Pre-bias optical proximity correction

Abstract

A pre-bias optical proximity correction (OPC) method allows faster convergence during OPC iterations, providing an initial set of conditions to edge fragments of a layout based on density conditions near the edge fragments.

Description

FIELD OF THE INVENTION

The present invention pertains to the field of Resolution Enhancement Techniques (RET) in photolithography. More particularly, it relates to the use of two-dimensional models to improve the performance of layout corrections using Optical Proximity Correction (OPC) by providing an improved set of initial conditions to the iterative OPC process.

BACKGROUND

In photolithography, a design is transferred onto a surface of a wafer by exposing and selectively etching a pattern of features onto a photo-sensitive material. Although advanced photolithography techniques routinely fit millions of circuit components onto a single chip, the wavelength of the exposing light is too long to produce undistorted layout replicas. OPC (Optical Proximity Correction, or sometimes Optical and Process Correction when effects other than proximity effects are included as well) is a technique that is used to adjust the mask features so that the transferred pattern will be a sufficiently accurate replica of the intended target.

OPC is a computationally expensive calculation that typically requires an iterative calculation of optical, resist, and etch effects. Typically, a figure of merit, such as edge placement error (EPE), is used in conjunction with a feedback factor to adjust the position of the mask edges for the next iteration. Ideally, each iteration improves the result, and the process continues until the EPE of each edge is near zero.

Since each iteration is a time consuming calculation, it is common to use a rule-based approach to provide an initial correction for edges in the layout. This improves the image quality and provides an approximation of the desired result, which normally reduces the number of model-based iterations required.

The rule-based modifications are typically expressed as tables, which characterize each feature of interest by its width and space (to the closest neighbor). While this method has proven to be sufficient for 90 nm processes and larger, it becomes very complex as feature sizes decrease and is no longer practical for advanced device fabrication. For that reason, there is a need for a technique of providing an improved initial set of conditions to a desired layout.

SUMMARY

The present invention is a technique for calculating bias values for layout features prior to the application of OPC modifications. In one embodiment, a model of the OPC process is calibrated by correlating density values in a test layout and a known OPC result. Once the model is calibrated, density signature values for edges in a desired layout pattern are applied to the OPC model to determine a pre-OPC bias value for the edges.

This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This summary is not intended to identify key features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.

DESCRIPTION OF THE DRAWINGS

The foregoing aspects and many of the attendant advantages of this invention will become more readily appreciated as the same become better understood by reference to the following detailed description, when taken in conjunction with the accompanying drawings, wherein:

FIG. 1 is a graph showing how critical dimensions of features printed on a wafer vary with changes in focus and dose;

FIG. 2 illustrates how a critical dimension of a feature printed on a wafer changes with distance between features;

FIG. 3 illustrates features printed with and without OPC;

FIG. 4 illustrates a conventional technique of determining OPC biases;

FIG. 5 illustrates how different initial conditions of a feature set can produce different OPC results;

FIG. 6 illustrates sample sites in a set of features and a representative two-dimensional function in accordance with an embodiment of the present invention;

FIG. 7 illustrates one method for calculating a density signature for the edge fragments in accordance with an embodiment of the present invention;

FIG. 8 illustrates calculating a density signature for edge fragments in accordance with an embodiment of the present invention;

FIG. 9 illustrates how edge fragments are pre-biased in accordance with an embodiment of the present invention;

FIGS. 10-15 illustrate cross sections of two-dimensional radially symmetric functions that can be used to compute a density signature of edge fragments in accordance with the present invention; and

FIGS. 16-19 illustrate a target pattern, a pre-biased feature pattern, and an OPC corrected feature pattern.

DETAILED DESCRIPTION

If only one type of structure is to be imaged onto a wafer, it is possible to select the process parameters to adjust the critical dimension to the desired target, as illustrated in FIG. 1. However, real world integrated circuits or other devices created by a photolithographic process include multiple types of structures in the layout that exhibit a different proximity behavior depending on their neighborhood. FIG. 2 illustrates that the critical dimension of objects in a layout changes with the distance or pitch between objects and that there is no process condition that can simultaneously image all features with the correct target critical dimension. For that reason, Optical Proximity Correction (OPC) is used to adjust the features on the mask by shifting edges either in or out so that the printed results on a wafer will match the target dimensions and allow otherwise non-printing features to resolve, as shown in FIG. 3.

Before the advent of model-based OPC, proximity correction was a simple matter of defining biases based on near-neighbor features. FIG. 4 illustrates a table of offsets or biases that is very well defined for regular one-dimensional structures (i.e., features having a length much greater than their width), but is non-trivial for locations where a feature is separated by two or more distances to its neighbors. In such instances, either new rules need to be applied or more complex tables need to be generated to capture such cases. In either case, the complexity of managing and finding the correct offsets or biases becomes cumbersome and unreliable.

This complexity accelerated the adoption of model-based OPC, in which such tables are not necessary. However, since most OPC techniques are iterative, these tables have persisted as a way to provide a better initial set of conditions before launching the iterative process, thus facilitating convergence and guiding the OPC towards a more robust solution. The rule-based “hints” enable OPC tools to achieve acceptable results with fewer iterations and, therefore, more quickly. FIG. 5 shows an extreme case (double exposure OPC), in which the aerial image produced by the OPC solution varies widely depending on different initial conditions. In this particular case the only difference between the top split and the bottom split is the improved pre-bias used as input to the OPC tool, thus guiding it to a more robust imaging solution. As this simple example illustrates, it is possible to create pre-OPC bias tables for simple one-dimensional structures, but such approaches tend to be very conservative when applied to more irregular configurations.

An embodiment of the present invention replaces the need to create very large pre-OPC bias tables for all structures by characterizing layout features according to layout density. As FIG. 6 shows, the method places a number of simulation sites 50a-50d, etc., on layout features where the characterization will be performed. Evaluating densities at the same site locations where the OPC tool will compute the figure of merit for an edge fragment is desirable. In one embodiment, the densities are computed by convolving the layout shapes with two-dimensional functions. This might be done for a single function or for multiple functions. As an example, orthogonal top-hat functions are used. However, this method is not limited to top-hat functions and can easily make use of Gaussian, Laguerrian, sinusoidal, or any arbitrary 2D function. Top-hats were selected for their simplicity and because once their coefficients are found, they can represent a discretized version of any arbitrary radially symmetric 2D function.

FIG. 7 shows an example of how a 2D function is evaluated at each simulation site on a feature and provides a “density signature” value for an edge fragment that encodes the nature of the layout neighborhood around the simulation site. It also shows how the functions are normalized over the integration surface. In this case, the reference layer is the target layer, which should include any re-targeting bias and other types of aggressive resolution enhancement techniques such as assist features.

In one embodiment of the invention, polygons defining the desired target layer to be created on a wafer are fragmented to define a number of edge fragments around the perimeter of the polygons. Simulation sites are defined for the edge fragments where the OPC tool will calculate the EPE of the edge fragment.

In one embodiment, the density signature D is a vector (i.e., an array) of numbers having values proportional to the area of a kernel overlapping polygons in the layout.

$\begin{array}{cc}D=\int {\int }_{S}F\left(x,y\right)·L\left(x,y\right)ds& \left[1\right]\end{array}$ where the integration covers all 2-dimensional space S, D is the Density value for a particular site, F(x,y) is the kernel, and L(x,y) a function corresponding to the layout, for example

$\begin{array}{cc}L\left(x,y\right)=\{\begin{array}{c}1\mathrm{if}.\mathrm{within}.a.\mathrm{polygon}\\ 0\mathrm{if}.\mathrm{outside}.a.\mathrm{polygon}\end{array}& \left[2\right]\end{array}$

A kernel may in turn be made up of a set of smaller kernels, such as a set of concentric top-hat functions. For example, if a top-hat based function F(x,y) is used having five different concentric circles (kernels), each centered over a simulation site, the density signature of the edge fragment may have the form[1,D_1,i,D_2,i . . . D_5,i]  [3]where i is the number of the edge fragment, and D₁ represents the percentage of the first top-hat kernel that is covering layout features. D₂ is the percentage of the second top-hat kernel covering layout features, etc. In one embodiment, the area inside a feature is defined as logic 1 and the area outside a feature is defined as logic 0. If a kernel covers layout features over 50% of its area, then the value of D for that kernel is 0.5. In practice, the a top hat function F(x,y) may include 50+ kernels. Different edge fragments typically will have different density signatures depending on the location of their neighbors. In the case of a top-hat function these results can be thought of as representing area but for the more general case of arbitrary kernels it is necessary to realize that a convolution is being done as shown in FIG. 7.

Typically, the functions F(x,y) are normalized

$\begin{array}{cc}\int {\int }_SF\left(x,y\right)ds=1& \left[4\right]\end{array}$ but this is not a requirement for their use. In FIGS. 7 and 8, the rings with different shadings represent that the value at different radial distances can be different. Different normalizations for each kernel and sub-kernal can be different as well.

FIG. 8 shows examples of three simulation sites and the calculated density value, D, at each of the three sites. The kernel is comprised of three concentric top-hat functions, each shown in a different shading If the two-dimensional form of the kernal is monotonically decreasing, then the density value will likewise be a monotonic function of the feature size and spacing. These simple functions do not, however, work well for calculating pre-OPC bias values because the through-pitch behavior of real features is not a monotonic function. To accurately capture and approximate the reference OPC, 2D functions that can sample discretely the layout configurations, such as top-hats or wavelets, are preferred

Before pre-OPC bias values can be calculated for edges in a desired layout pattern, the OPC model is calibrated to a known OPC solution. Therefore, test layout data, including feature patterns and spacings likely to be used in the actual desired layout, is analyzed with an OPC tool to determine how the feature edges should be moved to print as desired. Next, the test layout data is analyzed to determine the density signatures of the edge fragments in the test data. Finally, a mathematical analysis is performed to correlate the various density signatures to the known OPC correction values. The result of the analysis is a model that models the OPC process itself. In one embodiment, the correlation is determined by solving the following system of non-linear equations set forth in Equation 5 below for various coefficients α₀, α₁, α₂ . . . α_m

$\begin{array}{cc}\left[\begin{array}{cccc}1& D_{1,1}& \dots & D_{m,1}\\ 1& \dots & \dots & \dots \\ 1& D_{1,n}& D_{2,n}& D_{m,n}\end{array}\right]\left[\begin{array}{c}{\alpha }_0\\ {\alpha }_1\\ \dots \\ {\alpha }_m\end{array}\right]=\left[\begin{array}{c}{\mathrm{Disp}}_1\\ \dots \\ {\mathrm{Disp}}_n\end{array}\right]& \left[5\right]\end{array}$ from the known density signature values D_m,n and the known displacements (Disp₁, Disp₂, Disp₃ etc) from the OPC results.

Once the OPC model is calibrated from the test layout data and the known OPC solution, the coefficients α_i represent the relationship between the Density values and the displacements that the OPC procedure will produce. A displacement function Disp that determines a pre-OPC bias amount for a feature edge given its density signature D_i can therefore take the form:Disp=α₀+α₁D₁+ . . . +α_mD_m  [6]Other forms are possible, as well, and could include cross terms (such as α₃D₁D₂) as well as optical terms such those used in model-based OPC.

Since the solution of this system provides a mechanism to predict a first order approximation of the OPC, it is expected that these functions will be very different between process layers and process technologies. Composite 2D functions, calibrated with this method, are shown for several 65 nm layers in FIGS. 10 through 15. These figures show cross-sections of an example of a complicated kernel F(x,y). To visualize the kernel itself, imagine that this pattern is rotated around the y-axis to form a surface where peaks and valleys are the peaks and valleys of the cross-section line. The displacement can be found by convolving the 2-dimensional shapes of the desired layout pattern with this kernel. This kernel could be a single arbitrary kernel or it could be the sum of many simpler kernels (multiplied by a coefficient as determined from the calibration process). These examples are for illustrative purposes and the correct kernel will vary from one process to another.

To qualitatively examine the accuracy of the fast pass OPC method, FIGS. 15-19 show examples of the intended target, the real OPC, and the result of the pre-OPC biases calculated. The pre-OPC bias solutions can qualitatively, and in many cases quantitatively, reproduce the morphology changes produced by the final OPC solution. While the result may not provide the accuracy needed for final photomask creation, the results provide a fast way to create an initial condition prior to beginning OPC.

TABLE 1
Runtime comparison between regular OPC and Pre-bias in seconds.
		CPU baseline	CPU Pre-
	Layer	(CPU/Real)	bias(CPU/Real)
	RX	185/55	8/3
	PC	1062/299	43/14
	CA	66/19	10/4
	M1	653/190	32/12

Table 1 summarizes the relative speed of pre-OPC models with respect to traditional (iterative) OPC.

In one embodiment, the present invention is implemented by a networked or stand alone computer system that executes a sequence of computer instructions. The instructions are stored on a computer storage media such as a CD-ROM, hard disc, DVD, flash memory of the like. Alternatively the instructions may be received by the computer system over a wired or wireless communication link. The computer executes the instructions to read a desired layout pattern or portion thereof and to compute pre-OPC bias amounts for at least some edge fragments in the layout in accordance with the techniques described above.

While illustrative embodiments have been illustrated and described, it will be appreciated that various changes can be made therein without departing from the scope of the invention. It is therefore intended that the scope of the invention is to be determined from the following claims and equivalents thereof.

Claims

1. A method of preparing layout data for optical and process correction (OPC), comprising: receiving data representing features of a desired layout to be created photolithographically;designating edges of the layout data for which pre-OPC bias values are to be determined prior to the application of OPC;for at least one of the edges for which pre-OPC bias values are to be determined, calculating a density of nearby features; andusing a computer, applying the calculated density of nearby features to a calibrated model that correlates the density of nearby features to an OPC correction to determine at least one pre-OPC bias value for the at least one of the edges.

2. The method of claim 1, followed by the execution of model-based OPC on the layout.

3. The method of claim 1, wherein the density for the at least one edge is determined by determining an area occupied by the feature of the layout data under a set of concentric functions with varying radii with their centerpoints on the edge.

4. The method of claim 1, wherein the density for the at least one edge is a density signature comprising the area occupied by the features of the layout data under a number of top-hat functions having different radii and with their centerpoints on the edge.

5. The method of claim 1, wherein the at least one pre-OPC bias value comprises an initial condition for the application of OPC to the at least one of the edges.

6. A computer-readable storage medium storing computer-executable instructions for causing a computer to perform a method, the method comprising: receiving data representing features of a desired layout to be created photolithographically;designating edges of the layout data for which pre-OPC bias values are to be determined prior to the application of OPC;for at least one of the edges for which pre-OPC bias values are to be determined, calculating a density of nearby features; andapplying the calculated density of nearby features to a calibrated model that correlates the density of nearby features to an OPC correction to determine at least one of the pre-OPC bias value for the at least one of the edges.

7. The computer-readable storage medium of claim 6, wherein the method further comprises executing model-based OPC on the layout.

8. The computer-readable storage medium of claim 6, wherein the density for the at least one edge is determined by determining an area occupied by the features of the layout data under a set of concentric functions with varying radii with their centerpoints on the edge.

9. The computer-readable storage medium of claim 6, wherein the density for the at least one edge is a density signature comprising the area occupied by the features of the layout data under a number of top-hat functions having different radii and with their centerpoints on the edge.

10. The computer-readable storage medium of claim 6, wherein the at least one pre-OPC bias value comprises an initial condition for the application of OPC to the at least one of the edges.

11. A method comprising: receiving layout data representing a desired layout design or portion thereof, the layout data defining one or more polygons;fragmenting the one or more polygons into one or more edge fragments;determining at least one density signature of the one or more edge fragments; andusing a computer, determining at least one pre-OPC bias value based at least on the at least one density signature.

12. The method of claim 11, wherein the at least one pre-OPC bias value is further based on a set of calibrated coefficients, the calibrated coefficients being determined from density signatures of edge fragments in a test layout and a set of known OPC corrections to the edge fragments in the test layout.

13. The method of claim 11, wherein the determining the at least one density signature comprises convolving at least one of the one or more polygons with a set of radially symmetric functions.

14. The method of claim 13, wherein at least one of the radially symmetric functions is a top-hat kernel.

15. The method of claim 13, wherein the radially symmetric functions are centered over at least one simulation site associated with the one or more edge fragments.

16. The method of claim 11, wherein the determining the at least one pre-OPC bias value comprises multiplying at least one density signature value with at least one calibrated coefficient.

17. The method of claim 11, further comprising determining one or more OPC corrections for the one or more edge fragments using the at least one pre-OPC bias value.

18. The method of claim 17, further comprising modifying the placement of at least one edge fragment of the one or more polygons based on the one or more OPC corrections.

19. The method of claim 18, further comprising creating a photomask based at least on part on the modified placement of the at least one of the one or more edge fragments.

20. The method of claim 19, further comprising fabricating a chip using the photomask.

21. A computer-readable storage medium storing computer-executable instructions for causing a computer to perform a method comprising: receiving layout data representing a desired layout design or portion thereof, the layout data defining one or more polygons;fragmenting the one or more polygons into one or more edge fragments;determining at least one density signature of the one or more edge fragments; andwith a computer, determining at least one pre-OPC bias value based at least on the at least one density signature.

22. The computer-readable storage medium of claim 21, wherein the determining the at least one pre-OPC bias value is further based on a set of calibrated coefficients, the calibrated coefficients being determined from density signatures of edge fragments in a test layout and a set of known OPC corrections to the edge fragments in the test layout.

23. The computer-readable storage medium of claim 21, wherein the determining the at least one density signature comprises convolving the one or more polygons with a set of radially symmetric functions.

24. A system, comprising: means for receiving layout data representing a desired layout design or portion thereof, the layout data defining one or more polygons;means for fragmenting the one or more polygons into one or more edge fragments;means for determining, at least one density signature of the one or more edge fragments; andmeans for determining at least one pre-OPC bias value based at least on the at least one density signature and a set of calibrated coefficients.