claims are intended to include within their scope such processes, machines, manufacture, means, methods, or steps.
The above description of illustrated embodiments of the systems and methods is not intended to be exhaustive or to limit the systems and methods to the precise form disclosed. While specific embodiments of, and examples for, the systems and methods are described herein for illustrative purposes, various equivalent modifications are possible within the scope of the systems and methods, as those skilled in the relevant art will recognize. The teachings of the systems and methods provided herein can be applied to other systems and methods, not only for the systems and methods described above.
In general, in the following claims, the terms used should not be construed to limit the systems and methods to the specific embodiments disclosed in the specification and the claims, but should be construed to include all systems that operate under the claims. Accordingly, the systems and methods are not limited by the disclosure, but instead the scope of the systems and methods are to be determined entirely by the claims.