A Methodology to Optimize Design Pattern Context ...
Recommend Documents
May 4, 2000 ... control module for each physical component of the system. .... a feedback control
system where the components are interconnected via a ...
DSP Architectures: TMS320C80 Results towards the Real-time Execution of Low. Level Image ... low-level image processing on the C80 and by extend on.
The effects of electrocoagulation on COD removal of Nam Phong swine slaughterhouse wastewater. (Binh Thanh District, Ho Chi Minh City, Vietnam) were ...
transportation of persons with disabilities in Flanders in the prediction years 2015, 2020 ...... Ben-Akiva, M., J. Benjamin, G.J. Lauprete, and A. Polydoropoulou.
Abstract Shot peening is a mechanical surface treatment widely used in the automotive and aerospace industries to improve the fatigue performance of metallic ...
Motion Detection Algorithm on a multi-processor DSP ... complex low-level image algorithms under real-time ... execution of processing operators (i.e nodes).
path for drilling of an industrial robotic manufacturing machine. ... to outperform solutions provided by application software ... reduce production costs.
designing context aware adaptive m-learning applications/ ... The development and deployment of context aware ... Developers can intelligently use context.
Siemens AG Corporate Technology ... Software systems with a layered architecture, such as mid- ..... a visitor proceeds through the park, her ticket is modified.
Oct 13, 2014 - tuition and an setting the s to solve it [17 ine and priorit .... Laboratory/Caltech, MIT Lincoln Laboratory) or other organizations in the given ...
Design/methodology/approach â A fractional factorial approach has been ... approach using fractional factorial experiment design to identify the influence of.
modelo de evaluación empleado por COTECMAR y la Armada Nacional de ... Este modelo, se compone de atributos disimiles que pueden ser parametrizables,.
Simplify Software Metric Models with Missing Data. W. Eric Wong & Jin Zhao ... dictor metric(s)), we propose a statistical hypothesis test- ing methodology to ...
Jul 15, 2011 - Abstract: In this paper, ultrasound-assisted extraction (UAE) was applied to the extraction of anthraquinones (aloe-emodin, rhein, emodin, ...
Fractional factorial design has been applied for the simultaneous study of the effects of ... Index Termsâ Electrocoagulation, Factorial Design, Model, ANOVA, ...
protein content (Y5). A central composite design was used and second-order models for Y1 were employed to generate response surfaces. When bread protein ...
Nov 16, 2016 - routine business processes by implementing 6 Sigma methodologies. The article .... is necessary to create a detailed process map for uniquely.
The percentages of soy flour (X1) and dry milk (X2) were varied for the production of gluten-free bread to optimize batter softness (Y1), specific volume (Y2), ...
313. Thermal Breakthrough Calculations to Optimize Design of a Multiple-Stage Enhanced Geothermal System. Tianyu Li, Sogo Shiozawa, and Mark McClure.
Apr 3, 2018 - principles has proved to be feasible on the elaboration of analytical ... Quality by design (QbD) concepts were initially applied to the de-.
building configuration and adaption according to the occupants' behaviour, this paper presents a ..... [29] Virtuoso Web page: http://virtuoso.openlinksw.com/.
Miguel de Cervantes 120. Complejo Industrial Chihuahua. 31109 Chihuahua, Chih. ..... the Design of a Solar Water Distillation System. Martin-Dominguez, I.
Keywords: Districting, hybrid algorithm, metaheuristic, graph, Tabu search, GRASP. RESUMEN ... procedure employed by the company to design districts is the ...
A Methodology to Optimize Design Pattern Context ...
A Methodology to Optimize Design Pattern. Context Size Using Pattern Association Tree (PAT). Piyush Pathak, Shikha Somani, Piyush Verma, and Sriram ...
A Methodology to Optimize Design Pattern Context Size Using Pattern Association Tree (PAT) Piyush Pathak, Shikha Somani, Piyush Verma, and Sriram Madhavan
Introduction : Configurations vs Patterns • Configurations are simple
geometric constructs like tip-to-line space • Patterns are configurations with
context • Configuration based DRCs are
insufficient to detect all design hotspots • Context plays a key role in the
design process interaction • Pattern based checks successful
alternative to complex rule based checks Dai. V., Proc. SPIE 6521, 65210A (March 28, 2007)
Conventional Pattern Deck Generation x
x
x x
x
x x
x x
x
x
x
VLSI design
Defect locations marked on VLSI design
Pattern classification with radius ‘r’
• All design hotspots classified into unique pattern
classes • All patterns in the deck have a fixed common radius
thus the same context size
Fixed radius pattern deck
Pattern context – Process correlation Etch
Photolithography
Pattern radius related to
Optical diameter
Chemical Mechanical Planarization
Microloading distance
Topography dependent radius of influence
VLSI design Variation in the density and geometry within a pattern
Design hotspots with varying context size depending upon the defect inducing process
Variable radius pattern deck Traditional fixed radius pattern deck
Variable radius pattern deck (generated using PAT)
•
A variable radius pattern deck consisting of ‘root cause’ patterns will have a much higher defect predictability than a fixed radius deck
•
‘Root cause’ patterns have optimal context for defect detection
•
A Pattern Association Tree (PAT) is used to build such a variable radius pattern deck
Pattern Association Tree (PAT) Parent
Child
Grandchildren
Features of PAT • A child node can have one and only one parent node • A parent node can have more than one children
• Number of children for a given parent node represents extent of
variation in context observed across different hotspots
PAT node annotation - Terminology Name
Description
Formula
Hit count (NC)
Hotspot markers covered
-
Match count (NM)
Total locations matched
-
Hotspot count (NH)
Total number of hotspots
-
Precision (P)
Fraction of matched locations that were true hotspots
NC/NM
Sensitivity (S)
Fraction of the total hotspots that were matched
NC/NH
Precision = 1/2, Sensitivity = 1/3
Step 1 – Generating nodes Pattern classification with radius r1 x
PAT construction set aka “Training set”
x
Generated pattern decks
FR–r1 Deck
x
x x x
x
x x
x
FR-r2 Deck
x
x
x VLSI design
Defect locations marked on VLSI design
x x
x
x x Pattern classification with radius r2
Step 2 – Building associations Pattern classification output x
FR-r1
x x
x x
x
FR-r2 x
x x
x x
x
Pattern Association Tree
Step 3 - Node annotation Generated pattern decks
Pattern matching with FR-r1
FR-r1
x Defect locations marked on VLSI design
Nm
P
S
4
9
4/9
4/6
2
6
2/6
2/6
Nc
Nm
P
S
1
3
1/3
1/6
3
4
3/4
3/6
1
2
1/2
1/6
1
2
1/2
1/6
x x
x x
x
x
Nc
x
FR-r2
x
x x
x
NH = 6
x
x x
x x
x Pattern matching with FR-r2
Filtering and ranking criteria PAT deck can be filtered based on the attribute of significance to the application NmP* : Detect those design constructs which are rarely used in the design but which have high defect correlation Nm > Nm* : Most frequently used P > P* : high defect predictability
Number of patterns in PAT deck equals the number of unique traversal paths in PAT
Nc
Nm
P
S
2
3
2/3
2/6
1
2
1/2
1/6
1
2
1/2
1/6
1
2
1/2
1/6
Patterns in the PAT deck can be further ranked: P : Patterns with higher correlation with defects Nc : Patterns which have covered more hotspots Nm: Patterns which are more frequently used S: Patterns which have covered more hotspots given the design space of training pool
Step 4: Optimal context pattern selection GEN-I
GEN-II GEN-III
Precision
GEN-IV
Sharp transition
Pattern radius
A significant transition in precision along a tree traversal path indicates the ‘root cause’ pattern Only ‘root cause’ patterns are selected to construct the PAT deck.
Results: PAT snippet
• Three randomly chosen tree traversal paths • Selection based on significant transitions in precision results in variable
radius pattern deck
Results: PAT deck composition • Pattern statistics from
training set – ~10000 markers used Nm/Nc (Unique pattern classes at the classification radius)
for classification (Nc) – Precision increases with increase in pattern classification radii (1/P = Nm/Nc)
• PAT deck – Is obtained after
filtering the PAT – Consists of patterns
ranging from 3r radius to 6r radius.
Results: PAT deck performance • Number of patterns in the
deck increase as the pattern radius increases • Precision increases as the
pattern radius increases • Sensitivity decreases as the
radius increases • PAT deck has half as many
patterns as the FR-6r deck and almost the same number of patterns as FR-5r deck
• PAT deck has much higher
precision than all the traditional fixed radius decks
• PAT deck has a reasonable
sensitivity given the small training pool
Conclusions Context sensitivity analysis to obtain ‘root cause’ patterns using
PAT PAT Deck has higher predictive capability than other traditional FR
decks The ranking and filtering criterion for PAT nodes can be fine tuned