Don Samuels, Ian Stobert, IBM Microelectronics, Hopewell Junction, NY

When you are using optical proximity correction (OPC) to optimize contacts and vias, the obvious solution is not always the correct solution.

Optical proximity correction is now well-established in the industry to allow for manufacturing in a low k₁ environment. OPC will typically attempt to modify the design shapes to allow for a simulated image to match the design exactly. Modifications to this process by the vendors of OPC software include

• Looking at the simulations through process at dose and focus values other than nominal;
• Altering the design target at corners to give a rounded target for OPC to converge to; and
• Allowing for less convergence at corners than “straight line” fragments.

In all cases, though, the metric is a convergence to a target image. In some cases this is not the correct metric.

Contacts are drawn as squares or rectangles (and in some cases, polygons), not because that is what is desired in the final product, but because that is the simplest representation in GDS II. CAD systems are integer-based systems. Other shapes lead to an increase in data volume and write times. However, minimum square contacts in a low k₁ regime are always going to print as circles, no matter what. All that really matters is that the contact is within a specification of minimum CD, area, and cross-section intersection to the layers above and below. Square contacts on the final product are not required.

An example of a better metric than convergence of the simulation to a target is a requirement that all contacts have the same area after lithography. Different area sizes will lead to degradation of the circuit. In the worst case, this variation in area will cause a time zero failure if the contact does not survive processing downstream of litho.

There are three methodologies used to understand what type of OPC leads to the minimum area variation.

OPC is done on a large array in one of the following ways:

1. No fragmentation, just edge movement with mask rule constraints (MRC);
2. Three fragments per edge with MRC constraints, equal to traditional serifs; and
3. Three fragments per edge without MRC constraints, also traditional serifs.

Typical serif options are illustrated in Fig. 1 along with corresponding simulations. At best dose and focus, the image contours are approximately the same, nearly circular.

The Bossung plots and exposure/defocus trees for all three cases were also essentially identical, other than the dose offset.

Other analysis methods

Changing the type of analysis to one that calculates the area of each contact, and then looking at the variation of the areas, the tables in Fig. 2 can be used to identify the optimum OPC solution among the choices above.

The numbers in each box indicate the range of contact areas occurring for the combinations of dose and focus. A larger number indicates more variation in area, which is detrimental to the performance of the circuit. In these tables, a “99” was used when the simulation indicated that the contacts failed to open.

In an arbitrary checkerboard layout, a different corner treatment may be called for- an inverse serif. In this case, the contact edge is divided up into three fragments, and the center fragment is allowed to move outward, as opposed to a traditional corner serif where the corner fragments move outwards and the center fragment may move inwards.

The objective here, as in the first example, is not to have the simulated image match the drawn image. Figure 3 shows the three new options: Type A, Type B, and Type C.

The three different OPC styles respond differently to variation in the mask size. Figure 4 shows the dependence of the simulated CD on the variation of the mask CD from nominal- in other words, the MEEF, and the new options.

As can be seen, the traditional serif has higher MEEF than the other two, which will lead to greater contact variability on the wafer. The traditional serif is what OPC would require in order to match the drawn design, unless the OPC engineer interferes and forces different metrics and OPC styles.

An inverse serif, however, may be what is needed in this checkerboard type layout. The inverse serif avoids any MRC constraints due to the corner-to-corner distance being the minimum for this layout. Putting in an inverse serif still allows for a circular printed image. It also allows for a lower dose to print on-size due to the greater amount of light passing through the mask opening.

This discussion about dose leads us into one of the more subtle issues associated with choosing an OPC style. Each style of OPC uses a different amount of mask real estate, and for a given dose, each style of OPC will print a different sized contact. To illustrate this, we drew four mask shapes for contacts: The first case was a square mask shape and the others were contacts with traditional outer-corner serifs of varying sizes. In each case, we picked dimensions to make the total area of the contact on the mask fixed. We then used our simulation engine to simulate printing at five different dose values, for each contact in an isolated context. For convenience, the four contacts are shown side-by-side in Fig. 5, even though the simulations were done with these shapes in isolated placements.

Looking at Fig. 5, we see that as the corner serifs become more aggressive relative to the main feature size, the simulated image becomes progressively smaller. As with many of our other simulation results, variability becomes much worse as the simulated image shrinks. Our simulation techniques also show us that for a fixed mask size, variability through-dose increases
dramatically as space between mask shapes is decreased.

Summary

When it comes to contact level OPC, simple is sometimes better. Square and rectangular contacts are good for maskmaking, involve simple OPC recipes, and have excellent characteristics relative to other OPC solutions. When we study variation in dose, focus, and mask sizing, we find that these simple contacts stand up very well.

Contact OPC with inverse serifs may be a good option when we need to print very tightly spaced layouts, where some of the smallest spaces between targets are encountered on diagonals. Their through-process characteristics are very similar to those of square and rectangular contacts.

The temptation to use aggressively serif’d (Type C in Fig. 3) OPC is strong when tying to pattern small rectangular contact targets, because slightly more aggressive aspect ratios are achievable at nominal process conditions. However, the data shows that these benefits may be offset by a variability. Another problem with this style of OPC is that to achieve a fixed size of contact, a much larger area of mask real estate needs to be used.

Acknowledgments

This work was performed at the IBM Microelectronics, Div. Semiconductor Research & Development Center, Hopewell Junction, NY 12533. This work is supported by the independent alliance programs for SOI technology development and bulk CMOS technology development. The authors wish to thank Scott Mansfield, Azalia Krasnoperova, Norman Chen, Brad Morgenfeld, Bernhard Liegl, and Alan Thomas for some helpful ideas and contributions to our learning about contact printing issues, and Alan Leslie for making this work possible.

Reference

R. Ferguson, “Data Analysis Methods for Evaluating Lithographic Performance,” Journal of Vacuum Science, 15(6), pp. 2387-2393, 1997.

Contact Don Samuels at PO Box 2129, 1228 Emerald Rd., Silverthorne, CO 80498; ph 970/513-9728, e-mail samuelsd@us.ibm.com.