Upload
lamtu
View
229
Download
5
Embed Size (px)
Citation preview
Silicon, Interconnect, Packaging and Test Challenges from a Foundry Viewpoint
SWTC June 2015
Keynote Perspective
Design Test Dev
(DFT) Reticle
Sets Wafer Fab
& FSI Bump Probe
Thin & BSI
Ass’y & Test
OSAT Partners
Process development and characterization
For 3D and 2.5D, μbump is owned by the OSAT
Design Partners
Design flow, rules
and tool alignment
Memory Partners
Integration and
characterization
EDA & IP Partners
Development
and validation
2
Customer Design
Bump and Probe
Assembly, Package Test, Board Mount
Agenda
Trends in Silicon
Packaging Drivers and Solutions
Materials and Design: Challenges and Learning
1
2
3
Test Challenges from Foundry Viewpoint 4
Supply Chain Solutions 5
GLOBALFOUNDRIES 4
Technology Node
Tra
nsis
tor
Cost
Cost of Scaling per Transistor is no Longer Decreasing
FinFET
Cost of Scaling is Rapidly Increasing
• Development cost for 10nm design will approach $400M
5
Cost of Scaling is Too High
• Requires multi-$B lifetime revenue to be economically feasible per design
• Forces foundries to push customers into “good enough” solutions to drive volume
• Must find other solutions
6
Trends in Silicon
Packaging Solutions
Material and Design: Challenges and Learning
1
2
3
Test Challenges 4
Supply Chain Solutions 5
Solutions have to balance performance, power, and cost
GLOBALFOUNDRIES Confidential 8
Silicon
• BEOL stacks
• Low K modulus
• Thinness (warpage)
Thermal
• Heatsink
• TIM
Packaging Challenges
10/01/2014 GLOBALFOUNDRIES Confidential
TSV Process Challenges
• TSV integration
• Device KOZ
• Pumping
Substrate
• Core CTE
• Warpage
• Routability
Assembly Challenges
• Die attach and die stacking
• Underfill flow vs gap and pitch
• Warpage management
Bump
• Pitch
• Cu Pillar induced stresses
• EM challenges
Wafer Test
• Contact technology
• Pitch / Parallelism
• Probe card cost
• Supply chain
Bump Pitch (µm) Bandwidth Drive I/O Density
0
20
40
60
80
100
120
140
160
2012 2013 2014 2015
Bump
µBump
GLOBALFOUNDRIES Confidential 10
Bum
p P
itch (
µm
)
TSV
But Single Die Bump Pitch Isn’t Enough Applying Packaging to Solve Scaling Limits
• SoC Partitioning
– Cost optimization: Logic-Logic side by size or stacked
• Reduced die size
• Increased yield
– Functional optimization: Best fit node per function
• Design reuse without redesign
• Lower risk – lower cost
• Logic-Memory Power & Performance
– Logic & memory integration:
• Increased bandwidth
• Higher performance per watt
• Logic-Logic for Performance
11
SoC Die
1
Die
2
Jan 19th, 2015 GLOBALFOUNDRIES Confidential
Interconnect Density: The Enabler
• Edge Routing provides challenges (side x side solutions (2.5D, WLFO) routing density:
– 0.5µm (silicon interposer) - 5µm (WLFO)
– Small vias needed to enable multi-layer routing
– Shielding
– 1 edge of 10mm die at 2µm L/S, 1/3 of lines used for shielding: 1675 interconnect
– 1 sided routing challenges on design
• Stacked (3D) solutions:
– Area based: 5um vias, 2um keep-outs, 1/3 of TSV used for shielding, limit TSV to <1% of die area
GLOBALFOUNDRIES Confidential 12
HBM Memory: Routing for 55um Bump Pitch
13 Glass will be technology/supply chain dependent
HBM Routing Capable
Routing Density Solutions
GLOBALFOUNDRIES Confidential 14
Package Size
IO D
ensity (
traces/m
m o
f ed
ge
)
POP
High Density Laminate
Silicon Packaging (2.5D)
wlCSP (FI)
Incre
asin
g P
erf
orm
ance
HD WLFO
Silicon Packaging (3D)
Low Density Laminate WLFO
Power Envelope is Limited and Costly to Manage
15
June 9, 2015
• Mobile: Maximum skin temperature in smart phones requires shutdown of cores
• Computing: Power required to drive off substrate memory reduces graphics or processor performance
• Networking: > 50% of data center cost is cooling
• Memory: Tj > 80C increases the required DRAM refresh rate, increasing power and heat
• SerDes greatly reduces I/O density required between processor and memory but generates significant heat
• Solution 1: Massively parallel I/O solves the heat problem but requires new technologies with very dense routing capability on substrate
Solution 2: Technological Innovation Power Management 2.5D Roadmap
16
pJ/bit
Memory
Memory
Memory
SoC
Memory
Memory
Memory
Solution 3: Are We Over-Designing?
• Architecture decisions drive interconnect density
• Decisions are often made without understanding impact on cost (and number of suppliers with capability to meet)
• Affects cost at multiple levels: interconnect, wafer test, substrate, assembly
• We need to drive cost impacts prior to design decisions Interconnect, wafer test, substrate, assembly
17
12/12µm L/S 2/2µm L/S
Trends in Silicon
Packaging Drivers and Solutions
Material and Design: Challenges and Learning
1
2
3
Test Challenges 4
Supply Chain Solutions 5
Chip – Package - Interaction
• Polyimide and BEOL layers on one side warp wafers
• There is a decreasing amount of solder to join each bump (or pillar) as pitch shrinks – ie more sensitive to warpage (O/S)
• Substrates, with higher bulk CTE than silicon, dominate warpage from reflow cool down
• Resulting stress is exerted on bump and on BEOL stack, which can break
• Detect separation by ultrasonic imaging
•
GLOBALFOUNDRIES Confidential 19
Silicon
Bump
or
Pillar
BEOL
Stress
Transmitter
Substrate
Driving CPI at Leading Edge Nodes
S.Kengeri, Semicon SGP 2014
Driven by multi-core processors and increase in I/O
Reduce electrical setbacks inherent with shrinking Cu dimensions
Pb Free CuPillar
Si to laminate CTE mismatch
Organic Substrate
Size
Higher Modulus
Bump
Dielectric Constant
(ILD)
Die Size
Greater stress
More fragile: cohesive strength modulus adhesion
Greater stress Greater stress
Lower-K ILDs vs. Modulus vs. Hardness
B.Chandran, intel Future Fab 2004
Bump Interconnect “Nodes” GLOBALFOUNDRIES Bump/Probe Shipments
Ja
n/
08
Mrz
/ 0
8
Ma
i/ 0
8
Ju
l/ 0
8
Sep
/ 08
Nov/ 0
8
Ja
n/
09
Mrz
/ 0
9
Ma
i/ 0
9
Ju
l/ 0
9
Sep
/ 09
Nov/ 0
9
Ja
n/
10
Mrz
/ 1
0
Ma
i/ 1
0
Ju
l/ 1
0
Sep
/ 10
Nov/ 1
0
Ja
n/
11
Mrz
/ 1
1
Ma
i/ 1
1
Ju
l/ 1
1
Sep
/ 11
Nov/ 1
1
Ja
n/
12
Mrz
/ 1
2
Ma
i/ 1
2
Ju
l/ 1
2
Sep
/ 12
Nov/ 1
2
Ja
n/
13
Mrz
/ 1
3
Ma
i/ 1
3
Ju
l/ 1
3
Sep
/ 13
Nov/ 1
3
Ja
n/
14
Mrz
/ 1
4
Ma
i/ 1
4
Ju
l/ 1
4
Sep
/ 14
Nov/ 1
4
SnAg.Bump
Pb.Bump
CuP
GLOBALFOUNDRIES Confidential 22
2011 2012 2013 2014 2015 2016 2017 2018 2019
28nm 10nm 14nm Platform
Technology
SnAg ≥ 180µm Pitch
Cu Pillar
µBump
finFET
≥ 130µm Pitch
≥ 100µm Pitch
≥ 80µm Pitch
2.5D/3D ≥ 40µm Pitch
TSV Interposer TSV Diameter/Thickness: 10µm/100µm
3D TSV 6µm Diam/55µm Deep TSV + M3
GLOBALFOUNDRIES Interconnect Roadmap
23
Interaction Between Design and BEOL Stack
• Weaker dielectric
• Cost driver to eliminate layers
• Tougher mechanical reliability requirements
High lead Solder Lead Free Solder
28nm
D.Breuer et al. MAM2014
100um bump pitch
64nm silicon routing pitch
Almost 3 orders of magnitude
GLOBALFOUNDRIES Confidential 25 23.12.2014
White Bump – Localized Crack Under Bump
CSAM example of white bump
FIB SEM
BABSI Bump Assisted BEOL Stability Indentation
GLOBALFOUNDRIES Confidential 26
[1] H. Geisler, M. Lehr, F. Kuechenmeister, M. Grillberger, German Patent DE 10 2010 002 453 A1 (September 01, 2011).
[2] H. Geisler, E. Schuchardt, M. Brueckner, P. Hofmann, K. Machani, F. Kuechenmeister, D. Breuer, H. Engelmann “
Experimental Analyses of the Mechanical Reliability of Advanced BEOL/fBEOL Stacks Regarding CPI Loading” IRPS 2013
Improve Robustness Die Edge Design
• Crack stop with interconnected metal walls; wider = stronger
• Metal fill with vias in corners to increase robustness
• Complex metal “walls” GLOBALFOUNDRIES Confidential 27
TSV Process Flow and Challenges
Post TSV Etch
Post resist strip
TSV clean
Post TSV Liner
TEOS 03
STI
Post TSV
Barrier Seed
Post TSV ECP
& Anneal
Silicon
STI
CAP
MOL FEOL /
Stack
Post TSV CMP
Land on Cap
Dense array
• TCD / Depth /
BCD uniformity
•Liner step coverage •Barrier Seed
continuity
•Cu step height
•Cap uniformity
•Void free gap fill
•Overburden
28
Via Chains and Pop-up Structures Electrical Test Summary
• Very low impact due on via chains with KOZ at 1μm and 2μm
• TSV pop-up structures show negligible impact of TSV and meet the target specs
29 No degradation as move from 5um to 1um KOZ
Trends in Silicon
Packaging Drivers and Solutions
Materials and Design: Challenges and Learning
1
2
3
Test Challenges 4
Supply Chain Solutions 5
Foundry Test Business
• The Foundry does wafer test to provide fast feedback on yield to the fab
– Critical for devices in leading edge fab processes
– Particularly FA drives probe at the Foundry
– Bump is required to get to wafer test
• Cost drives location of bump and probe for mature devices and nodes
• Thus bump and probe are generally at the foundry for leading edge and generally at the OSAT for HVM and trailing edge
• Customers generally provide probe cards because for the Foundry, the customer controls all input probe card costs (I/O, parallelism, product ramp, product lifetime)
• The Foundry (mostly) does not do final package test
GLOBALFOUNDRIES Confidential 31
Bump Pitch (µm)
0
20
40
60
80
100
120
140
160
2012 2013 2014 2015
Bump
µBump
GLOBALFOUNDRIES Confidential 32
Bum
p P
itch (
µm
)
TSV
Bump Cost Trend
2012 2013 2014 2015
GLOBALFOUNDRIES Confidential 33
But Wafer Test Cost Composition is Changing
Dep
PCs
Util + Fac
Maint
I/L
D/L
Other
Today
GLOBALFOUNDRIES Confidential 34
Dep
PCs
Util + Fac
Maint
I/L
D/L
Other
1 yr Ago
Probe Card Cost
• Cost is increasing with
– Decreasing pitch
– Increasing probe count
– Increasing parallelism
• Which also drives tester cost
• Design impacts
– Cost of test is determined largely in device design
GLOBALFOUNDRIES Confidential 35
Current cost model ($10/probe) is not scalable with IO quantity and pitch
x8 Total Cost
Head Only
$K
per
Card
Increasing # pins
Probe Card Lifetime
• Pitch reduction drives PC lifetime reduction
– Pin diameter reduction
– Shortening usable tip length
GLOBALFOUNDRIES Confidential 36
0
0.2
0.4
0.6
0.8
1
1.2
1.4
0
100
200
300
400
500
600
700
800
900
4.0 2.5 2.0
Usable TipLength (µm)
Life time (Mtouchdowns)
Probe pin diameter (mils)
Useable
tip
length
(µ
m)
Life
Tim
e (m
illion to
uchdow
ns)
Typical Product Ramps
1st Quarter 1st Year 2nd Year 3rd Year
Mobile
Graphics or PCProcessor
Automotive
GLOBALFOUNDRIES Confidential 37
Rela
tive V
olu
me
Net Probe Card Cost per Wafer Shipped
160um 110um 80um
GLOBALFOUNDRIES Confidential 38
$ p
er
wafe
r ship
ped
Bump Pitch
Not acceptable Forcing a re-think on technology
Wafer Level (wlCSP) Test
• Large Turn-key Business for Foundries
• Business model is selling net-good-die
– Close OSAT partner relationships
• Challenge is processing after test: no electrical failure detection
– Capability and process control at blade and laser saw, pick, inspection, and in tape design critical and tends to drift out of control
– Saw more difficult on wlCSP because of high stress in thick one-sided RDL layers
• WLFO enables molding a small ring around die. Issues:
– Cost: no satisfactory solution
– Need either cheap safe die bare die handling or
– Installed capacity
GLOBALFOUNDRIES Confidential 39
WL Fan-In
wlCSP Crack Risk from Saw Process
Step 1 Step 2A Step 2B
Note: Drawing is not to scale.
July 23, 2013
Confi
denti
al 40
Street width makes a net good die difference at these small die sizes Laser energy decreases over time Detection difficulty post test Continuous attention on process control essential
Wafer Probe Roadmap Challenges
• Probe card cost , particularly <100µm
• Probe technology < 80µm
– Probe on Microbump?
√ Sacrificial pad?
• Utilize DfT with wrappers to reduce the number of I/O that need to be contacted?
– Ability to re-use tester assets
– Ability to use less dense and cheaper cards
– But difficult for the Foundry to implement
• wlCSP testing methodology
GLOBALFOUNDRIES Confidential 41
Trends in Silicon
Packaging Drivers and Solutions
Materials and Design: Challenges and Learning
1
2
3
Test Challenges 4
Supply Chain Solutions 5
Deep Partnerships A Collaborative Supply Chain
Customer benefits
– Open, flexible, cost-effective supply chain
– Aligns with customer’s preferred partners
– Active co-development to ensure smooth volume ramp
– Suppliers don’t duplicate investment and thus cost is lower
Design Test Dev
(DFT) Reticle
Sets Wafer Fab
& FSI Bump Probe
Thin & BSI
Ass’y & Test
OSAT Partners
Process development and characterization
For 3D and 2.5D, μbump is owned by the OSAT
Design Partners
Design flow, rules
and tool alignment
Memory Partners
Integration and
characterization
EDA & IP Partners
Development
and validation
43
Business Model Challenges for the Foundry
• Don’t invest in what an OSAT can do better/cheaper
• Do invest in leading edge differentiating technology
• For multi-chip techologies:
– Will competitors ship die to a competitor for multi-chip product assembly?
– Who owns the cost of failed IC’s on multi-chip products?
• Close collaboration on development
• Priority as a ‘non-customer’
GLOBALFOUNDRIES Confidential 44
Collaboration Drives GLOBALFOUNDRIES Packaging
Validated Interposer reference flow
DFT enabled interposer flows
60K/month Bump & Probe Capacity
>15 years experience providing turnkey solutions
2.5D and 3D TSV enabled at Leading Edge
2.5D HBM integration demonstrator
Seamless supply chain model in setup
CPI Qualified at 28nm with low cost FC solutions at two major OSATs
14LPP 2D, 2.5D & 3D Package qualifications at multiple OSATs
2.5D package qualified with GF interposer
3D TSV package qualified
Seamless supply chain models in setup for 2.5D products
Design
Capabilities • Co-Design
Reference Flows
• DFx support and
tools
Internal
Capabilities • Bump & probe
• Turnkey solutions
• Investment in
emerging
technologies
Memory
Partnerships • Logic-Memory
Demonstrators
• Innovative
business models
OSAT
Partnerships • JDAs at leading
edge nodes
• Innovative
business models
Summary
GLOBALFOUNDRIES Confidential 46
Summary
• End of scaling at always lower cost
• Alternatives driver higher I/O density and new packaging solutions
• The new solutions drive higher cost test and downstream costs (substrate)
• Advancing silicon technology presents material and stress challenges that must be characterized.
• Must have significantly cheaper probe card technology and questioning of cost of test at design
• Need high density substrate solutions at lower cost
• The collaborative supply chain enables accessing best expertise at lowest cost
Thank you to: Jens Kober – Probe Dev and Production John Carulli – Test Development TM Mak – DfT and IP 2.5D/3D Jens Paul – CPI Reliability Rama Alapati – Product Mgt for Packaging
Trademark Attribution
GLOBALFOUNDRIES®, the GLOBALFOUNDRIES logo and combinations thereof, and GLOBALFOUNDRIES’ other trademarks and service marks are owned by GLOBALFOUNDRIES Inc. in the United States and/or other jurisdictions. All other brand names, product names, or trademarks belong to their respective owners and are used herein solely to identify the products and/or services offered by those trademark owners.
© 2014 GLOBALFOUNDRIES Inc. All rights reserved.
Thank you