OPC modeling game changer: Rigorously-tuned compact modeling
Model-based optical proximity correction (OPC) has been an effective application to help achieve design intent in the lithographic process since the late 1990’s with the introduction of the 180 nanometer (nm) technology nodes, but with continued push of the current ArF immersion imaging hardware into the sub-20nm nodes, traditional OPC modeling approaches are struggling to keep pace. The quality of the OPC results is dependent on the accuracy and predictability of the model, which is heavily reliant on the input metrology data and physicality of the model terms.
Accuracy is defined here as the ability of the model to achieve a true value within a constrained range of design variance, whereas, predictability is the ability of the model to achieve a true value outside of this constrained range. A good allegory for accuracy and predictability is when your weatherman utilizes and tunes a model to provide the weather forecast. If he tunes his weather model for a specific metropolitan area near the ocean, it is pretty accurate for that region, but this same model would not be able to predict weather for another metropolitan area in the mountains, for example. The reason for this is that the weather model terms and data utilized were ocean-centric and it neglected the required model terms and data for mountainous regions.
With the introduction of sub-20nm nodes, OPC models are failing to provide the accuracy and predictability as efficiently as previous nodes for two reasons. First, the resist processes and materials have become more complex and yet the traditional OPC model terms have remained very simplistic, which is great for runtime friendliness but fails to provide the predictability required for advanced nodes. Second, the traditional metrology technique (top-down scanning electron microscope or SEM) is failing to provide realistic data due to non-ideal resist profiles, which results in inaccurate models during the tuning process. These two issues require innovative solutions to sustain OPC model accuracy and predictability beyond the 20nm nodes.
The metrology accuracy problem
The purpose of metrology for OPC models is to capture the most critical dimensions (CD) of the design, which assists in the model tuning process and helps to converge each term within the model. These CDs are typically taken with a top-down SEM, which makes an assumption that the backscatter electron-beam energy signal correlates to a particular 2D width (or CD) of a resist profile as shown in Figure 1. This metrology technique has been the primary method for generating data for OPC models since the 1990s.
Figure 1. The process of collecting critical dimensions from resist profiles using a scanning electron microscope. The electron beam signal provides a good top-down 2D width for an ideal resist profile.
Top-down SEM metrology has been a sufficient technique down to the 20nm node since the resist profiles have been smooth conical shapes up to this point, but at 20nm and below these profiles have begun to degrade due to low-contrast imaging conditions, leading to marginal process windows. Sub-20nm resist profiles might have a wide variety of shapes that make the top-down 2D SEMs inept for the most critical design features as shown in Figure 2. Here, the most common profile degradation types would result in an incorrect CD if a top-down SEM is utilized. For example, the 3D failures in resist profiles c) and e) could result in a poor model since traditional OPC tuning software assumes a CD threshold at a particular point in the resist height (red arrows) whereas the reported SEM CD would incorrectly report a larger CD because of the profiles’ mushroomed shapes. As for profiles b) and d), the CD SEM might report the correct CD, but it would fail to capture the resist top loss or footing problems, which is importation process information for the post lithographic steps. For example, profile d) might result in good measured and modeled CD at the lithography step, but the resist would punch through during the etch process, resulting in poor yielding hotspots.
Figure 2. Resist profile types that could be encountered in sub-20nm processes. A top-down SEM would fail to capture the 3D failures noted in b) to e) whereas a traditional OPC model acquires the CD at a certain threshold (red bar) within the resist.
The obvious solution is to deploy a metrology technique that provides the 3D resist profile data from the most critical areas to supplement the CD’s derived from a backscattered SEM. There are currently several metrology tools that can provide the resist profile data in addition to a top-down SEM CD: atomic force microscope (AFM), focused ion beam (FIB), and scanning tunneling microscope (STM). These profile metrology tools are typically slower and more costly than traditional top-down SEMs, so there are also options to supplement the metrology profile data with resist profiles from rigorous simulators, like Synopsys Sentaurus Lithography,1 where volumes of 3D resist profiles can be provided with minimal effort2. Both of the 3D metrology and rigorous simulation approaches can provide the required profile data to enhance model accuracy and predictability at the sub-20nm nodes, but this metrology problem is not completely resolved since profile data still cannot be used to improve OPC models because the traditional model tuning software has not been updated to accept this 3D data format.