Information Systems and Cross-Enterprise Learning
in Support of New Product Introduction
Enterprise Needs and Information Systems Based
Decision Management
Professor Ram Akella
Product and
Process
Development
Customer
Information Systems and Technology Management
University of California at Santa Cruz
Talk Outline
• A framework for solving ISTM problems
• Computer Based Process Learning and Information Systems
–Yield Management Systems
Sales and
Marketing
MIS Research Center
Carlson School of Management
University of Minnesota
February 20, 2004
• Incorporating Learning Into New Product Introduction through
competitive use of Information Systems
–An ontological framework (ongoing research)
• Enterprise Management and ISTM
IT
–B2B Exchanges: E-Business and Supply Chain Management
–IT Management, Outsourcing and Real Options
–Product Portfolios in High Tech and Food
Finance and Investment
E-Business and Supply
Chain Management
Outline of Talk: ISTM Framework
IS-based
Process Learning
Technology
Management
Information
Systems
Completed
Partially
completed
TBD
Partially
completed
Partially
completed
In progress
Complete,
in-progress
TBD
TBD
Integration
• Business Problem
– Business Need
– Technology Context
IT -based Product Process
Reengineering
An Ontology Framework
Enterprise management
B2B: E-Biz/SCM
IT Management, Outsourcing,
Product Portfolios, Real
Options
2
Computer-based Process Reengineering
Yield Management Systems
New product
introduction with
learning
1
• Technology Management
– Data to Information
– Information to Knowledge Management
• Information Systems
• Integration
3
5
1
IT-based Product Process Reengineering:
Business Need
Computer-based Process Reengineering
In-line Monitoring for Semiconductor ProductProcess Learning and Manufacturing
Isolation
• Need at KLA: Grow market and revenues
–Opportunity: Koreans claim that KLA in-line inspection
machines help dominate Japanese
–Need: Lack an internal understanding of
• Methodology: For rapid product and process learning
– How to use inspection machines to improve yield
per machine
• Economics: Of process and monitoring tools
– How many machines are worth buying to further
improve yield
• Information Systems: Not yet seen as a major need!
• Required to persuade customers, to sell more
machines
–Company size: $200M
• Business Problem
– Business Need
– Technology Context
Depo
Poly 1
Poly2
Metal 1
Metal 2
Off-line
Review
Etch
KLA KLA
Þ
Þ
2110
2110
or
In-line
ADC
Photo
• Technology Management
– Data to Information
– Information to Knowledge Management
Wafer
Probe
Wafer processing
Inspection
Classification
10 days
• Information Systems
• University Mission: Move area from “Black Art” to
30 days
Yield
• Integration
“Science”
6
7
Technology Implications for Speed and Accuracy of
Excursion Detection:
Review Methodologies and Images
Inspection: Defects on Wafers- Surrogates for
Yield in Monitoring
8
Excursion Detection of Killer Defects in the
Presence of Non-killers
Random defects
Equipment defect X
Process defect Y
Wafer Map
Image from
optical
Image from SEM
Defect trend chart at Poly in Fab A
1.4
Normalized defect D
1.2
Multiple defect types
• Killer defects – Kill die
• Non-killer defects – Do
not kill die
• Smaller pixel size ? finer
resolution, increased scan time
• Larger pixel size ? lower
resolution, faster scan time
9
0.6
0.4
0.2
0
Non-killers can “mask”
killer defects
Trade-offs
1
0.8
1
21
41
61
81
101
121
141
161
181
201
Wafer sequence
10
Excursions of killer defects(in red) can be masked on the SPC
chart of random defects by the non-killer defects (in green) 11
2
Sampling Strategy for Wafer Inspection and Defect
Review : Minimize Excursion Detection Time
Where?
% lots?
Wafers
per lot?
Computer-based Process Reengineering
Defects
Defect
to review? classification
Challenge
Poly
Poly 2
Þ
KLA
2110
Þ
KLA
2110
Via
Metal 2
UCL
?
<
Result
Generalizing Neyman-Pearson
results on minimizing risk of
not detecting an excursion,
with a constraint on probability
of false alarms
• Developed new model and
Model fab-wide yield
learning and optimize the
rate of learning and
investmentfor maximum
profitability
Integrating statistics used for
excursion detection with
queuing models for capturing
resource usage and the
resultant delays in integrated
inspection-review systems
• Developed and solved novel
Integrated defect type and
yield data not available,
difficulties
using heterogeneous
databases
Fab partnerships, BS/MS intern
training, 2 -3 years onsite data
extraction, data analysis and
Business Intelligence reports
on integrating disparate fab
databases
2
# of type
estimate
• Technology Management
– Data to Information
– Information to Knowledge Management
OoC
Process
In-control
Action Required
Model and detect a killer
defect type excursion when
it is masked by the
presence of other defect
types
1
• Business Problem
– Business Need
– Technology Context
Isolation
Process
Key Challenges in Framing and Solving Yield
Learning Problem
Excursion
Excursion
Occurs
Detected
Detection delay
Objective
To reduce the detection delay by trending individual defect types with
integrated wafer inspection and defect review strategy
• Methods : Decide how much inspection sampling and how much review sampling
• Economics: Decide how many machines of each type and associated personnel
analytics incorporating sampling
error
• Data to Information
•
• Information Systems
3
• Integration
12
IS-based Process Learning
13
Key Issues
• Validation of Solutions
• Information Systems
development and Integration to
achieve Business Intelligence
14
Conversion Of Defect Data To Yield Information
And Action
Firms unclear about “system” level functioning and performance
- Caught up in technology
- Data to information not clear to firms
- Information to knowledge is much worse
- Goal is not clear; consequently data and information systems
- Direct consequence – poor integration of information systems
- Concept of “meta model” needed rather than just “meta data”
• Business Problem
– Business Need
– Technology Context
models:
– Economically Optimized Yield
Learning
– Benefit of Ownership (in place
of traditional Cost of
Ownership)
Information to Knowledge
Management
Isolation
Poly1
Poly2
Metal1 Metal2
Market
• Technology Management
– Data to Information
– Information to Knowledge Management
Resources &
strategies
Validation
Inspection
Corrective
actions
Review &
classification
Root-cause
analysis
Source
identification
? DD(t) &
? Y(t)
Data/information flow
• Information Systems
• Integration
15
16
Goal
Detect killer defect excursion faster through
efficient integrated inspection-review cycles
Trade-off: Time versus benefit and cost
17
3
Defect Control Charts: Single-dimensional
Multi-dimensional Excursion Monitoring With
Defect Classification
Killer
Defect Distributions
Upper Control
Limit (UCL)
Control Chart
Probability
InControl
(INC)
f(x)
Out -ofControl
(OOC)
g(x)
In-control
x
Y
UCL
Þ
KLA
2110
y
Non-killer
^
x
>?
Þ
KLA
2110
Non-killers
z
m
Total defects
on wafer
Out-of-Control
# of random defects x
– s.t. false alarm probability ? ?is less than a pre-specified value
that determines whether in-control or out of control , when we have only
one defect type, by the Neyman-Pearson Lemma (the regions are given
by f(x)/g(x) < c, and UCL can be computed from this)
^ so that the decision surface is obtained, and
mean shifted version of f(x,z),
^ space
approximate it by a hyperplane in the (x,z)
19
Assessment: What is Technology Worth?
Computer-based Process Reengineering
• Business Problem
– Business Need
– Technology Context
Review accuracy (r = q = probability of misclassification)
150
OOC
100
50
IC
197
183
169
155
141
127
99
113
85
71
57
43
29
1
0
Prob. of missing excursion ?
Upper
Control
Limit (UCL)
15
Additional observations:
1. Increasing levels of sophistications can include misclassification probabilities,
fixed or adaptive control limits
2. We have used dynamic programming to generalize sequential sampling
approaches to these environments, with an appropriate sufficient statistic 20
B-Risk Vs. Review Sample Fraction at Different Review
Accuracies, with different review level compensation
250
200
Control), where ^x is the killer defect estimate
^
^ since g(x,z)
^ is a
5. Reorganize the terms with the density functions f(x,z)/g(x,z),
18
Multi-dimensional SPC Chart For
Integrated Inspection-review Sampling:
Killer And Non-killer Defects
^ z) (In Control) and g(x,z)
^ (Out of
3. Consider the joint density functions f(x,
standard result and using the Euler equation to determine the optimal policy
to minimize the probability of missed excursions (subject to a given
probability of false alarms)
Random Defect Sampling with Classification Errors
• Incomplete and imperfect defect information
• Fixed versus adaptive control limits
• This is achieved through a control chart with an Upper Control Limit UCL
Compute conditional distributions of killer defects , condit ional on
observed total defects z, and the fraction (or equivalently, number) sampled
4. Use a generalized Neyman-Pearson Lemma, extending the results of the
Random Defect Sampling with Perfect Classification:
• Incomplete but perfect defect information
• Fixed versus adaptive control limits
• Minimize excursion detection time => equivalent to
• Minimizing the probability of missing an excursion ????
1.
2. Use standard central limit theorem for normal approximations of distributions
UCL(z)
(sampled
defects) ,
equivalent to
f = fraction sampled)
Total Defect Count Approach:
Complete and Perfect Defect Information
# of random defects x
Killer defect count estimator
Killer defect
estimate
Killers on wafer
X
Development of Multi-dimensional Control Chart
0.7
To achieve b=0.3
r=q=70% need
review 50% defects
0.6
0.5
• Technology Management
– Data to Information
– Information to Knowledge Management
0.4
0.3
0.2
r = 70%
r = 80%
r = 90%
0.1
0
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
• Information Systems
1
• Integration
Review sample fraction
Total random defect count
21
22
23
4
Information to Knowledge: Yield Management and
Learning Model
Goals, Issues and Economics: In-Line Defect
Sampling Strategy
A Yield Learning Cycle-time Optimization Problem
Þ
KLA
2110
DATA
INFORMATION
Isolation
Poly1
Poly2
KNOWLEDGE
Inspection &
Review Tools?
Which & how
many?
Metal1 Metal2
Market
Validation
Inspection
Corrective
actions
Review &
classification
Root-cause
analysis
Source
identification
Tool
Settings
Where?
% lots?
Wafers
per lot?
Inspection
% Wafer?
Sensitivity?
Isolation
In-control
Beta
Risk
1.2
? DD(t) &
? Y(t)
Poly 2
Metal
Wafer
Defect
Inspection Review
h -h
1-???? 2
Source
Corrective
Identification
Action
IC
Time
T0
SI(h)
0.4
Via
SR(h,f)
or
SI(h,f)
T1
T2
for In-line ADC
The process of excursion-correction can be represented
by a renewal reward process with expected unit costs
Metal 2
• Cycle time of each learning loop determine the
overall learning rate
• Effect of learning bottlenecks and economics
E[AC]= E[C]
E[T]
24
Illustrative Objective Function and Decision
Variables
25
Solution highlights
Objective
To minimize the detection delay (including queuing) of an excurs ion
26
Results: Quantified Advantage of Integrated
Inspection-review Sampling Methodologies
Optimization of sampling, control limits, and choice of
machines (and personnel)
Min Td (h, m) ? T? ( h,m) ? E[?] ? TI ( h) ? TR( h,m)
h ,m
h
T? (h, m) ?
Delay due to the beta risk of SPC Chart
1? ? (m)
h
E[?] ?
Excursion time in the wafer inspection interval
2
2
2
(? ? ? S )
T I (h) ? ? S ? S
Total inspection time per wafer (M/G/1)
2(h? ? S )
(? 2 (m) ? ? 2R( m))
T R(h, m) ? ? R (m) ? R
Total review time per wafer (M/G/1)
2(h ? ? R( m))
Decision Variables:
• Control limits,
• Sample fraction for defect classification f
• Wafer inspection intervals h;
1
?
0.6
Via
Data/information flow
Prod.
Delay
Tool Depreciation and Labor Costs
Poly
Resources &
strategies
Þ
KLA
2110
Costs
8
Use supermodularity of cost function, which results in
monotone non-increasing (in control limit)
property of sampling interval
Equivalence of cost minimization, cycle time minimization
under constant false alarm levels
1 Optimal inspection sampling
plan (no review)
7
2 Optimal inspection and review
sampling plan (separately)
5
3 Optimal integrated inspectionreview sampling plan
3
This is based on stable demand
Use real options and zero level asset pricing to handle
demand and price uncertainties
?=
18%
6
4
2
1
Additional work on sample path approaches, for
optimizing lot sequencing (LIFO in place of ubiquitous
FIFO)
27
?=
55%
9
0
28
1
2
3
29
5
Results on Review Methodologies:
Economics of Technology Investments in Fab A
Summary of Technology Management
Contributions
Expected yield lost due
to excursions per year
• Development of a novel generalized control chart that is
useful in the new integrated inspection & review context:
Data to Information
$35,000,000
$30,000,000
17,397,793
13,816,845
$10,000,000
$5,000,000
$Optical
• Technology Management
– Data to Information
– Information to Knowledge Management
• Development and analysis of a new prescriptive model that
determines the optimal inspection & review policies to
maximize the yield learning rate for a fab: Information to
Knowledge
$15,000,000
Traditional
• Business Problem
– Business Need
– Technology Context
Our research contributions are three-fold:
38,481,819
$40,000,000
$25,000,000
$20,000,000
Computer-based Process Reengineering
• Information Systems
• Demonstration of the value of this model in actual fab
context through Information Systems Integration
SEM
30
Yield Management Systems Required to Integrate
Defect, Yield, and Lot Data Do NOT Communicate
• Integration
31
Key to Success: ISM Solution
32
Computer-based Process Reengineering
In place today
Partially in place
Not in place
WorkStream/
Informix DB
SAP loader for offline
created data
Offline summary
of orders
Offline summary of
reports
Management
reports
Management
reports
SAP loader for
offline created
data
•
5 key partner firms and two dozen facilities globally
•
Doctoral students
•
Masters and undergraduate interns at facilities,
extracting data, developing business intelligence,
and integrating information systems for several
years nonstop to produce ISM manuals for fabs!
Mini -DSS (Decision Support System) solutions.
• Business Problem
– Business Need
– Technology Context
• Technology Management
– Data to Information
– Information to Knowledge Management
Order (PM & CM)
MDS
(Equipment
Interface Table)
Oracle DB
RTD for prev.
maintenance
PM & CM order data
Planned wafer starts
Global SAP D B
(SQL)
KLA Yield
Management
Systems
Defect data
YIELD DATA
• Information Systems
• Integration
33
34
35
6
Integration: Yield Management Systems
Successes in $200M-$1B Growth of K-T
Remaining Challenge
QC stories – “ Data Drownage”
• Training of marketing and application engineers during
The real need is to develop a large scale unified Yield
Management System integrating:
Monterey retreat on methodology/economics
• Marketing,including product definition, and “collateral”
• The overall business need for enterprise profitability
including marketing and sales
•
•
•
•
•
•
• Defect, parametric, and yield data and processes
• Lot movement data and processes
• Knowledge Management for Yield Learning overlaid on
the Product Development Process
development, guidance to engineering
Global “Customer” interface on methodology!
Seminars worldwide to thousands!
Executive impact and awards (including stocks!)
Academic Impact
Resources: Multi-million dollar effort
Faculty from Engineering (Systems and Domain) and
Management from Stanford, Berkeley, Carnegie Mellon
36
Backups
37
Information Systems and Technology
Management (ISTM) to Solve Business Needs
38
IS-based Process Learning
Business Need: Technology context – use or development of technology
• Generic Business Problem in High Tech
Integration: Business intelligence to maximize enterprise profitability
• Technology Management (TM) Challenges
–Facilities are challenged to develop new products and
process technologies rapidly
–Methods for rapid product and process learning
–Economics of process and monitoring tools
• Technology Management
– Identify specific goals
– Delineate Business Processes
– Model Economic trade-offs
– Capture Strategic Information Needs
• Domain Knowledge
• Analytic Model
– Stochastic Optimization
– Economics
39
• Information Systems
– Procedural or software system
– Knowledge Management System to
enhance
– Local, enterprise, or value network
performance
• Information Systems (IS) Issues
–Enhancements and limitations imposed on Technology
Management by IS/IT
–Knowledge Management
• Context -based Business Intelligence
• Needs: Domain knowledge, software technology
PLATFORM
• Software,hardware, and networks
• Psychology
• Integration
–TM and IS/IT integration challenges, including data
–Human interactions with Decision Support Systems
40
41
7
REDEFINED GOAL: MINIMIZING TIME TO DETECT
AND FIX YIELD EXCURSION
State-of-the-Art YMS: data ConductorEP ®
Key Solution Concept
( AN EXCURSION @ ETCHER IN FAB A)
Event
Occurring
Event
Detected
50
InControl
Source
Isolated
17
Detection Delay
7
Source
Isolation
In type-based excursion detection the sampling distribution
statistics of the killer defect estimates, given the total defects
on the wafer, are described by:
• Simple random sampling results in approximately Normal
distribution with statistics:
Fix
Validated
20
2
9
Hours
E[ Xˆ | X , z ] ? X
1
m ?1
VAR ( Xˆ | X , z ) ? ( )(1 ?
) X (z ? X )
m
z? 1
E[ Xˆ | z] ? ?
Root
Corrective
Cause
Actions
Analysis
X| z
1
m ?1
V A R(Xˆ | z) ? ( )(1 ?
)(Z? X| z ? ( ? X2 | z ? ? 2X / z )) ? ? X2 | z
m
z ?1
FXˆ | z ( xˆ ) is determined by the distribution of X & Y, and the sampling scheme
Defect level
Goal
Optimize procedures and inspection-review machine
usage to reduce delay to detect and fix yield excursion
• Using defects as surrogates ( linking defects to yield is a
technology problem in electrical/computer engineering)
• Trending by individual defect types (killer, non-killer)
• The two- dimensional control “limit”
hyperplane can be
characterized in terms of these parameters in (X, z) or ( X̂, z), by
generalizing the Neyman-Pearson Result
• The estimated standard error of the killer defect estimates is a
function of m and z; i.e., VAR ( Xˆ | z) ? g ( m , z)
42
43
Semiconductor Manufacturing: In-Line Monitoring
Wafer
Preparation
Wafer
Fabrication
Silicon
Ingot
Lot
Wafer
Probing
Agenda
Packaging/
Testing
Wafer
Triggered Learning Process
from Production to Product
Development
In-line Process Control
Yield
Time (days)
0
10
20
30
Days
30
40
50
5
10
15
20
25
30
35
40
45
50
Process Flow
Isolation
Gate 1
Poly
Gate 2
Poly 2
Contact
Metal
Via
Metal 2
Time (days)
0
10 20
Motivation
Triggered Learning Process
Current Implementation
Managerial Insights
Bigger picture & Future Work
Chip
Die
30
40
44
John Voit Delphi
Ram Akella UC SC/Silicon Valley Center
Rajeev Kishore SUNY
Ramesh Ramaswamy SUNY
50
5
Days
TEST
45
47
8
Motivation
Many problems throughout the production, assembly, and customer use are
solved by different parts of the organization. The lessons lear ned are then
archived in different formats and different levels of detail. These lesson learned
are not formally communicated to NPD due to organizational boundaries (real or
perceived), diverse storage media, and access privileges. If communicated, the
documents are sometimes too long, or are written in a context th at is not
immediately understandable for NPD use or absorb. This can result in the NPD
activity launching products that contain past problems (Busby, 1 999; Von Hipple
& Tyre, 1994).
First Time
Quality
Triggered Learning Process
Other attempts
Safety &
Formal
Ergonomics OEM Complaints
Post-Mortems
• Tend to be long reports that require discipline to prepare. Forexample, Microsoft sites 3-6 months
to prepare a 10-100 page post- mortem ( Thomke & Fujimoto, 2000)
• Ambiguous on how NPD will integrate information into new programs
Design Reviews with downstream stakeholders
• Downstream personnel cannot readily relate to NPD artifacts (i.e .; digital models) (Black & Carlile,
2002)
• Cross organizational information transfer (verbal and written) h ave problem of context and jargon
causing poor communication (Uschold & Grunninger , 1996)
• Time between reviews causes ‘batched’ learning and a greater chance and cost of iteration (Ha &
Porteus, 1995)
General Lessons Learned Database
• no process to make sure reviewed
Long Term
Durability
GOAL: An ontology-assisted triggered learning process (TLP) for getting Lessons Learned
communicated and used in NPD activities.
Lessons Learned
Long Term
Durability
Safety &
Formal
Ergonomics OEM Complaints
First Time
Quality
Warranty
Warranty
NPD = New Product Development
OEM = Original Equipment Manufacturer
NDP
2 PDP Production
?3 Delphi
3 OEM
OEM
NDP
2 PDP Production
?3 Delphi
Field <3 yrs Field >3 yrs
48
Definitions
Field <3 yrs Field >3 yrs
50
Triggered Learning Process
Lessons Learned Ontology Model
Ontology – “a set of concepts (e.g. entities, attributes, processes), their definition and their interrelationships; this is referred to as a conceptualization” ( Uschold & Grunninger, 1996)
Trigger - An event, called a Lesson Learned, that is communicated and used by NPD
• Assumption: the lesson learned was a ‘big enough’ problem that it was documented in
some manner by a part of the organization.
3 OEM
OEM
49
L e s s o n
L e a r n e d
F
p
u
e
n c t i o n
c i f i c a t
1
i
F
a
E
o
n
i
f
l
f
M
e
c
o
t
N
1
C
o
1
1
N
S
1
d
1
e
{
N
|
S
{
|
S
t
r r e c t i v e
D e l e t e ( )
A d d ( )
A
M o d i f y ( )
e p | + | F e a
t
c
u
t
r
i
e
o
|
n
>
=
1
N
1
N
R o o t
t e p | +
C a u s e
| F e a t u r
0 . . N
0 . . N
e
|
>
=
1
}
}
TLP is a structured approach that
Feeds back lessons learned created by downstream organizationalpersonnel
To a staff that condenses these lessons learned into an ontology a n d
Communicates these items to NPD
NPD personnel reacts to this information as it arrives by incorp orating
it into the newL product
or process under development.
L e s s o n s
e a r n e d
0 . . N
F
0 . . N
i
r
s
t
Q u a l
i
T
t
i
y
m
e
E
S
a
f
r
g
o
e
t
n
y
o
&
F
m
i
3
O
E OM E
c
s
O
E
o
r
M
C
m
a
o
m
L
0 . . N
C l a s s
n a m e
a t t r i b u t e s
o p e r a t i o n s ( )
{ c o n s t r a i n t s }
C
F
0 . . N
u
e
r
a
r
t
e
u
r
n
e
l
a
i
s
0 . . N
C u
S
t
r
e
r
p
e
n
W
0 . . N
t
n
o
D
l
p
K E Y
A s s o c i a t i o n
H a s
O n e
t
n
u
g
r
a
T
b
i
y
r
e
l
i
r
t
m
y
s
a
r
r
s
r
a
n
t
y
t
s
N
2 D
PP
D
P
P? r 3 o
d D u ec
lt
ip
oh n i
M
F
i
e
l
d
<
3
y
F
i
e
l
d
>
3
s
Step #1
Feedback
Step #2
Update
Lessons
Learned
Ontology
Step #3
Communicate
Step #4 React
51
52
53
9
Perspectives/Context
Step 2: Update Ontology
Step 1: Feedback Details
TLP is consistent with the theory that organizational learning is triggered by external shocks
(e.g.; lessons learned) that makes adaptation necessary (Cyert & March, 1963)
NPD as a problem-solving activity ( Thomke & Fujimoto, 2000). Enterprise and Customer
requirements are considered problems that must be solved by new product and process
development. Often downstream lessons learned are manifestations of a failure to meet
downstream stakeholders requirements.
What: Documentation that was created
in resolving the problem for the value
stream
Staff used
• To summarize the
Lesson Learned in
the ontology
• “Attach” the Lesson
Learned documents
provided
When: Feedback is initiated once a lesson
is learned and a value stream problem is
resolved.
Who: Each production lessons learned
source is owned by a different organization
(function) within the enterprise
54
Step 3: Communicate
i=1
Step 4: Reaction
Only potentially
relevant lessons go to
the appropriate teams
i=3
R 41 = R 51 = 0
Lessons
Learned
Ontology
i=5
Rij = Relevance
of Source i to PD
Team (PDT) j
Rij ? [0,1]
Lesson
Learned
j=1
RXNijk = Reaction to
source i from PDT j
regarding lesson k
Root Cause Present
=> Failure Mode
1
j=2
New Product Line
Profit
Quality
1
j=3
j=5
57
2a. Check
()
Root Ca
use
1
2
Ro b. C
ot hec
Ca k()
us
e
Reaction Alternatives
1. Implement Current
Controls
2. Innovate product
features or process steps
3. Need to plan
4. Not relevant - Do Nothing
j=4
i=4
56
Control Loop
1. Communicate
(Root Cause, Corrective Action)
R 11 = R 21 = R 31 = 1
i=2
55
New
Product
Design
present
3a. If root cause
ctive Action
Require() Corre
3b. If root cause present
Require() Corrective Action
1
1
1
3
Lesson
Reused
N
Team not
responding
New
Features
Lessons
Learned
Ontology
1
1
New
Process
Design
1
RXN3j1 RXN3j2 RXN3j3
Manager
Monitors
Response
s
1
4
3
1
4
3
1
2
2
N
New
Steps
Innovation
58
59
10
Triggered Learning Take Aways
Current Implementation
Triggered Learning Take Aways
ACTIVE Knowledge Movement process
• Right People: Email to the people developing the next generationproducts
• Right Information: Real Stakeholder Dissatifiers
• Right Amount:
Feedback established in United States and Mexico Production Value Streams (4 production facilities)
Product Development teams in US and Mexico are reacting to the information
Plans to expand to European Production facilities and Product de velopment.
Execution
• Improves Design Reviews
– Batched leaning to triggered learning - do not wait until reviews to share downstream lessons.
– Reduces Surprises & Opinions based comments
• Supports flawless launched to ensure past mistakes not repeated
• Helps maintain FTTQ and Health & Safety gains made in Value Stre am by communicating current fixes
– Lesson Overview Ontology line item
– Detailed solution information if required: Attached documents
•
Right Time: When A Lesson is Learned (Trigger)
Knowledge Application
PDTs have a reaction plan to use the new information
Controls & Standard Management Ensure process is followed
•
LL Update
Personnel
U.S.A.
Lessons
Learned
Ontology
LL Update
Personnel
Mexico
60
Managerial Insights
61
Required Resources
Some Keys for Implementation
Feedback:
When is this systems most valuable?
• Product Maturity High (i.e.; many small issues)
• Past product highly relevant to new products
• High Cost of making or repeating mistakes
• Project time short & project teams highly utilized
• Number of independent future projects high
• Organization structure: Information Bucket Owners different that product process
developers
•
•
Information technology already present
Needed to spend time to develop process and get buy-in.
• Feedback People
– Minimal Impact
• Update People
– New Responsibilities
• Communication People
– Standard Management Responsibilities
• Reaction People
– Spend time now or spend time later
– Not new responsibility, but new information, not always enough time to address
issues
Add steps to current problem solving processes to send Lessons L earned to update person
Be on guard for those partial lessons learned
Update:
•
•
•
Development of the Ontology Structure
Populate with historical projects to prove concept and work out bugs
Not a Clerk Job, people must have some domain knowledge
Communicate:
•
Works Best when communication person is also the Manager of Reaction People
React:
•
•
•
63
62
Be patient.
Team will be getting information they never received before
Standard Management/coaching commitment in beginning must be high
64
65
11
A Bigger Picture
Interesting Research area slicing across traditional disciplines
•
•
•
•
•
Management Science
– Q: How do decision makers know when having system and staff is important?
– Tools: Analytic Queuing Models
Knowledge Management
– How do we get information to the people who need it?
– Tools: Organizational Design, Communication Processes.
Organization Learning
– How do groups and people learn?
– Tools: Observational studies, and experiments
Philosophy
– Q: How do we structure knowledge to best get our questions answered?
– Tools: Ontology, Taxonomy
Computer Science
– Q: How can this be an even more automatic/active system?
– Tools: Object oriented programming and systems.
66
Future Research & Work at Delphi
Bibliography
Increase scope of buckets and plants, deeper into supply chain
Link to “Double Loop” learning (Agyris, 1976). The idea is that a concentration of lessons
learned indicates a systematic problem, and Organization policy should be changed.
Argyris, C. 1976. Single-Loop and Double-Loop Models in Research on Decision Making. Administrative Science Quarterly, 21(3): 363-375.
Black, L. J., & Carlile, P. R. 2002. Managing Knowledge in a Product Devlopment Process: What to Do and When,Working Paper: 40.
Busby, J. S. 1999. Problems in error correction, learning and knowledge of performance in design organizations. IIE Transactions, 31: 49 -59.
Cyert, R. M., & March, J. G. 1963. A Behavioral Theory of the Firm. Englewood Cliffs, NJ: Prentice Hall.
Ha, A. Y., & Porteus, E. L. 1995. Optimal timing of reviews in concurrent design for manufacturability. Management Science, 41(9): 1431-1447.
Thomke , S., & Fujimoto, T. 2000. The Effect of " Front-Loading" Problem -Solving on Product Development Performance. Journal of Product Innovation Management,
17: 128 -142.
Uschold , M., & Gruninger, M. 1996. Ontologies: principles, methods, and applications. The Knowledge Engineering Review, 11(2): 93 -136.
v o n Hipple, E., & Tyre , M. 1994. How "learning by doing" is done: Problem identificati on in novel process equipment. Research Policy, 19: 1 -12.
67
68
12