1. Native-Conflict-Aware Wire Perturbation for Double Patterning Technology Szu-Yu Chen, Yao-Wen Chang ICCAD 2010

2. Outline Introduction Preliminaries Native conflict prediction NC-Aware wire perturbation Experimental result Conclusion

3. Introduction Foundries have been systematically reducing the printed feature size (half pitch) for years. From Rayleigh criterion, the half pitch is limited to 36 nm, which cannot satisfy the desired sub-22nm technology nodes. The most popular solution for sub-22nm node is double patterning technology (DPT).

4. Introduction Two feature have to be assigned to different masks if the distance of them is less than the minimum double patterning spacing (DP- spacing).

5. Introduction Minimum pitch size = minimum width (Wmin) +minimum spacing (Smin) DP-Spacing = 2Pmin - Wmin

6. Introduction Not all the patterns can be directly decomposed into two masks.

7. Introduction Nevertheless, the stitch insertion might not be powerful enough doe resolving all the conflicts. A conflict that cannot be resolved by inserting stitches anywhere is called a native conflict (NC). In this paper, it deals with the NC issue by two stages: – (1) NC prediction. – (2) NC removal

8. Preliminaries NC conflict

9. Preliminaries DPT-Compliant Redesign – A native conflict can only be resolved by DPT- compliant redesign. – DPT-compliant redesign should first predict the occurrence and the locations of NCs, and then further correct them.

10. Problem Formulation Given a post-routing layout, the minimum spacing, and the double patterning spacing, the problem finds a perturbed layout so that the number of native conflicts is minimized, subject to minimum-spacing rule and the double-patterning constraint.

11. Native Conflict Prediction An odd cycle in the conflict graph is not necessarily a NC because of the possible solution by stitch insertion.

12. Native Conflict Prediction Pattern projection

13. Native Conflict Prediction Segmentation

14. Native Conflict Prediction Odd-Cycle Detection – The existence of odd cycles can be determined by a DFS algorithm. – Finding all the odd cycles is time-consuming. Cycle basis – A cycle basis is a basis if we treat every cycle as a vector represented by series of edges, and the linear combination of cycles is a ring sum operation of cycles. Then the basis form a linearly independent spanning set for all cycles.

15. Native Conflict Prediction Ring sum operation – Let the edges in the cycles C 1 and C 2 be sets E 1 and E 2, respectively. A ring sum operation on C 1 and C 2 is denoted as C 1 ⊕ C 2, which is a cycle with its edge being a set, (E 1 ’∩E 2 ) ∪ (E 1 ∩E 2 ’).

16. Native Conflict Prediction Cycle basis of a graph can be found by applying a modified DFS algorithm. For a connected graph G = (V,E), the result of DFS traversal forms a spanning tree T on V. Inserting any other edge e ϵ (E - T) causes a cycle. If there is no odd cycle in the cycle basis of a graph, then the graph contains no odd cycle.

17. NC-Aware Wire Perturbation

18. Symbolic Layout Representation – Every wire (via) is represented by two points. – By the symbolic representation, it can perform the compaction across layers simultaneously.

19. NC-Aware Wire Perturbation DPT-Compliance Checking Wire Perturbation – Add DPT constraints to remove odd cycles. – Perform a trial compaction to see whether the level of DPT-compliance is improved.

20. NC-Aware Wire Perturbation Compaction – The compaction problem is modeled as a longest- path problem on the constraint graph. – (1) Wire Constraint Graph Construction An edge (u,v) → wire u is on the left or bottom of v. The transitive edges in the graph can be removed.

21. NC-Aware Wire Perturbation Compaction – (2) Point Constraint Graph Construction An edge (u,v) with its weight w → point v should be on the right of point u, and the distance between them should be at least w. There are four constraints: – (1) Design-rule constraint – (2) Wire shape constraint – (3) Fixed-pin constraint – (4) Layout-boundary constraint

22. NC-Aware Wire Perturbation

23. DPT Constraint Generation – (1) Separating Wire Pairs Generation

24. Experimental Result

26. Conclusion It presents a NC-prediction method based on geometry relation of features and a NC-aware wire perturbation algorithm to minimize as many NC’s in layout as possible. Perform an efficient method to find odd cycles.