improving quality and factory performance with real data · improving quality and factory...
TRANSCRIPT
2015. 10. 27
Daejung (Joseph) Ahn
Samsung SDS
2015 Oracle OpenWorld
Improving Quality and
Factory Performance
with Real data
Agenda
I. Issues of Manufacturing field
II. Directions for Future Factory
III. Review useful Practices
IV.Opinion for Success
1.1 Factory says a lot through data I. Issues of Manufacturing field
Processing Steps & TAT Machine Data Utilization Status
Day Hour Minute Sec msec
Lot
Wafer
Shot
Chip
Point
[Unit]
[Cycle]
(# of Sensor/Parameter)
2,000
(400)
Total TAT
# of Total Step
(Main Step)
35nm Process
(~2013)
20nm Process
(2014~)
4,000
(600)
※ TAT: Turnaround Time
65.8%
34.2%
※ Semiconductor, Display company case
0
1
2
3
4Data gathering and store
Data Search
Virtual item data
Data summary
Equipment sensor
monitoring
Operating condition
management
Analysis of the cause of
anomaly
Analysis of equipment
maintenance
Facilities abnormal pre-
analysis
Equipment code
standardization
Sensor code
standardization
Sensor Limit line selection
1.2 What is required capabilities?
Data-driven manufacturing capability
Subjects Top-tier General GAP
Machine data collection 3.8 1.9 -1.9
Data gathering and store 3.5 2.1 -1.4
Data Search 4.0 1.8 -2.2
Virtual item data 3.5 2.0 -1.0
Data summary 4.0 1.8 -2.2
Anomaly detection 4.0 2.4 -1.6
Machine sensor monitoring 4.0 1.8 -2.2
Automatic yield/defect monitoring 4.0 3.0 -1.0
Root cause analysis of anomaly 3.3 2.5 -0.8
Analysis of the cause of anomaly 3.0 3.0 0.0
Analysis of Machine maintenance 4.0 2.0 -2.0
Facilities abnormal pre-analysis 3.0 2.5 -0.5
Machine master data 3.3 3.5 0.2
Machine code standardization 3.0 4.0 1.0
Sensor code standardization 3.0 3.0 0.0
Sensor Limit line selection 4.0 3.5 -0.5
Top-tier company: 3.6
General company : 2.6
For general company, Equipment data summary and analysis capabilities are needed In order to
control their production process in stable condition
Top-Class
General
What are you making an effort for improving quality and performance of manufacturing?
Gaps between Top to General
※ Some items are omitted
※ Energy device company case
I. Issues of Manufacturing field
2.1 Directions for Future Factory
‘2014~‘1990~2013
“intelligent manufacturing” is the keyword for next generation factory to improve performance
and quality through Mobility, Analytics, IoT
Smart machine care (Real-time Analytics)
Real-time detection & corrective action
Big data analytics for unknown yield loss
Unmanned surveillance
for safe factory
“intelligent Manufacturing”
Mobility
Analytics
IoT…
• Anomaly Detection& Failure Analysis
• Feed-Back/Forward Control
• Machine On-line
• Semi automated Operation
• Auto Scheduling & Dispatching
roductionautomation
aterialandling
automation
ualitynalysis
automation
OperatorDependency
Full-automatedProduction
Yield & Effectiveness
※Semiconductor Display PCB, Battery, Mobile Assembly, etc
II. Directions for Future Factory
2.2 What is your Question?
Key questions for innovation in the manufacturing factory are as follows:
What kinds of valuable questions do you have for innovation?
① Can you detect abnormal condition of machine in real time using all machine data, e.g.,
sensor, transaction, event data, and so on?
② How to maintain optimal condition of machine considering its health condition, not only a
number of usage?
Product
Quality
Equipment
Performance
① Is there an alternative to identify root causes of chronic and unknown defects, not able to
solve with conventional statistical approaches?
② What is the way of dealing with manufacturing big data, such as all factory data during a
long time, e.g., for one year?
Our questions for innovation
II. Directions for Future Factory
3.1.1 Intelligent Machine Care -Furnace, Coating Machine
Data driven machine control using a single EQP health index
→ Optimize machine condition with health diagnosis and maintenance
Data Collection > Monitoring > Anomaly Detection > Analysis > PM Recommend
• Data Collection and Real-time Monitoring: Machine anomaly is detected through real-time sensor data collection, data pre-processing
• Machine Health Diagnosis: The condition of machine is diagnosed by a health index. Once machine anomaly is detected, root cause investigation takes place immediately
• Preventive Maintenance Prediction: The aging level of each module is quantified to predict the best timing for preventive maintenance
• A Single Intelligent Device, Anytime, Anywhere : Data collection, anomaly detection and machine diagnosis are all available from a single device. You can also check the condition of machines anytime, anywhere from your mobile device
How Resolved
② Key Parameter
Selection
① Data
Collection ③ Monitoring
④ Detection ⑤ Factor Analysis
Dimension Reduction
⑥ PM Recommendation
III. Review Useful Practices
for Equipment Performance
※MLCC: Multi Layer Ceramic Condenser, TBM: Time based Maintenance
(All-in-one device)
3.1.2 Sensor Profile Classification-Etching Machine
Cluster has been classified by the sensor patterns and the cluster with
high failure rate products can be identified
Single Products
Multi Products
III. Review Useful Practices
for Equipment Performance
Sensor Profile Clustering Quality Classification by Cluster
Failure
Clusters
Failure
Cluster
3.1.3 Pump Breakdown Prediction–Diffusion Pump
Problem Features Breakdown Prediction Results
• Irregular lifespan of pump (1 ~ N months)
• Working under abnormal condition
(Limitations of cognitive against pipe blockage)
• Two-step detection procedure : Detect abnormal conditions →
Detect breakdown signs → Predict breakdown point 13 days
before failure with prediction accuracy 80%
Replacement BreakdownAbnormal
Exhaust Temp.
Continue
Variable abnormality
Electric current of Dry
Pump
Electric current of
Booster Pump
Exhaust
Temp
Body
TempN2 Flow
Exhaust
Press
ProfileVacc.
Pump
1st Step detection
2nd Step detection
1st step raw data
Control Chart
Integrated indicator
Contribution
Start 2nd step only afterMeeting 1st step detection criteria
Increase breakdown preventions with productivity & quality risk indicators
→ Failure feature extraction and abnormal condition detection
III. Review Useful Practices
for Equipment Performance
3.2.1 Failure Analysis using Sensor profile–FAB
Problem Features
Able to increase possibilities of identifying unknown factors to explain chronic failures
Explain chronic failure root causes with high accuracy
→ Micro sensor pattern differences against time series of sensor data
Pattern comparing with Golden Signal
Difference in temperature profile between pass and failed products
Reference Signal
Bad #2
Bad #1
Identified profile pattern difference
Bad #2
Bad #1Difference in torque profile
between pass and failed products
Display Low Yield Problem
Semiconductor Low Yield Problem
Root cause analysis
with sensor profile?
Machine
Data
Failure root-cause
Process
Data
Product
Data
Above 3.5 billion data records
with half million sensors
III. Review Useful Practices
for Product Quality
3.2.2 Defect Pattern Analysis–Flat Panel Device FAB
Defect Segmentation Unmanned Analytics for Defect Reduction
Result I Result II
Full-automated Analytics (Accounting for 80% )
Fault DetectionFault Type
Classification
Cluster 2
Cluster 1
Engineer knowledge based fault
classification
Clustering based fault
classification
Root Cause
Analysis
Cluster 3
EQP
Analysis
Defect
Analysis
Inspection
Analysis
Metrology
Analysis
.
.
.
Systematic Analysis
Engineer dependent
analysis
SUMMARY
Automated Analysis Reports
CLUSTER
Materials
Glass
Coil
Etc.
Increase fault defection analysis coverage from under 50% to more 90%
→ Automatic root cause analysis with defect pattern segmentation
※ADC: Automatic Defect Classification, RCA: Root Cause Analysis,
III. Review Useful Practices
for Product Quality
3.2.3 Micro Quality Grading–Battery Factory
Fair quality grading Predict Fault Signs
Screening the main parameter affecting the final quality
Microstructure grading of fair quality goods(Hotelling’s T2 Statistic)
5th grade
…
2nd grade
1st grade
Predict defect-induced goodsthrough finding the time
when 5th grade goods occurs frequently
Increase
5th grade
products
Appear
defective
products
Derive a paradigm shift of fault analysis by micro-specific quality grades
→ Predict defective product trends not only monitoring but optimization
III. Review Useful Practices
for Product Quality
ValueEnjoy an economical, efficient, compatible and scalable option by using analysis functions with large-scale and high-speed parallel and distributed computing technologies based on in-memory.
3.3 Big data Analytics Architecturefor Semiconductor, FPD, Energy, Mobile Assembly, Steel industry
Accelerate big data performance with high-speed parallel and distributed computing
technologies
• Hadoop enables deployment of
big data analytics system at
a remarkably low cost.
• Wide range of statistical functions based on
in-memory parallel computation.
• Workflow type modeling tool is utilized to
make it easier to build analytical model and
share the results.
• User can work interactively with various types of
charts in the
graphical user interface.
Analytic Model Library
Product Map Analysis
Root Cause Analysis
Sensor DataProfile Classification
Sensor Data Profile Pattern Analysis
Product Grade Controller
Automatic Quality Anomaly Detection
Best PM Time Prediction
Machine Aging Modeling
Oracle EXA Oracle BDA X5-2 Cloud DeviceStore
HDFS Spark PostgreSQL HBaseHive
Data Handler Data Analyzer Visualization
High Performance
Data Processing
Functions (33EA)
Script ModelerWorkflow Modeler
Parallel Computing
Statistical Functions
(77EA)
R AdapterData Extractor
III. Review Useful Practices
4. Make a valuable Question…
If you clearly understand your goals and objectives, you can come
up with a substantial problem.
In order to solve the problem, outstanding question is
required. What are your questions?
IV. Opinion for Success