Yen-Ting Yu
Iris Hui-Ru Jiang
Yumin Zhang
Charles Chiang
DRC-Based Hotspot Detection Considering Edge Tolerance and Incomplete Specification
ICCAD’14
Outline
Introduction Preliminaries Hotspot Detection Framework Experimental Result Conclusion
Introduction
In modern manufacturing processes, certain layout configurations are susceptible to lithographic process
Patterns with similar layouts could become process-hotspots
Represent these similar patterns by a representative pattern with edge tolerances and incomplete specified regions
String-matching-based Each pattern and layout window are encoded by
strings
Overview
The key features of this work Redefine MTCG and the extraction rules to
reflect the impacts of don’t care regions and edge tolerances
DRC searching space reduction technique Longest common subsequence on strings to
handle the impact of don’t care regions
Preliminaries
Design Rule Checking (DRC) Design rules are a set of parameters to ensure
the manufacturability of a layout Fundamental rules include the minimum width,
minimum spacing, and minimum enclosure rules
Modified Transitive Closure Graph (MTCG)
Problem Formulation
Given Hotspot pattern with edge tolerances and
incompletely specified regions (don’t care regions)
A layout Report
All hotspot locations considering eight possible orientations in the layout
Hotspot Detection Framework
Pattern Enumeration
Edge tolerances within a given pattern may lead to different pattern topologies
Extend the idea of All-Pair Min-Range Path (APMRP) algorithm to form pattern enumeration algorithm
APMRP m and n denote the minimum and maximum
distance between two edges minimize the (n – m) value If m < 0 and n > 0
(m, n) set contains three subsets: {(m, −1), (0, 0), (1, n)}
If m < 0 and n = 0 (m, n) set contains two subsets: {(m, −1), (0, 0)}
If m = 0 and n > 0 (m, n) set contains two subsets: {(0, 0), (1, n)}
Else only one subset {(m, n)}
MTCG with Don’t Care Regions and Critical DRC Rule Extraction
To use the aid of DRC to realize hotspot detection
Interpret all edge constraints to design rules Redefine five types of rules in [1] All rules can be extracted only from Ch,h and
Cv,v, Ch,v and Cv,h are serve for boundary checking
Rule 1(internal rule)–the width and height of a block tile find the dimension of each block tile that does
not touch the window boundary
Rule 2(external rule)–the distance between two adjacent block tiles find the dimensions of all space tiles that do not
touch the window boundary and are located in between block tiles
Rule 3(diagonal rule)–the diagonal relations between two convex corners of block (space) tiles find the diagonal relations between any two
convex corners of block (space) tiles
Rule 4(longedge rule)–the space or block tile with one edge touching the window boundary
Rule 5(segment rule)–the space tile with two or three adjacent edges touching the window boundary or space tiles
The dimensions of each extracted rule can be represented by a rule rectangle
The height and width of a rule rectangle are defined by its corresponding edge constraints
Define two types of don’t care regions Don’t region with two or three adjacent edges
fully facing the window boundaries
Don’t region in between two facing edges of polygons
Rule 6––the space tile with one edge or two opposite edges touching the boundary tiles
Searching Space Reduction
A pattern may have eight possible orientations Divide these eight orientations into two sets Generate a runset file for each set and run DRC
twice to obtain the locations that hit any generated rule
The region AND technique
Rule Ordering
Even a simple range pattern may generate tons of different pattern topologies after pattern enumeration
With the region AND technique, how to cover the whole pattern topologies during DRC with fewest DRC rules becomes an issue
The topology covering problem is NP-hard U = {1, 2, 3, 4, 5}
four subsets S = {{1, 2, 3}, {2, 4}, {3, 4}, {4, 5}}
subsets{1, 2, 3} {4, 5} A greedy heuristic can be applied to this
problem
Rules priority {internal rule, external rule, diagonal rule}
v
{longedge rule, sixth rule}
v
{segment rule}
Candidate Identification
Each generated pattern topology is represented by a set of DRC rules
Encoding rule rectangles to two strings, one in the vertical, one in the horizontal
To identify the potential hotspot locations in the layout, based on DRC results and rule priorities
Finalization
Some locations contain extra polygons that are not related to any of our extracted DRC rules and are not within the don’t care regions
Experimental Result
Implemented in the C programming language on a Linux platform
Hotspot patterns
Integrate a state-of-the-art industrial DRC engine into our framework
Conclusion
Proposed an accurate and efficient hotspot detection framework to handle hotspot patterns with edge tolerances and incompletely specified regions
Compared with the state-of-the-art work, our approach can reach promising success rate with significant speedups