Enhanced resist and etch CD control by design perturbation
Summary (2 min read)
1. INTRODUCTION
- RET such as assist feature insertion and OPC are mandatory post tapeout steps to ensure printability of features in sub-90nm technology nodes.
- Since the model provides mask cost as a function of layout parameters, library designers can modify device layout to minimize mask cost without running OPC and fracturing repeatedly.
- Instead, the authors model the response of the OPC algorithm as a function of OPC tolerances and layout parameters.
- Layout dimensions and geometries of devices in standard cells is different from that of wires.
- Hence, MCC approach is different for each.
2. LIBRARY MCC (LMCC)
- Fracture count of a standard cell is a function of OPC tolerances (IT, OT, SSIDE, FRAG) and its layout context.
- MCC for isolated context can be performed by running OPC followed by fracturing on individual standard cells for different IT, OT, SSIDE and FRAG.
- Outer tolerance (OT) specifies the maximum tolerable edge movement outside the drawn feature.
- Fragmentation (FRAG) is one of the parameters that can have a significant impact on fracture count.
- The authors identify layout parameters that are the source of fracture count variation between any two standard cells for a given tolerance combination.
2.1. Layout Parameter Extraction
- The authors explore different characteristics of standard cell layouts that are the source of variation in fracture count for a given tolerance combination.
- The source of fracture count discrepancy is the layout of the cells.
- The authors only consider the average spacing between poly, defined as the ratio of cell width to the number of poly features.
- A simple vertical poly has just four vertices.
2.2. Experimental Setup
- The authors give details of their OPC setup and regression studies.
- The authors construct OPC recipes to run OPC exhaustively on all 15 cells for all combinations of IT, OT, SSIDE and FRAG given in Table 2.
- Based on the fracture count data and layout parameter values for 15 cells, the authors perform regression studies to construct FC model using SPLUS7 software.
- From the plots the authors can observe that PFC is strongly correlated with NP, CW, PVC and PW.
- From the results the authors observe that around 62% of the cells have less than 5% error between predicted FC and actual FC.
3. WIRE MCC (WMCC)
- Wire mask cost (WMC) model predicts the FC of wires before running OPC using pre-characterized models and look-up tables.
- In addition to tolerance optimization, WMC model can be used for guiding wire sizing optimizations to minimize FC.
- The model captures the three major contexts of a wire such as the line-body (L), line-end (LE) and the line-corner (C) as shown in Figure 7.
- Using the FC data, the authors construct closed-form expressions and populate LUTs representing the model.
- Section 3.1 describes FC saturation in detail.
3.1. FC Saturation
- The geometries of wires in the layout along with context can be very complex.
- Parameters for the main pattern start with the letter ’M’ and those of the neighbors start with the letter ’N’.
- The diffraction effects caused by a small neighbor in the vicinity of the main pattern are corrected aggressively by the OPC tool.
- The OPC treatment of line-ends and corners of a wire pattern is different from that of the line-body.
- The general trend in saturation points of various parameters are shown in Figure 10 and their values are summarized in Table 4.
3.2. Line-body, End and Corner Models
- Based on the saturation points for different parameters, the authors construct different configurations of the line-body, line-end and line-corners and run OPC followed by fracturing.
- To compute the slope of the model, the authors construct the test patterns shown in Figure 12(a) which shows two main lines M1 and M2 of lengths ML1 and ML2 respectively.
- Table 6 gives the LUTs for line-end for single and double neighbor cases.
- The presence of wire patterns around the line-end does not change the fragmentation significantly and hence, the authors see a small change in FC with change in neighbor spacing.
- Hence, the authors construct LUTs with CCL and EPEtol as parameters and use it for predicting FC of line-corners in real layouts.
3.3. FC Prediction
- To validate the line-body, line-end and line-corner WMC models presented above, the authors analyze a real layout and predict its FC and compare it with real FC after OPC and fracturing.
- The authors then extract the optical radius (OR) of the OPC model from a line and space test pattern.
- These FCs are used to compute α and populate LUTs.
- To predict FC of real layouts, the predictor decomposes wire patterns from real layouts into the three contexts based on FC windows.
- Table 8 shows the real and predicted FC for metal layer 2 of ALU128 benchmark in the 90nm technology.
4. CONCLUSIONS
- The authors have presented methodologies for characterizing OPC of standard cells and wire patterns in terms of fracture count.
- These FC models can be used by designers to choose between different OPC tolerance combinations to minimize mask cost.
- The placement of standard cells and the type of standard cells surrounding a given cell have significant impact on its FC.
- The authors are currently working on extending the library MCC approach to include the impact of layout context.
- Optimizing line-end corner fragmentation parameters can enable further reductions in FC.
Did you find this useful? Give us your feedback
Citations
217 citations
163 citations
152 citations
139 citations
101 citations
References
281 citations
274 citations
181 citations
130 citations
130 citations
Related Papers (5)
Frequently Asked Questions (12)
Q2. How can The authorperform MCC for isolated context?
MCC for isolated context can be performed by running OPC followed by fracturing on individual standard cells for different IT, OT, SSIDE and FRAG.
Q3. What are the steps to ensure printability of features in sub-90nm technology nodes?
RET such as assist feature insertion and OPC are mandatory post tapeout steps to ensure printability of features in sub-90nm technology nodes.
Q4. What is the procedure for fitting the fracture count of standard cells?
The authors then perform linear regression to fit the fracture count of standard cells as a function of layout parameters for each tolerance combination.
Q5. How do the authors construct a model of the fracture count?
Based on the fracture count data and layout parameter values for 15 cells, the authors perform regression studies to construct FC model using SPLUS7 software.
Q6. How many FCs can be predicted for a given set of library cells?
FC model constructed for library MCC using a limited set of library cells for a given tolerance combination can predict the FC trend of up to 75% of cells in the library within 5% error.
Q7. What is the main reason for the RET?
Doubling of layout data volume every technology node combined with aggressive RET is driving mask set cost to prohibitive levels.
Q8. What can be done to extend the presented models?
RET engineers can extend the presented models by adding other OPC parameters and use it for tuning OPC recipes and optical model parameters.
Q9. What is the basic step for a complete design-aware OPC flow?
59921W, (2005) · 0277-786X/05/$15 · doi: 10.1117/12.633416Proc. of SPIE Vol. 5992 59921W-1Characterization of mask cost and timing impact of different OPC tolerances is the basic step for a complete design-aware OPC flow.
Q10. What is the second approach to minimizing mask cost?
In the second approach, tolerance optimization is performed by choosing OPC tolerance combination specific to the standard cell or wire pattern by analyzing its impact on mask cost and design performance simultaneously.
Q11. What is the significance of the trend in FC?
Even though the predicted FC differs from the actual FC, the trend in FC is useful for performing tolerance optimization, since the designer needs to be aware of the change in FC rather than the absolute value.
Q12. What are the parameters that affect fracture count in a real layout context?
In addition to the parametersProc. of SPIE Vol. 5992 59921W-2described above, fracture count also depends on OPC corrections performed at line-ends and corners.