Brian Subirana & Sanjay Sarma, MIT Subirana & Sanjay Sarma, MIT. History z 1998: DISC founded ... Huge channel inventories ... Distribution Center 10%

ESD.290Special Topics in Supply Chain

Management

Brian Subirana & Sanjay Sarma, MIT

History

z 1998: DISC founded z 1999: Auto-ID Center founded Auto-ID Field Trial

started 2000 z 2001: First standards presented z 2002: Gillette orders 500,000,000 tags from Alien z 2003: Wal-Mart, DoD Mandates

EPCglobal launched, Center retired z 2004: More mandates

Outline

z RFID and the Auto-ID Center z An in-depth look at some issues

Outline, Part I

z RFID and the Auto-ID Center z What and why of RFID z The cost issue z Manufacturing low-cost RFID z Handling the data z Current status

z An in-depth look at some issues

Outline, Part I



Magnitude of Challenges

z Inventory Management: z Inventory uncertainty:

z 65 % of 370,000 records inaccurate (HBS study of one major retailer) z Transportation uncertainty: Perfect delivery is dismal

z Stock-outs:z Average 9% out of stock in retailers world-widez Lost sales due to stock-outs: 4%

z Overstock: Huge channel inventoriesz CPG average 11 weeks inventoryz Retailers average 7 weeks inventoryz Locked up capital, industry-wide

z Brand Management: z Counterfeit:

z $500B pharmaceuticals business, $50B counterfeit z Diversion:

z Market size difficult to estimate

The problems are everywhere (

0.0 4.0 8.0 10.0

l i

Across geographies

Hai

Toil i

Sal

)

Across product lines

Overall OOS Extent Averages)

7.9

8.6 8.2

8.3

2.0 6.0

USA

Europe

Other Regions

Wor dw de

Percent OOS

OOS Averages by Category

9.8

7.7

7.0

6.8

6.6

5.3

8.3

0.0 2.0 4.0 6.0 8.0 10.0 12.0

r Care

Laundry

Diapers

Fem Hygiene

et T ssue

ty Snacks

World Avg (18 categories

Percent

(

7.3

9.1

Fri

Across time

OOS by Day of Week Average of 13 studies)

10.9

8.7

9.8

10.0

10.9

0.0 2.0 4.0 6.0 8.0 10.0 12.0

Sun

Sat

Thur

Wed

Tues

Mon

Percent

Other Cause 4% Store Forecasting

Retail HQ or 13% Manufacturer

14%

Distribution Center 10%

Store Ordering 34%

Store Shelving25%

Across the supply chain

The Vicious CycleUpstream In-store

causes causes Demand variations

28% 59% 13% 14%

Upstream problems

50% Store Problems at ordering

manufacturer’s DC Shelf 35%

Problems at retailer’s DC

Replenishment

8% Manufacturer Manufacturer’s DC Retailer’s DC Retailer World

11 weeks inventory Wide

6 weeks inventory

Consider stock-outs

RFID System

RFID

Low cost rfidhandling cost

die

size

/cos

t, ce

n ts

Silicon: 4c/mm2

20

15

10

5

time

Why is RFID expensive today?

reduce functionality

increased chip sizegreater functionality

(Networking & software)

reduce chip size (handle small chips)

Cheap protocol

10 20 30 40 50 60 70 80 90

i i

100

Noi

se

Gauss an F t 0.8 912 100 2 Meters, 4 KHz, SQW, 1 dBm

905 910 915 920

Frequency (MHz)

0

are

context- ware router

context-aware router

router

context-rout

router

co

the hypothesis or bet01. 203D2A. 916E8B. 8719BAE03C

Header 8 bits Serial Number 40 bits

Product 24 bits

Manufacturer 24 bits

z Place unique number on tag z Electronic Product Code, EPC z 64 bit, 96 bit, and upwards

z Develop manufacturing technology for small chips and tags

z Move data on the network z Network service for resolving EPC z Network architecture for gathering

and routing data

a

er

context-aw router context-aware

router

context-aware

aware context-aware


ntext-aware router



sensor sensor sensor sensor

Outline, Part I



ICDesign

Low cost RFID

Antenna Manufacture

EndIC Manufacture

IC Antenna/IC Assembly

Conversion to Package users Design

$X Million 20¢ 5¢ 5¢ 20¢ Billions Number of tags

1-2¢ 1¢1¢1¢

Challenges of IC minimalism

z 0.25 mm2: does it make life tougher? z ……..

z Street width will dominate

z Still have to test the IC’s (?)z Die handling costs are high

z Die-attach/wire-bonding techniques do not scale











low cost rfid challenges

Antenna Manufacture

IC Manufacture

Antenna/IC Assembly

Conversion to Package

20¢ 5¢ 5¢ 20¢

1-2¢ 1¢ 1¢ 1¢

Testing

z Economics today: z $500 - $1000 per wafer

z But minimal functionality means z High reliability z Don’t test on wafer z Test wirelessly at conversion

Slicing and Dicing

z

z Instead, use separation by thinning

Standard saw-dicing wasteful

C. Landesberger, S. Scherbaum, G. Schwinn, H. Spöhrle: “New Process Scheme for Wafer Thinning and Stress-free Separation of Ultra Thin IC’s,” Proceedings of Microsystems Technologies 2001, Mesago, Stuttgart, pp. 431-436, 2001.

Low cost RFID challenges

Antenna Manufacture

IC Manufacture

Antenna/IC Assembly


20¢ 5¢ 5¢ 20¢

1-2¢ 1¢ 1¢ 1¢

Antenna

z Screen printing z Etching z Forming


Antenna Manufacture

IC Manufacture

Antenna/IC Assembly


20¢ 5¢ 5¢ 20¢

1-2¢ 1¢1¢ 1¢

Assembly

z Fluidic Self Assembly z Vibratory Assembly z Pick and place

Vibratory AssemblyChip

Assembly Chip

Inlet

End User/ Design

Label Converting Assembly

Silicon Manufacturing

Antenna Manufacturing

Label Tag Manufacturing

Wafer Treatment


Antenna Manufacture

IC Manufacture

Antenna/IC Assembly


20¢ 5¢ 5¢ 20¢

1-2¢ 1¢ 1¢ 1¢

Conversion

z Paper/package/label industry expertise z Scales well with mass production z Capital equipment expenditure

software

paperhardware

Outline, Part I



Architecture: Local

Reader

Local database

Reader

Local network

Data Processing

Middleware

01.203D2A.916E8B.8719BAE03C

Tag

AE03C

Architecture: Global

Reader

Local database

Reader

Local network

Local system

ONS

Internet <EPC= >

</EPC> </TIME>

????

Quality control specialist

XQL

PML Server

18.72.100.100

18.72.100.100

PML

… … ...

40 DEG C </MaximumTempearture>

5 DEG C

Data Processing

Middleware

01.203D2A.916E8B.8719BAE03C

<PML> <TIME=2000.4.28:10:05.05HRS>

01.203D2A.916E8B.8719BAE03C <TEMPERATURE=15 DEG C>

</PML>

01.203D2A.916E8B.8719B <PML>

<MaximumTempearture>

<MinimumTemperature>

</MinimumTempearture> </PML>

23AB.36C2.AB21.6733

AE03C

Tag

Reader

Local database

Reader

Local network

Local system

ONS

Internet <EPC= >

15 DEG C> </EPC>

</TIME>

????

Quality control specialist

XQL

PML Server

18.72.100.100

18.72.100.100

PML

… … ...

40 DEG C

5 DEG C

Data Processing

Inference Temperature OK

Middleware

<PML> <TIME=2000.4.28:10:05.05HRS>

01.203D2A.916E8B.8719BAE03C <TEMPERATURE=

</PML>

01.203D2A.916E8B.8719B <PML>

<MaximumTempearture>

</MaximumTempearture> <MinimumTemperature>

</MinimumTempearture> </PML>

01.203D2A.916E8B.8719BAE03C

Trading PartnersTrading Part

Three Layers of an EPC Architecture

ONS

EPC Enterprise •EPC Enterprise data store

nersTrading Partners

Business IDs + •Data migration from Edge

•Standard API •Enterprise systems •Trading partners

Enterprise

Edge

Edge 3EPC Edge Edge 2 object Edge 1

loc time

•••

•EPC Edge data store ••

Exis

ting

Syst

ems

/ EA

I

RF Abstraction layer Device mgmt Event mgmt

Standard API Data migration to Enterprise

Software

Hardware Readers •Physical data capture

Tags

Outline, Part I



Field Trial

II

II I I

I

I IO

COCA COLA BOTTLER CLEVELAND, TN

JOHNSON & JOHNSON OL VE BRANCH, TN

KRAFT FOODS FORTH WORTH, TX

GILLETTE D ST. CENTER CHICAGO IL

PILOT TEST FAC LITY BENTONV LLE, AR

WAL-M ART DEPOT BENTONV LLE, AR.

WAL M ART STORE CLEVELAND, TN

Warehouse Retail Floor Staging Area Retail Floor

SAM'S DEPOT KANSAS C TY, M O

WAL-MART STORE BROKEN ARROW, OK

SAM'S STORE TULSA

P & G FACTORY CAPE G RADEAU , MO

UNILIVER DIST. CENTER BALT MORE, MD

P & G DC IOWA CITY,

Pilot facility is being used as a mini warehouse

The Commercialization of EPC

z Landmark Event: EPCglobal is formed z Many companies have significant tests and pilots underway z Mandates: z DoD

z Marks & Spencerz Tesco

z Wal-Martz Metro Group

z Targetz Albertsons

z Best Buyz Other major retailers are continuing to announce their strategies

RFID Status

z 3 protocols z Class 0 UHF z Class I UHF z Class I HF

z Tens of manufacturers z Tags: Alien, Matrics, Philips, ST Micro, Rafsec, …. z Readers: Alien, Matrics, AWID, ThingMagic, Tyco, Symbol,

Samsys,…

z New versions being designed z Gen 2 taking off z Intermec patent still issue

Key philosophy #1: interoperability

Sensors

Open tag

Agile reader

Software

Internet

Key philosophy #2: Layers

Communicate with Classes IV and V.

Active tags with

Clsemi-passive RFID tags

Clpassive tags with additional

functionality

read-only passive tags

Upw

ard

com

patib

ility

Dow

nwar

d fa

ilsaf

e

Class V tags Readers. Can power other Class I, II and III tags;

Class IV tags:

broad-band peer-to-peer communication

ass III tags:

ass II tags:

Class 0/Class I:

Vendors

z Chips: Alien, Matrics, Philips, ST Micro

z Readers: Alien, Matrics, Philips, Tagsys, Samsys, ThingMagic, Tyco, Symbol, Markem, AWID

z Software: Sun, Oat Systems, Manhattan, Globe Ranger, Conecterra, SAP, Tibco, Verisign, Vizional, ..

z Systems: Accenture, PWC/IBM, GEA, … z End-Users: Gillette, Wal*Mart, P&G, TESCO, Metro,

Target, Wegmans, ….

Research Issues

z Tag anti-collision z Reader anti-collision z Security and privacy z Advanced sensor networks z Data routing and handling z IC Design z IC manufacturing

z Silicon processing z Chip assembly

z Polymers z Controls/automation z Manufacturing systems z System Synthesis z Supply chain issues

Outline, Part I



Outline, Part II

z RFID and the Auto-ID Center z An in-depth look at some issues z A peek at the protocol z Security and Privacy Issues z Software z Vibration analysis z Silicon manufacturing

Components

z Signaling z Anti-collision z Functions

reader to tag: modulation

100%

10%

Low Time

45%

45%

Reader to Tag ModulationUHFEnvelope

On Off Keying (OOK), Min 90% Modulation Depth

Modulation: reader to tag

16 us

Cl i li

62.5 Khz

k Low nterva for data = 1

Clk Low nterval for data = 0

tag to reader

0

1

1 Bit Cell• Bit Cell Time: ~8 µs Tag to Reader (128 kbs) • 2 Transitions = 0• 4 Transitions = 1• Always Transitions Within a Bit

• Bit Cell Time: ~8 µs Tag to Reader (128 kbs) • 2 Transitions = 0• 4 Transitions = 1• Always Transitions Within a Bit

Anti-collision

z A Reader Talks First (RTF) System z Commands Issued from Reader z Tags Reply at a Later Time While Reader

Listens

z Transactions are Self-Contained Operations (Minimal Persistent State Information Required)

Contention Detection

Anti-Collision Algorithm Relies on Detecting Contention (When More than One Tag is Responding to a Reader Command).

0 1 0 1 1 0 0 10

100% Tags, Same Clock Rate, 1-Bit Difference

Contention- Two

Anti-Collision

0110

01100 01101

011001 011011

0111

01110 01111

011100 011101 011110

0

00 01

001 010 011

1

10 11

100 101 110 111

001 100 111

[CMD] = 00001000 (Ping)[PTR] = 00000000 [LEN] = 00000011 (3)[VALUE] = 011

[CMD] = 00001000 (Ping)[PTR] = 00000000 [LEN] = 00000000 (0)[VALUE] = 0

011000 011010 011111

000

Query 1

Query 2

000 010 011 101 110

Functions

z Write address z Lock address z Preload address mask z Read ID (anti-collision) z Read payload z Write payload z Sleep z Wake z Destroy

Outline, Part II


Does protocol compromise privacy?

Not necessarily. Your choice.

z You can destroy the tag and opt out

z or

z You can keep tag for later use z (physics is your friend)

Mass hijack of tags

z Could happen in destroy or re-programming

z Physics our friend z Bandwidth limited: 100’s of tags a second anti

collision

z Destroy must be individually addressed

z So it takes time to killz Surveillance

Issues

z Tags are light-weight

z Anyone can read the tags (promiscuity) z The same number shows up all the time z Channel is open and shared

01. 203D2A. 916E8B. 8719BAE03C

Header 8 bits Serial Number 40 bits

Product 24 bits

Manufacturer 24 bits

Problem: unique and promiscuous

Kill Serial number?z Product still readable

z Person can be tracked by constellation

Personalize the number?z Repeated reads yield same numberz You could still be tracked by constellation

Check out EPCglobal

Public policy

www.epcglobalinc.org/public_policy/ public_policy_guidelines.html

Outline, Part II


Outline, Part II


Outline

z Introduction

z Kinematics z In-plane surface motion z Out-of-plane motion

z Adhesion and fluid effects z Experiments z Conclusions

Introduction

Motivation

z Need methods for handling of smallmicroscopic parts in tag production processes

z A traditional technique to solve scalingproblems is parallelization

Topics

Basic kinematics of vibratory parttransport

Effects of fluid, andsurface/adhesion forces

Outline

z Introduction z Kinematics z In-plane surface motion z Out-of-plane motion

z Adhesion and fluid effects z Conclusions

Vibration Kinematics

z Want relative motion z Approaches: z Surface micro-features

z MEMS cilia, rollers, etc. z Gas flow nozzle arrays

z Electrical/magnetic fields z Moving Fluid medium flowing over surface

z Fluidic Self Assembly (FSA Alien Tech’s) z Vibrating surfaces

z Time asymmetric in-plane vibrations z Out-of-plane vibrations (such as in Bowl Feeders)

Vibrating Surfaces

In-plane vibrations of a surface if time-asymmetric

z Part moves if surface acceleration is more than friction

z Forward-backward motion? z Much literature

Example: Stick Slip

Surface accelerating faster than friction can

sustain

Gravity friction cone

v

x

Out-of-plane vibrations

z Vertical vibration: extra degree offreedom

z Create hop, move platen back z Small-scale forces z Increased accelerations may be

needed to compensate

Part in flight

Hopping example

Surface accelerating faster than gravity downwards

Gravity friction cone

Possible return path

Part in collision with surface

Outline



Adhesion

z Adhesion effects become important at smallscales

z Van der Waals forces z Due to static and quantum mechanically induced dipoles z Strong role in inter-molecular and surface phenomena z Become important at < 100nm surface separation z Adhesion surface energies of ~100mJ/m2

z Clean atomically smooth surfaces in contact may haveadhesion pressures of the order of thousands ofatmospheres

Adhesion effect on vibratory transport z Larger accelerations are needed

z Too contaminated or too clean are bad

z Can be alleviated by: Roughening or Surfactants

z Strong secondary excitation can “levitate” the parts

z If done for fractions of the cycle, “creeping” transport can be achieved, where the part moves when vibrations are on and sticks when they are off

Example: hopping with adhesion

adhesion friction cone

collision

Flight after unstick

Surface acceleration

Sticky

Fluid dynamic scaling

z Sticking due tofluids

z Alleviation z use surfaces that

are “leaky” z Make chip

surface bumpy z Run in vacuum

velocity loss due to film damping

Frequency effect on

gap effect on

400 µm dia. 200 µ

damping

Starting

damping

Si disk

m thick

Outline



Conclusions

z Basic physics

z Ongoing work: z Measurements z Effects of geometries z Test methods

z Design: z Chip delivery methods and roll to roll packaging

Outline, Part II


RFID manufacturing simulation

Gita Swamy Sanjay Sarma

Outline

z Components of an RFID tag z Understanding the Experiment z Manufacture zSemiconductor zTag

z Results z Worldwide Fab Capacity

Results

z Total RFID Cost zIC + Traditional Assembly: 4.351¢ zIC + Flip Chip Assembly: 3.311¢

z IC Cost: 1.151¢

z Antenna Cost: 11¢

z Assembly: 2.25c ~ 1.15 1¢

z At 10’s of billion tags a year

RFID Tag Components

z Antenna z Mixed-Mode IC z Packaging

Inputs to modelProcess StepsEquipment Benchmarkz Throughputz Raw material & Utilitiesz Laborz Yield

Overhead Maintenance Depreciation

Semiconductor Processing

Process modeling

z R. Leachman, J. Plummer & N. Sato-Misawa, “Understanding fab economics” Competitive Semiconductor Manufacturing, University of California, Berkeley, 1999.

z Leachman & Hodges, “Berkeley Semiconductor Manufacturing” IEEE Transactions on Semiconductor Manufacturing, May 1996.

z J. Bloomsburg, “RFID Tag Manufacturing” MIT UROP, 2002.

z R. Wright, “Cost Resource Model Detail” Economic Model Workshop, International Sematech, 2001.

semiconductor process modeling

z Sematechdeveloped benchmark

z 250_A1_82 Process Benchmarkz 0.25 micron

z 282 Step

z 19 Mask

z 3 Metalz 2 Poly

Driving Variables

z Mask, Metal and Poly Layers

z Wafer starts: 300,000 per year

z 100,000 dies/wafer z Wafer size: 200mm

Assembly Process

z Assembly Process Steps z Thinning z Dicing z Assembly z Traditional z Flip Chip z Fluidic Self Assembly z Vibratory Assembly

z Tag Test

Flip chip

Assembly Process

z Types of Runsz Unit Machine z Line Maximized

Results

z Total RFID Cost zIC + Traditional Assembly: 4.351¢ zIC + Flip Chip Assembly: 3.311¢

z IC Cost: 1.151¢

z Antenna Cost: 11¢

z Assembly: 2.25c ~ 1.15 1¢

z At 10’s of billion tags a year

Results: Die Cost

Results: Assembly Cost Breakdown

Unit Machine

Results: Cumulative Assembly Cost

Dici ily

Li i

Dicing Flily

imi Uni i

Assembly Costs

0.00002772 0.00005651

0.02922279 0.03025432

0.000342826 0.000650941

0.032175597 0.036584412

0.00001000

0.00010000

0.00100000

0.01000000

0.10000000

1.00000000 Thinning ng Tradit onal

AssembTag Test

$ C

os t

Pe r

Die

ne Max mized

Assembly Costs

0.00002772

0.00083171

0.01857550 0.01959650

0.000342826

$0.00142614

0.024655643 0.026883013

0.00001000

0.00010000

0.00100000

0.01000000

0.10000000

1.00000000 Thinning p Chip

AssembTag Test

$ C

ost P

e r D

ie

Line Max zed t Mach ne

Worldwide Fab Capacity z Assume 1 billion

tags a day z 100,000 dies/wafer

z 20% of world silicon capacity

z Fabs 15% idle today

Outline, Part II


Conclusions

Physics and God

The LimitsThe LimitsPa

ram

eter

#2

Parameter #1

The limits of physics

God

vendors

Documents

Brian Subirana & Sanjay Sarma, MIT Subirana & Sanjay Sarma, MIT. History z 1998: DISC founded ... Huge channel inventories ... Distribution Center 10%