8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
1/39
Evan Rash and Karl Kempf
Decision Engineering Group
Intel Corporation
Product Line Design
and Schedulingat Intel
INFORMS ANNUAL MEETING 2011
Charlotte, North CarolinaCPMS Daniel H. Wagner Prize Competition
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
2/39
2
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Agenda
1.Business Background
2.The Strategic Business Problem
3.Mathematical Formulation
4.Our New Solution
5.Our Custom Implementation
6.Growing Business Impact
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
3/39
Founded
July 18th, 1968
3
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
4/39
Leading Edge Process Technology
32nm45nm
1x
0.1x
0.01x
0.001x
65nm 22nm*
LowerTrans
istorLeakage
LowerTrans
istorLeakage
Higher Transistor Performance (Switching Speed)Higher Transistor Performance (Switching Speed)
* projected
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
5/39
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
6/39
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
7/39
7
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Market1 $13 240,000Market2 $15 300,000Market3 $14 450,000Market4 $12 880,000Market5 $ 9 900,000
Marketing ASP Vol
Different Markets Need
a Different Mix of Featuresmarke
ts
features
Selling the product in the market brings in revenue
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
8/39
8
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Feature1 Feature2 Feature3 Feature4 Feature5 Feature6 Eng&
Mfg
$300,000 $400,000 $400,000 $250,000 $300,000 $200,000 EngCost$ 1.50 $ 0.35 $ 1.25 $ 0.50 $ 0.50 $ 0.25 MfgCost/u
Market1 $13 240,000Market2 $15 300,000Market3 $14 450,000Market4 $12 880,000Market5 $ 9 900,000
Marketing ASP Vol
Different Markets Need
a Different Mix of Featuresmarke
ts
features
Engineering and manufacturing incurs costs
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
9/39
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
10/39
1 0
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Different Markets Have
Different Timings
Markets are not all synchronized in time
markets
features
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
11/39
1 1
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Different Features Have
Different Availabilities
timeFe
ature1
Featu
re4
Feature3
Feature
2
Feature5
REV 1.0
BUILDBUY
REV 2.0
Partial Reuse
markets
features
Feature development must besynchronized with market
windows
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
12/39
1 2
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Constraints
Feature sets in the products must meet (or exceed) theneeds of the target markets
Features must be engineered in time to be integratedinto the products
Products must be engineered and manufactured to hitthe market timings
The engineering budget is finite (leading to an emphasison reuse)
Maximize Profit (Max Revenue, Min Eng and Mfg Cost)
Objective
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
13/39
1 3
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Business questions include (at least):
Given an engineering budget, what set of productsmaximize revenue or profit?
Given a revenue target, what set of products
minimize cost, with what engineering budget? Given a number of Features to engineer, what is
the profit maximizing order of development?
Given a Feature build vs. buy decision (cost,timing), which generates the most profit?
Difficult to solve with standard techniques due to manydifferent constraints, competing objectives, and interrelated
tradeoffs
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
14/39
1 4
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Math
Define Problem & Formulate as MathematicalProgramming
Show Complexity & Difficulties involved with
Traditional techniques
Solution Methodology & Implementation
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
15/39
1 5
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
The Core Problem
Generate a Product Line Strategic
Map products into markets
Schedule product development
Generate Product FeaturesTactical
Meet or exceed market requirements
Schedule feature development
Optimize for Profitability StrategicProduct line must optimize profitability
Must consider engineering budgets
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
16/39
1 6
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Generating the Product Line
Inputs
Set of marketsNumber of products
Time horizon
DecisionsHow many products to
build
When to introduce
products
Which markets to sellproducts into?
M,...,1
TT ,...,,...,1 0
p
TTzp ,...,0
pmt
Binary
Integer
Binary
At most one product per market P
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
17/39
1 7
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Generating Product Features
Inputs
Set of featuresMarket Requirements
DecisionsProduct Features
Feature Availability
F,...,1
pfx
TTy f,...,
0
Units of Feature f in Product p
Integer
Integer
mfD
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
18/39
1 8
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Optimize for Profitability
InputsMarket Volumes andPrices
Feature EngineeringCost (with Reuse)
Product Engineering Cost
Feature Mfg. Cost
ExpressionsRevenue
Engineering Cost
mtmt pv ,
fc
A
tRf
P
p
M
m
T
t
F
f
pffmtmtpmt xcpv
0 0 0 0
F
f
T
tf
P
pp tRA
0 00
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
19/39
1 9
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Reuse Function
Engineering featurespresents reuseopportunitiesDeveloping Feature 3 may
cause developing Feature4 to be cheaper/faster
The Reuse Functiondefines these reusesynergies
Typically dynamic andcomplextime
Feature1
Feature4
Feature3
Feature2
Feature5
REV 1.0
Feature 3
Feature 4
BUILD
BUY
REV 2.0
Partial Reuse
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
20/39
2 0
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Reuse Function Example
A hypothetical Reuse Function where developing one feature in agroup causes subsequent feature development to be 50% cheaper
Featuref Group G(f) when 1 1 1 1
2 2 .5 if
2 3, else 1
3 2 .5 if3 2, else 14 3 .5 if4 5, else 15 3 .5 if5 4, else 1
if
=
if
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
21/39
2 1
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Full FormulationObjective: Maximize Profit
max 0 0
0
0
T0
0
0
0 1 , One Product per Market ,, Market Satisfaction Constraint max: 0 Product Availability Constraint 0,, Market Coverage Availability Constraint 00 Product Selling Requirement
0 Resource Constraint
0,1 Binary Constraint
0,1 Binary Constraint 0, ,max Integral Units of Features Constraint 0 , , Scheduling Window Constraint 0 , , Scheduling Window Constraint
Subject to:
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
22/39
2 2
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Why Is This a Hard Problem?
Non-linearity
Reuse Function
Objective Function & Constraints
Integral & Binary Decisions
Scheduling
Mapping Combinatorics & Problem Size
Difficult to solve by traditional techniques!
Linear/Mixed-Integer ProgrammingConstraint Programming
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
23/39
2 3
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Our Solution
Integrate diverse OR techniques
Resource-Constrained Job Scheduling
Optimal Set Covering
Portfolio Optimization
Dynamic Programming
Decompose Problem into Multiple Stages
Outer strategic Genetic Algorithm
Inner tactical Heuristics and MIPs Financial Optimization through Genetic Algorithm
Fitness
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
24/39
2 4
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Decomposition Product Line Design
Strategic Tactical
Out e r Gene t i c Algo r i th mI n n e r Se t Co v er i n g / H eu r i s t i cs
1. Product Release Schedule
2. Product to Market Mappings
3. Product FeaturesExt : Featu re Substit ut ions
4. Feature Schedules
5. Feature Reuse6. Resource Constraints
7. Financial Objectives
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
25/39
2 5
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Decomposition Generate Product Features
Strategic Tactical
Out e r Gene t i c Algo r i th mI nne r M I Ps and Heu r i st i cs
1. Product Release Schedule
2. Product to Market Mappings
3. Product FeaturesExt : Featu re Substit ut ions
4. Feature Schedules
5. Feature Reuse6. Resource Constraints
7. Financial Objectives
Per Product
Per Product
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
26/39
2 6
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Decomposition Financial Optimization
Out e r Gene t i c Algo r i th mI n n e r Se t Co v er i n g / H eu r i s t i cs
1. Product Release Schedule
2. Product to Market Mappings
3. Product FeaturesExt : Featu re Substit ut ions
4. Feature Schedules
5. Feature Reuse6. Resource Constraints
7. Financial Objectives
Selection
Mutation & Crossover
Per Product
Per Product
Strategic Tactical
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
27/39
2 7
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Outer Strategic Algorithm
1. Outer: Creating Product Schedules
Generate a random chronologically sorted productschedule, with some products turned off.
Use crossover to zip different schedules togetherand mutations to randomly permute schedule by
pushing products out and pulling products in
2. Outer: Creating Market to Product Mappings
For each market randomly cover or skip the market.If covered, select a random product from the listgenerated in 1
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
28/39
2 8
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Inner Tactical Algorithm
3. Inner: Determine Product Features (MIPs)
- Cover market requirements with minimum manufacturing cost
- Cover market requirements with minimum engineering costRandomly alternate and allow the evolutionary process to pick thebest
4. Inner: Deduce Feature Schedules
Back out the feature engineering schedule based on when thefeatures need to be available for the products availability (1)
5. Inner: Evaluate Reuse
Evaluate the reuse of the feature schedule from (4)
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
29/39
2 9
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Outer Strategic Algorithm
6. Evaluate Resource Constraint
Evaluate the engineering resources for the entireroadmap
Model engineering resource constraints as softconstraints
Use a Lagrangian penalty approach similar to theconcept of an overtime cost of exceeding the availableengineering resource supply
7. Evaluate NPV & Fitness
Evaluate the fitness of the product line by determining its
NPV and subtracting out any resource overage penalties
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
30/39
3 0
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Pinning Parts of the Solution
Planning involves many strategic aspects
Not always possible to solve with a clean slate
Solver must be able to pin portions of thesolution in place and solve using remainingdegrees of freedom
Examples
Locking products onto the roadmap
Locking feature availability schedules
Forcing entry into particular markets
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
31/39
3 1
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Implementation
Custom Implementation(C# .NET)
Required CustomMutation/Crossover andSolution Flow
Inner sub-problem
solved via modularheuristics plugged intolarger GA
Most Heuristics: C#
Feature Substitution:OPL CPLEX
Database
Reporting and Analysis
Scenarios
Solver
Inner Sub-Problems
Outer GA
C#
C#
C#
C#
CPLEX
SQL
90%
10%
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
32/39
3 2
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
The Business Process
BEFORE
1) Many spreadsheets with localdatabases
2) Local view by product,sometimes by division
3) Few what-ifs
AFTER
1) One tool with global database (HWand SW)
2) Holistic view across divisions andproducts
3) Many what-ifs
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
33/39
3 3
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
The Business Process
BEFORE
1) Many spreadsheets with localdatabases
2) Local view by product,sometimes by division
3) Few what-ifs
4) Difficult decision making betweenfinance, planning, and engineering(design and mfg)
5) No global optimization and little
(if any) local optimization
6) Little reuse between divisionsand within divisions
AFTER
1) One tool with global database (HWand SW)
2) Holistic view across divisions andproducts
3) Many what-ifs
4) Collaborative decision makingbetween all of the product functions
5) Global profit optimization
6) Increasing reuse across divs andproducts (few%/mo)
U D d F db k
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
34/39
3 4
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
User Data and Feedback
0
2 0
4 0
6 0
8 0
1 0 0
1 2 0
1 4 0
0
7
-1
0
0
8
-1
0
0
9
-1
0
1
0
-1
0
1
1
-1
0
1
2
-1
0
0
1
-1
1
0
2
-1
1
0
3
-1
1
0
4
-1
1
0
5
-1
1
0
6
-1
1
0
7
-1
1
0
8
-1
1
0
9
-1
1
1
0
-1
1
USERS BYMONTH
U D t d F db k
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
35/39
3 5
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
User Data and Feedback
JOBTITLE #USERS HRSUSED
STRATEGICPLANNERS 16 996
PROJECT/PROGRAMMANAGER 39 476
PRODUCTDESIGNENGINEER 23 394
FINANCIALANALYST 39 342
PRODUCTMARKETINGENGINEER 5 63
OPERATIONMANAGER 3 21
PRODUCTSOFTWAREENGINEER 8 4
TOTAL 133 2296
0
2 0
4 0
6 0
8 0
1 0 0
1 2 0
1 4 0
0
7
-1
0
0
8
-1
0
0
9
-1
0
1
0
-1
0
1
1
-1
0
1
2
-1
0
0
1
-1
1
0
2
-1
1
0
3
-1
1
0
4
-1
1
0
5
-1
1
0
6
-1
1
0
7
-1
1
0
8
-1
1
0
9
-1
1
1
0
-1
1 PRODUCT #OF
DIVISION USERS
Div1 58
Div2 25
Div3 12
Div4 11
Div5 9
Div6 4
Div7 3
Div8 3
Div9 2
DivAdmin 3
Misc 3
TOTAL 133
USERS BYMONTH
U D t d F db k
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
36/39
3 6
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
User Data and Feedback
JOBTITLE #USERS HRSUSED
STRATEGICPLANNERS 16 996
PROJECT/PROGRAMMANAGER 39 476
PRODUCTDESIGNENGINEER 23 394
FINANCIALANALYST 39 342
PRODUCTMARKETINGENGINEER 5 63
OPERATIONMANAGER 3 21
PRODUCTSOFTWAREENGINEER 8 4
TOTAL 133 2296
0
2 0
4 0
6 0
8 0
1 0 0
1 2 0
1 4 0
0
7
-1
0
0
8
-1
0
0
9
-1
0
1
0
-1
0
1
1
-1
0
1
2
-1
0
0
1
-1
1
0
2
-1
1
0
3
-1
1
0
4
-1
1
0
5
-1
1
0
6
-1
1
0
7
-1
1
0
8
-1
1
0
9
-1
1
1
0
-1
1 PRODUCT #OF
DIVISION USERS
Div1 58
Div2 25
Div3 12
Div4 11
Div5 9
Div6 4
Div7 3
Div8 3
Div9 2
DivAdmin 3
Misc 3
TOTAL 133
USERS BYMONTH
W e a r e f i n al l y w o r k i n g i na t r ansparen t sys tem
ins tead o f sp readshee ts onrandom shared d r i v es.
Use fu l as an acum en too las w e l l as lea rn ing abou tw he re syne rg ies ex i st f o r
ou r p roduc t s .
N o id e ah o w w e ca n
op t im i zem a r k e t
coveragew i t h ou t atoo l l i ke
th is .
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
37/39
3 7
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Conclus ion
This is a complex problem consideringmarket, feature, and product time dynamics
Extremely difficult to solve with traditionaltechniques
Developed and implemented a custom
solution to the problemThe system currently has users acrossdivisions and job roles
We believe the system (over time) willbecome crucial to Intels continuing success
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
38/39
8/2/2019 Product Line and Scheduling at Intel_Kempf_Wagner Presentation
39/39
3 9
CPMS 2011 Daniel H. Wagner Prize Competition Intel Corp 2011
Extensions
Feature Substitution
Feature A or Feature B can be interchanged
Time to Market Penalties
Late products suffer in the marketplace
Minimum vs. Target Market Requirements Feature A is a must-have, Feature B is a value-add
Build vs. Buy decisions
Develop in house or license?NPV Optimization
Top Related