0

Applying AVM and Advanced Manufacturing Cloud of Things

for Industry 4.1

e-Manufacturing Research Center National Cheng Kung University

Taiwan, ROC

Foresight Technology Company, Ltd. Taiwan, ROC

March 3, 2016

鄭芳田 Fan-Tien Cheng

1

VM Definition

Key Features of the AVM System

AVM Demo

Integrating GAVM into WMA

Requirements of CPA

Advanced Manufacturing Cloud of Things (AMCoT)

Applying AMCoT to WMA

Industry 4.0 & Industry 4.1

Selected References

Contents

2

VM Definition

3

Virtual Metrology (VM) is a method to conjecture manufacturing quality of a process tool based on data sensed from the process tool and without physical metrology operation.

Virtual Metrology

it 1+it 2+it 3+it 4+it jt 1+jt 2+jt 3+jt 4+jt

UCL

LCL

Process TimeVir

tual

Met

rolo

gy D

ata

iy()

^OOC

PHASE 5

DEPT 6

DEPT 5

DEPT 4

DEPT 3

DEPT 2

DEPT 1

PHASE 4PHASE 3PHASE 2PHASE 1

DEPLOYMENT CHART

Metrology Equipment

Virtual Metrology System

Sampling Products

Products

Sensor Data ijx

Real -Time & On-Line

T∆Transportation &

Measurement Time

T∆

Tti ∆+ Ttj ∆+it jt

UCL

LCL

Process Time

T∆

Rea

l Met

rolo

gy D

ata

iy()it

jt

Production Equipment

IBM

VM can convert sampling inspections with metrology delay into real-time and on-line total inspection.

4

Key Features of the AVM System

5

Conjecture Model

RI Module

GSI

RI

ProcessData

GSI Module

VMI

VMII

Only forTraining& Tuning

√√√√√√√√√√√

MetrologyData

√

Dual-PhaseAlgorithm

Data Preprocessing

DQI Z-Scorey

DQIx Z-Score

Data Preprocessing

AVM Server

Data Quality Evaluation

Prediction Kernel

Reliability Indexes

6

Advanced Dual-Phase VM Algorithm Metrology Data

Collection

Collection Completed

Metrology Data vs. Process Data

Correlation Check

Correlation Successful

Re-TrainingDQI, VM & RI/GSI

Models

TuningDQI, VM & RI/GSI

Models

No

No

Yes

Update VM, RI/GSI & DQI Models

Phase II

1st Cassette after Idling

YesNo

Manual Activation or Refreshing

Yes

No

Re-compute VMII with RI/GSI of Each Workpiece

in the Entire Cassette

Start Start

Update Models

DQIy CheckBad

Good

Yes

Compute VMI with RI/GSI

for the Workpiece

No

Yes

Phase I

Process Data Collection of a Certain

Workpiece

Collection Completed

DQIX CheckBad

Good Send Warning & Ask for Analysis

Disable RefreshingIf VM Accuracy Is Gained

Send Warning & Ask for Analysis

Correlative Process Data is Good

No

Yes

The DQIX and DQIy models are added into the algorithm to perform automatic data quality evaluation.

A model refreshing mechanism is added into the algorithm for automatic model-refreshing control.

When VM accuracy is gained, the refreshing procedure will stop automatically and the system will enter the normal operation state. Taiwan Patent No.: I349867;

US Patent No.: 8,095,484 B2; Korea Patent No.: 10-1098037; Japan Patent No.: 4914457; China Patent No.: 843932

7

Automatic Virtual Metrology (AVM) System

VMClient

SOAP

VMManager

MetrologyEquipment

SOAP

VM Server-1

MetrologyEquipment

...

...

Model Creation Server

CentralDatabase

ProcessEquipment

ProcessEquipment

VMClient

VM Server-n

8

(a) TFT Process Production Line

ProcessData

ProcessData

MetrologyCD

MetrologyDepthProcess

Data

MetrologyWidthProcess

DataProcess

Data

MESMaterial flowInformation flow

Film Deposition

(Stage I) Exposure Developing Etching Stripping

Positive Photoresist

Coating

Film Deposition (Stage II)

Metrology THK (G1)

(a)

Substrate (Glass)

IN

GG1G2

Substrate (Glass)Substrate (Glass)

G1Gate Layer

Photoresist

Substrate (Glass) Substrate (Glass)

UV

Mask

Substrate (Glass) Substrate (Glass)

MetrologyTHK (G) Process

DataProcess

DataMetrology

CDMetrology

DepthProcessData

MetrologyWidthProcess

DataProcess

Data

VMI or VMII

MES

AVMCD

AVMTHK (G1)

AVMTHK (G2)

~

Material flowInformation flow

(G1)^

VMI )) VMII&

Film Deposition

(Stage I) Exposure Developing Etching Stripping

Positive Photoresist

Coating

AVMDepth

AVMWidth

Film Deposition (Stage II)

VMI )) VMII&VMI )) VMII&VMI )) VMII&VMI )) VMII&

Metrology THK (G1)

(b)

(a)

Substrate (Glass)

IN

GG1G2

Substrate (Glass)Substrate (Glass)

G1Gate Layer

Photoresist

Substrate (Glass) Substrate (Glass)

UV Mask

Substrate (Glass)

+-

Substrate (Glass)

MetrologyTHK (G)

~G2 = G - G1^

(b) Deployment of AVM Servers

Semiconductor Layer of the TFT Process Flow with Deployment of AVM Servers

9

The purpose is to integrate the functions of AVM with those of MES. The interfaces among AVM, other MES components, and R2R (run-to-run) modules in the novel manufacturing system are defined so that the total quality inspection system can be realized and the R2R capability can be migrated from lot-to-lot control to wafer-to-wafer control.

Integrating AVM into MES

Equipment i Equipment i+1

ProcessData

Material flowInformation flow

Metrology

)25 pcs)

)1 pc)

MES

R2R i R2R i+1

Alarm Manager

WIP Tracking

SPC Scheduler

Equipment Manager

Material Manager Reporting

)25 pcs)

Equipment i Equipment i+1

ProcessData

Material flowInformation flow

Metrology

AVM

)25 pcs)

)1 pc)

MES

R2R i R2R i+1

Alarm Manager

WIP Tracking

SPC Scheduler

Equipment Manager

Material Manager Reporting

)25 pcs)

VMI for FB R2R Control

VMII for FF R2R Control

10

A key challenge preventing effective utilization of VM in R2R control is the inability to take the reliance level in the VM feedback loop of R2R control into consideration. The reason is that adopting an unreliable VM value may be worse than if no VM is utilized. The AVM system possesses the RI of VM to gauge the reliability of VM results [2], [A]. Therefore, this novelty is to invent a novel scheme of R2R control that utilizes AVM with RI/GSI in the feedback loop.

APC Utilizing AVM with RI/GSI

Metrology

Tgt

Material flowInformation flow

Process

1ku +

1α

Product

zy

: No. of RunskZ: No. of Actural Measurements

Sampled product

ηk

R2R

0 1 kuβ β+

EWMA Filter

Metrology

Tgt

Material flowInformation flow

Process

AVM ˆkykX

1ku +

1α

Product

ProcessData

For Training or Tuning

zy

: No. of RunskZ: No. of Actural Measurements

Sampled product

VMI or VMII

ηk

2α /RI GSI2 ( ,f RIα =

R2R

Upstream Metrology Data

0 1 kuβ β+

EWMA Filter

1)GSI α×

11

AVM Demo

12

Live Demo of the AVM System for CNC Precision Machining

Video showing the precision machining on cellphone backplanes.

(At the 2012 Taiwan International Machine Tool Exhibition)

GUI displaying real-time and online VM values of straightness 2.

(The CNC tool was located in a machine tool factory in Taiwan)

(The GUI was shown at the Exhibition Hall in Taipei, Taiwan)

The engineer proceeds to press the “Start Machining” button

The CNC machine is performing precision machining on the 25th cellphone backplane

The total duration of the precision machining is about 10 seconds

After the precision machining is completed, the GUI will display the predicted machining-precision value of the 25th cellphone backplane within 10 seconds

VM value of the 25th sample is displayed within 10 sec after

processing.

13

420408396384372360

348

Curr

ent

1

336

(b)

1

Dual Phase

Free Run

(a)

GSI

79

79PM3PM1 PM2

GSI = 9.55

50

GSI = 36.44GSI = 11.11

GSI

ESC Leakage Current

ESC Leakage CurrentLCL=130, UCL = 330

LCL=130, UCL = 330

VM (M

ean)

Curr

ent

Curr

ent

I I

RI

RIT

GSIT

RIT

GSIT

RI

420408396384372360

348336

VM (M

ean) The VM conjecture models will be tuned/re-

trained by the dual-phase scheme run-to-run.

The VM conjecture models will not be refreshed.

Smooth manufacturing process

Superior VM accuracy (Sampling rate can be reduced)

Demo of the RI, GSI, & Dual-Phase Schemes (1/2)

14

420408396384372360

348

Curre

nt

1

336

(b)

1

Dual Phase

Free Run

(a)

GSI

79

79PM3PM1 PM2

GSI = 9.55

50

GSI = 36.44GSI = 11.11

GSI

ESC Leakage Current

ESC Leakage CurrentLCL=130, UCL = 330

LCL=130, UCL = 330

VM (M

ean)

Cur

rent

Cur

rent

I I

RI RIT

GSIT

RIT

GSIT

RI

420408396384372360

348336

VM (M

ean)

GSI values are higher than GSIT

VM accuracy can be recovered and maintained

PM3PM1

RI values are less than RIT

VM accuracy becomes poor.

The VM conjecture models will be tuned/re-trained by the dual-phase scheme run-to-run.

The VM conjecture models will not be refreshed.

Demo of the RI, GSI, & Dual-Phase Schemes (2/2)

15

Sampling Rates 2/25 vs. 1/50

Sampling Rate 2/25

Sampling Rate 1/50

16

Integrating GAVM into WMA

17

The Wheel Machining Automation (WMA) cell layout

• two lathes (Lathe 1 and

Lathe 2) • one drilling machine

(Drill) • one off-machine

metrology (OMM) tool for sampling inspection of all precision items

• one robot is placed in the middle of the cell for material handling and two buffers are installed for input/output purposes.

Integrating GAVM into Wheel Machining Automation for Total Inspection (1/4)

Metrology Tool

OutIn

Lathe 1

Drill

Lathe 2

AVM Server

GED

Sampling Inspection (e.g. 1 out of 20) Total Inspection

ReaderLaser

Marker

18

Conjecture Model

RI Module

GSI

RI

GSI Module

VMI

VMII

Only forTraining& Tuning

AdvancedDual-Phase

VM Algorithm

Data Preprocessing

DQI Z-Scorey

DQIx Z-Score

Data Preprocessing

AVM ServerGED

Comm

unicationAgent

ApplicationInterface

PluggableApplication

Module

(PAM)

EquipmentDriver

DataCollectionManager

DCPlan

DCReport

InterfaceB

ox STDB

Process Data

Metrology Data

Machine T

oolM

etrology Tool

Sensor Data

Machining Parameter

Metrology D

ata

GAVM System for Machine Tools.

Integrating GAVM into Wheel Machining Automation for Total Inspection (2/4)

19

• Traditional Wheel Machining Inspection: wheel quality can only be checked by one in-line metrology (ILM) equipment for total inspection of primary precision items, and one off-machine metrology (OMM) tool for sampling inspection of secondary precision items.

• GED-plus-AVM (GAVM) System: The integration system of generic embedded device (GED) and AVM play

an inspector role of achieving nearly ZD of automated total inspection for the machine-tool industry.

Sufficient process data for the GAVM system are not only collected from sensors, but also can be: segmented to extract the essential parts from the original numerical-

control (NC) file for synchronizing various machining quality/metrology data

filtered for improving signal-to-noise (S/N) ratios transformed into essential features for building VM models, and the

corresponding solutions of those three challenges were implemented in GED

Integrating GAVM into Wheel Machining Automation for Total Inspection (3/4)

20

Integrate GAVM into WMA: a Miniature of Industry 4.0 • GED is the prototype of an IoT agent • AVM is an example of CPS To Do: Enhance the functions of GED, and then adopt Advanced Manufacturing Cloud of Things (AMCoT) for wheel production industries based on the miniature GAVM system.

Integrating GAVM into Wheel Machining Automation for Total Inspection (4/4)

21

Requirements of CPA

22

GED: focus on data collection and implementation of pluggable applications

• To reach the goal of Industry 4.0, the functions of 3C, including computation, communication, and control, should be implemented into GED so as to become a CPA for achieving “man-machine collaboration”

Differences between CPA and GED: 1) control of interaction with physical objects and cyber systems 2) the communication among the cyber systems (e.g. AMCoT), the physical

objects (e.g. machine tools), and human operators

Requirements of CPA (1/2)

23

Conjecture Model

RI Module

GSI

RI

GSI Module

VMI

VMII

Only forTraining& Tuning

AdvancedDual-Phase

VM Algorithm

Data Preprocessing

DQI Z-Scorey

DQIx Z-Score

Data Preprocessing

AVM ServerGED

Comm

unicationAgent

ApplicationInterface

PluggableApplication

Module

(PAM)

EquipmentDriver

DataCollectionManager

DCPlan

DCReport

InterfaceB

ox STDB

Process Data

Metrology Data

Machine T

oolM

etrology Tool

Sensor Data

Machining Parameter

Metrology D

ata

GAVM system for machine tools.

Requirements of CPA (2/2)

Turns GED to CPA

24

Cyber-Physical Agent (CPA)

CPA consists of 1) CPA Control Kernel 2) Communication Service 3) Data Collection Manager 4) Data-Collection Plan (DCPlan) 5) Data-Collection Report

(DCReport) 6) Equipment Driver 7) Application Interface 8) Database

CPA Architecture

CPA Control Kernel

Data Collection Manager

Equipment DriverApplication Interface

Pluggable Application

Module

Pluggable Application

Module

...

Communication SerivceREST SOAP

DCR

DCPDatabaseController

PageMaker

Command Handler...

Database

GPIO Driver

ZigBee/WSNDriver

(USB/COM)

Ipv4 Driver

(Wi-Fi/Ethernet)

IPv6/WSNDriver

(6LoWPN)

Taiwan Patent : I225606 US Patent：7,162,394

Japan Patent：特許4303640號

25

Advanced Manufacturing Cloud of Things (AMCoT)

26

Advanced Manufacturing Cloud of Things AMCoT provides a cloud-based platform for connecting and sharing all

information of things for bridging the technology supports among vender, customers, and manufacturing tools.

AMCoT

Customer 2

Customer 1

Customer 3

Customer 4

Vender

One-to-Many Relationship among a Vender and its Customers via AMCoT

Application Example -- AVM Model-refreshing: Each customer does not need to build its own AVM models but just downloads all the preliminary AVM models (provided by the vender) from AMCoT.

27

Applying AMCoT to WMA

28

…

AMCoTSTDB/CDB

AVM System

Tool Life Management

Service

Collision Detection Service

Metrology Planning Service

IntelligentPredictive

Maintenance

CPA CPA1Tool RUL

Vender Customer 1CPA2Cell 1 Cell 2

Customer 2

Customer 3

Customer 4

Applying AMCoT to WMA

Integrating WMA’s Vender and Customers into AMCoT

29

UCL

LCL

Vender

Sample No.

(mm

)

SC2

Cell 1

SC3SC1

DQIy Error

The first wheel in Cell 1

GSI

Mai

n M

otor

C

urre

nt: R

MS

(A)

30

20

010

GSI

Global Cyber-Physical Interactions (AVM Models Refreshing). LCL: lower control limit (72.93 mm); UCL: upper control limit (73.15 mm).

Illustrative Example−Experiment

30

Industry 4.0 & Industry 4.1

31

Industry 4.0

Industry 4.0 = IoT + CPS + Cloud Manufacturing (CM) + Big Data Analytics.

Industry 4.0 only keeps the faith of achieving nearly Zero Defects (ZD) state.

Zero Defects (ZD) • ZD has been one of the quality-improvement objectives for

accomplishing manufacturing quality [A]. • Through prevention methods, ZD aims to boost production and

minimize waste. • ZD is based on the concept that the amount of mistakes a worker

makes doesn't matter since inspectors will catch them before they reach the customer [A].

[A] Halpin, James F. Zero Defects: A New Dimension in Quality Assurance, New York: McGraw-Hill, 1966.

32

Industry 4.1

Industry 4.0 + AVM = Industry 4.1

→ To achieve the goal of Zero Defects (ZD) • Stage I：Achieving the goal of ZD for all the Deliverables • Stage II：Achieving the goal of nearly ZD for all the Products by Continuous Improvements

33

Process Flow of TFT-LCD Manufacturing

The TFT-LCD manufacturing flow consists of four processes: TFT, CF (color filter), LCD, and LCM (liquid crystal module).

TFT substrate

CF substrate

34

TFT-LCD Front-End Process TFT Process

CF Process LCD Process

Chromaticity

THKTHKTHKTHK

Chromaticity

PS LayerG LayerR LayerBM Layer ITO LayerB Layer OC Layer FinalInspection

Mura

Totalpitch

Chromaticity

ProcessData

ProcessData

ProcessData

ProcessData

ProcessData

ProcessData

ProcessData

BM line width

OD

Mura

BMTHK

Totalpitch

MarkPOS

THK

Chromaticity

MuraMarkPOS

Pixel Dimension

THK

Substrate (Glass)

GSubstrate (Glass)

BM BM BM BM

Substrate (Glass)

R Substrate (Glass)

BSubstrate (Glass)

PS

Substrate (Glass)

ITO

Unifornity SheetResistance

PS Height

PS CD

SheetResistance

TEGParticle

PatternInspection

Pin Hole

Particle

THK

Width

SheetResistance

Pin Hole

Particle

THK

Width

SheetResistance

Pin Hole

Particle

THK

Width

SheetResistance

Gate Metal (Stage I)

AS (Stage II)

Particle

GETHK GE

Width

Sheet Resistance Pin Hole

Particle

THK

Width

SheetResistance

SD (Stage III)

BP (Stage IV)

ITO (Stage V)

FinalInspection

ProcessData

ProcessData

ProcessData

ProcessData

ProcessData

Substrate (Glass)

Gate Metal Source/Drain Metal

Substrate (Glass) Substrate (Glass)

SiNx

Substrate (Glass)

ITOSiN A-Si N+ A-Si

Substrate (Glass)

THK

PretilyAngle

THKProcessData

ProcessData

Scribe Location

ProcessData

Polarizer Attach

ProcessData

Assemble

Polarizer Location

Scribe

Polyimide Print ODF

Substrate (CF)

Substrate (TFT)

Seal Dispense

ProcessData

Substrate (TFT) Substrate (TFT)

Substrate (CF)

Substrate (TFT)

FinalInspection

Cell Gap

Luminance

ContrastRate

RubbingSystem

ProcessData

THKProcessData

Polyimide Print

Seal Dispense

ProcessData

RubbingSystem

ProcessData

ProcessData

Substrate (TFT)

Substrate (CF) Substrate (CF)

HOT PressSubstrate (TFT)

Substrate (CF)

Substrate (CF)

Substrate (TFT)

PretilyAngle

THK

ProcessData

35

Selected References

36

Selected References Journal Papers F.-T. Cheng, C.-A. Kao, C.-F. Chen, and W.-H. Tsai, “Tutorial on Applying the VM Technology for TFT-LCD

Manufacturing,” IEEE Transactions on Semiconductor Manufacturing, vol. 28, no. 1, pp. 55-69, February 2015. F.-T. Cheng, C.-F. Chen, Y.-S. Hsieh, H.-H. Huang, and C.-C. Wu “Intelligent Sampling Decision Scheme Based on the

AVM System,” International Journal of Production Research, vol. 53, no. 7, pp. 2073–2088, 2015. H.-C. Yang, H. Tieng, and F.-T. Cheng, “Automatic Virtual Metrology for Wheel Machining Automation,” International

Journal of Production Research. DOI: 10.1080/00207543.2015.1109724, published online: Nov 2015. F.-T. Cheng, H. Tieng, H.-C. Yang, M.-H. Hung, Y.-C. Lin, C.-F. Wei, and Z.-Y. Shieh, “Industry 4.1 for Wheel Machining

Automation,” IEEE Robotics and Automation Letters, vol. 1, no.1, pp. 332-339, January 2016.

Patents Fan-Tien Cheng, Hsien-Cheng Huang, and Chi-An Kao, “Dual-Phase Virtual Metrology Method,” U.S. Patent no.:

7,603,328 B2, Taiwan R.O.C. Patent no.: I338916, Japan Patent no.: 4584295, China Patent no.: 823284, and Korea Patent no.: 10-0915339.

{行政院2011年傑出科技貢獻獎} {2011經濟部國家發明創作獎銀牌}

Fan-Tien Cheng, Hsien-Cheng Huang, Yi-Ting Huang, and Jia-Mau Jian, “System and Method for Automatic Virtual Metrology,” U.S. Patent no.: 8,095,484 B2, Taiwan R.O.C. Patent no.: I349867, Japan Patent no.: 4914457, China Patent no.: 843932, and Korea Patent no.: 10-1098037. {2012經濟部國家發明創作獎 發明獎金牌} {2013 IEEE Inaba Technical Award for Innovation Leading to Production}

Haw-Ching Yang, Hao Tieng, Min-Hsiung Hung, and Fan-Tien Cheng, “Method for Predicting Machine Quality of Machine Tool,” Taiwan R.O.C. Patent no.: I481978; with U.S. and China Patents Pending under applications 14/069,382 and 201310593639.0, respectively.

Fan-Tien Cheng, Chun-Fan Chen, Yao-Sheng Hsieh, Hsuan-Heng Huang, and Chu-Chieh Wu, “Metrology Sampling Method and Computer Program Product Thereof,” Taiwan R.O.C. Patent no.: I521360; U.S. and China Patents Pending under applications 14/666,324 and 201410196529.5, respectively.

37

Companies or Organizations that have technology transferred and/or deployed AVM related Patents or Technologies

Semiconductor Industry: TSMC (台積電), UMC (聯電), ASE (日月光)

TFT-LCD Industry: Innolux (群創), AUO (友達), CPT (華映)

Photovoltaic Industry: Motech (茂迪)

Machine Tool Industry: FEMCO (遠東機械) /FATEK (發得科技)

Aerospace Industry: AIDC (漢翔)

Organizations: ITRI (工研院) (Machine Tool Technology Center工具機科技中心 & Big Data Technology Center巨資中心) , MIRDC (金工中心)

38

Thank you for your listening !

Q&A

39

Foresight Technology

Tool & Info Automation • PC/PLC

Control • SECS/GEM

Capability

Data Collection & Storage • MES • Tool Process

Log & Parser

Data Visualization & Control • Tool Real-Time

Monitor • Data Viewer

Data Featuring • RTM FDC:

Retrieve Real-Time Data to Physical Index

• eRunCard: Retrieve Process/ Mfg Information

Data Analysis • YMS/EDA:

Engineering Data Analysis

• ePK: Key Parameter Analysis

Big Data Application • Control: Run-

to-Run Control • Prediction:

AVM & IPM • Tool Matching • Tool Health

機台與資訊 自動化

資料搜集 與儲存

資料視覺化與監控

資料特徵萃取

資料分析

Big Data應用

IoT Big Data

• Website: http://www.fs-technology.com/ • Phone: (06)236-6981 • Mobile: +886-977-158-326 • e-Mail: tltseng@fs-technology.com

Industry 4.1