Letting Process Drive Design
The RocketMEMS® Model
Alissa M. Fitzgerald | 6 October 2014
11th
anniversary
Overview
•
•
•
•
•
About us
The „Long Tail‟ of the MEMS market
Customer profile: System integrators
What‟s wrong with MEMS development
The RocketMEMS model
Page 2
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
Company background
• Founded 2003 by Alissa M.
Fitzgerald, self-funded
• Burlingame, CA: near SFO and
Silicon Valley
• Goal: to be the premier MEMS
product development firm
Headquarters in Burlingame, CA
• Consistent growth
– Over 125 clients served to date
• Active member of the MEMS
Industry Group
Fab operations at 1500m2
UCBerkeley Marvell Nanolab
Page 3
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
Full development services from concept to production
AMFitzgerald in-house
Strategic
partners
• Custom MEMS development from start to finish
– Multi-disciplinary, expert engineering team
• Design and process integration for volume production
• In-house prototype fabrication, easy transition to
production partners
Page 4
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
A Tale of Two MEMS Markets
Page 5
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
The Top 30 earn ~80% of
industry revenue
Page 6
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
The Long Tail concept of market distribution
Harry Potter
Book sales on Amazon
MEMS
Design
[Source: C. Anderson, Wired, October 2004.]
Page 7
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
The MEMS Industry has a Long Tail
500M-
Units/ yr.
100M-
Everybody
else
The Top
30
10M MEMS for
consumer,
automotive
Applications
[Source: C. Anderson, Wired, October 2004.]
Page 8
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
Some applications in the MEMS Long Tail
Units
per year
Diagnostic chips
Low volume
1M
Surgical tools
High profit margin
Cell phone towers
Little price erosion
Robots
100K
Industrial printing
Scientific instruments
Factory monitoring
Commercial aircraft
10K
Energy exploration
Server farms
Military aircraft
1K
Spacecraft
Page 9
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
Long Tail company profile
• A well-established system integrator or OEM
– Has a mature product line
– Deeply understands their market
– Examples: Schlumberger, Medtronic
• Sees MEMS as an opportunity to make new products and
gain a competitive advantage
– New sensing capabilities
– Lower existing product cost to expand market
• Sees market opportunity, needs < 2 years development
to enter market at right time
Page 10
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
Long Tail company profile - continued
• No experience with MEMS chip design or integration
– No confidence nor desire to hire a MEMS team or to develop
MEMS capabilities (yet)
• Views MEMS as a component in their system
– Developing new patents not that important
– IP is in package and system integration
• Would prefer to buy finished MEMS chips to minimize
overall product development risk
• Does not understand why MEMS development takes so
long and is so risky!
Page 11
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
How to serve the Long Tail market
• Serving the Long Tail requires selling many different
items in small unit volumes
• Amazon is the master of the Long Tail market
– Super-efficient operations model
• Current barriers in MEMS:
– Long development times (usually 5+ years)
– Hard to recover cost of custom development when unit
volumes are so low
Page 12
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
The challenge of MEMS development
Lack of
process
rules
Desire
for
patents
Limited
physical
models
Designer forces
new process to
foundry
“One product, one process”
Jean-Christophe Eloy, Yole Développment
Page 13
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
Current MEMS fabless model → design-fab-test cycles
Digital ASIC Fabless Model
Reference
designs,
IP blocks
Verified
physics
models
New
Design
Foundryspecific
design rules
Process
modeling
Design
simulation
18 months, 95+% yield
Rule
checking,
Verification
Product
MEMS Fabless Model
Reference
designs,
IP blocks
Verified
physics
models
Foundryspecific
design rules
New
Design
© AMFitzgerald
Semicon
Europa2014
2014 - Grenoble
Design
simulation
5+ years, ?% yield
GOOD
Page 14
Process
modeling
FAIR
POOR
Rule
checking,
Verification
Product
Process development consumes a lot of time
Process Development
Redesign
Redesign
Design
Example
MEMS
Custom
Development
3
months
Process
Development:
Attempt 2
Process
Development:
Attempt 1
3
1
3
Redesign
Process
Development:
Attempt 3
1
3
Attempt N
1
3
>~18 months
Page 15
Semicon Europa 2014 - Grenoble
First working
samples:
© AMFitzgerald 2014
It‟s a different story for the Top 30 MEMS companies
• They own their fabs and have characterized their
processes
– Mature process libraries, design rules
– Customized EDA tools, IP blocks (“cells”)
Captive Fab MEMS Development
Reference
designs,
IP blocks
New
Design
Verified
physics
models
Foundryspecific
design rules
Process
modeling
Design
simulation
2+ years, >75% yield
GOOD
FAIR
Product
POOR
Page 16
Semicon Europa 2014 - Grenoble
Rule
checking,
Verification
© AMFitzgerald 2014
Letting process drive MEMS design
*But customization
is desired!
Process
and
design
rules
No need
for
novelty*
Tuned,
verified
models
Foundry process
library sets
boundaries for
designs
One process library,
many products…rapidly
Page 17
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
Why can‟t MEMS copy the CMOS fabless model?
CMOS
MEMS
• Process flow fixes Z axis
• 2D design of circuits →
many chip types
• Needs 3D design freedom
• Process flow fixes Z axis →
few sensor types
• Many ASIC designers
• Few MEMS designers
• CMOS foundry strongly
motivated to create standard
process flows
• TSMC, GF, etc.
• MEMS foundry has little
motivation to create standard
process flows
• Contract mfg. model
Page 18
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
What about MEMS “Platforms”?
• Mature process flows
opened to public
• Customer must provide the
chip design
– Most Long Tail companies
cannot provide a design!
Source: CMP
Some MEMS platforms:
Facility
Service Name
Target Market
Process
Wafer size
ST
Thelma
Motion sensors
Thick epi-poly
200
Dalsa
MIDIS
Motion sensors
SOI with vacuum
and pressure cavity
200
X-FAB
XMB-10
Motion sensors
Cavity SOI
150, 200
NF Shuttle
Motion sensors
CMOS cap, SOI
wafer
200
InvenSense
Page 19
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
Platforms are not effective without MEMS Designers
• There are very few of us!
– MEMS is introduced very late in engineering education
– Graduate-level only, usually Ph.D.
• Lack of full-featured design tools means experts must do
design
Long Tail
Companies
MEMS
Designers
MEMS
Foundries
Page 20
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
RocketMEMS: A Model for Designer-Foundry Cooperation
• Designer and foundry define design/process together
• Use stable, characterized process modules
– Silex SmartBlocksTM library
• Collaboratively assemble process flow from modules
• Create reference designs relevant to market needs
– Many, but not all, sensor needs may be met
• Ownership:
– Designer owns reference designs
– Foundry owns process library
– Customer gets silicon – fast!
Page 21
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
RocketMEMS® vs. Custom Development
Process Development
Redesign
Redesign
Design
Example
MEMS
Custom
Development
Process
Development:
Attempt 2
Process
Development:
Attempt 1
3
months
3
1
3
Redesign
Process
Development:
Attempt 3
1
3
Attempt N
1
First working
samples:
3
>~18 months
RocketMEMS: Skips Process Development
Start with
reference
designs and
process
Tailor
Design
RocketMEMS
Development
1-2
mo.
Fabrication
3
Prototyping
3Low Volume
Production
First working
samples:
4-5 months
Pilot Production
Page 22
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
First RocketMEMS sensor type: Pressure
What can be customized?
• Pressure range
• Sensitivity
• Die size & thickness
• Bond pad location & size
• Bridge resistance
• Full vs. Half bridge
Examples of various
RocketMEMS
pressure sensors
(with test structures)
Varying membrane
sizes
Page 23
Semicon Europa 2014 - Grenoble
Varying resistor
sizes
Varying die sizes
© AMFitzgerald 2014
Economical production that scales
• Multi-Project Wafer (MPW) format
for cost saving
– Multiple customers served by
AMFitzgerald
• Production options at different
volumes
– Prototype
 500+ die
– Low volume production
 10,000+ die
– High volume dedicated runs at Silex
 100,000+ die
Multi-Project Wafer: Many different chip
designs are processed on the same wafer
Page 24
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
RocketMEMS: Our philosophy
“Design for Verified Process”
• No process development
• Use proven reference designs and
tailor them to customer
specification
One
Product,
• Our long-term vision for
RocketMEMS:
One
Process
– A menu of sensor types and foundries
to suit a wide range of customer
needs
Page 25
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014
Summary
• Two MEMS industries: the Head and the Long Tail
– Good business opportunities exist in the Long Tail – if one
can serve it efficiently
• New models are needed to speed up MEMS development
and serve the Long Tail
– RocketMEMS is a new model – evolved from the ASIC design
house model
• The RocketMEMS model can benefit the entire industry
– Helps to grow the MEMS Fabless model and ecosystem
– Connect OEMs/System Integrators to rapid MEMS solutions
– IoT, etc.
Page 26
Semicon Europa 2014 - Grenoble
© AMFitzgerald 2014