the continuing growth of silicon integration technologies and...
TRANSCRIPT
The Continuing Growth of Silicon
Integration Technologies and
Application / System Level Testing
Prompts a Reflection on the EDA to
HVM Highway
Zoë Conroy
Senior Manager – DFT, Test & PE
Component Quality & Technology
Sept 21, 2017
Today’s Reflection
• Electronics market is demanding high growth,
high function, yet cost-effective solutions
• Multi die heterogeneous integrations offer More
than Moore
• Applications are requiring minimal defect rates
• Are today’s development tools enough to sustain
and improve required quality and reliability?
• How will EDA and manufacturing test reshape
themselves to meet these needs?
1
Continuous Innovation: More than Moore
Networking - Graphics - Computing
• 2.5/3D/SiP/WLP
• Optics, photonics
• MEMs, Sensors
• RF, analog
• Power
2
Where is the Industry at Today?
• Applications demand continuous uptime, with
cloud/data-centers allowing more applications
worldwide
• Automotive arena - <<100 DPPM
• More companies adding system level test to get
highest level of test coverage and quality
• Gate counts/ heterogeneous integration and
advanced packaging continue to push boundary
on technologies requiring ‘perfect’, yet cost
effective silicon
3
The Impact of Bad Quality
4
The Impact of the Network
• Our lives NOW DEPEND on the network
• Highest Quality is a MUST, not a nice to have!
5
Hardware is Here to Stay!
Catching Defects - What Comes First?
Fab / Process
DesignWafer / Package
Test
Board / System/
Application Test
Defect Free?
DFT to detect all defects?
Test Pattern coverage to
catch ALL defects?
Application testing to catch
remainder of defects?
Where do Problems Occur?
Fab / Process
DesignWafer / Package
Test
Board / System/
Application Test
• Silicon technology Temp inversion affect
• Memory test gaps• Structural Test gaps
• Power simulation vs. silicon
• T/V sensors do not reflect reality
• Hard to close timing
• Wide power variation chip to chip
• No DFD• Interoperability
margin
• Varying Defects• Process center
movement
EDA to HVM Highway
9
SystemTraffic
test
Pack
Out
Burn
- inSMT ICT
3D X-Ray
Functional Diagnostic
testParametric
Board Build System Integration
Component
Supplier
Equivalent SLT
ComponentDesign &
verification
WaferMfg & Test
Package/Heterogeneous
/stack test
SLT
End To End Manufacturing Flow
ATE to System Differences
• Functional application test
• Complex board / system design
• Multiple components interfacing to
each other
• Noisy environment
• Varying temperatures
• High coverage structural test program
• Low noise DUT-board design
• Low noise floor
• Very accurate power supplies
• Stable temperature
10
Typical networking board: ASICs, uP, FPGA, PHYs,
CDRs, memories (DRAM, SRAM, TCAM, HDD, etc)
Different types of coverage
tests, environments,
11
IP Guardian 10GE
Mac
(XM)
Port Logic
(PL)
Security
(SC)
10GE
Mac
(XM)
Port Logic
(PL)
Security
(SC)
Ingress Packet
Buffer
(IB)
Decision Engine Ingress
(DI)
Filter Block
(FB)
Egress Packet
Buffer
(EB)
Virtual
Output
Queues
(VQ)
Reorder
(RO)
Arbiter
Manager
(AM)
Decision
Engine
Egress
(DE)
Fabric Receive
(FR)
Chico RX
(CR)
Fabric Transmit
(FT)
Chico TX
(CT)Arbiter
Exhange
(AX)
“Packet Storage”
Port Logic
Buffer
1 pkt every 16.79 clk
1 pkt every 16.79 clk
1 pkt every 16.79 clk
1 pkt every
8.39 clk
1 pkt every
8 clk
1 pkt every 4 clk
1 req
every 4 clk
1 pkt every 4 clk (FC)
1 req
every 4 clk
1 gid
every 2 clk
1 gid
every 4 clk
1 pkt every 4 clk
1 pkt every 8 clk (Eth)
1 pkt every 4 clk (FC)
1 pkt every 8 clk (Eth)
1 pkt every 16 clk
1 pkt every 4 clk (FC)1 pkt every 8 clk (Eth)
1 pkt every 16 clk
1 pkt every
8 clk
1 pkt every 16 clk
1 pkt every 16 clk
1 pkt every 16 clk
1 page rls every
8 clk
1 credit
every 2 clk
1 pkt every 16.79 clk
1 pkt every 16.79 clk
1 pkt desc. every 4 clk
Decision Engine
Lookup
(DL)
1 cred rls every 4 clk
System Traffic Test
Ext M
em
BIS
T
Internal
Memory
BIST
IOBIST
TCAM
BIST
JTAG
1149.1
CPU
2JTAG
JTAG
2CPU
Em
bed
de
d In
stru
me
nt
Logic BIST
SerDes PRBS &
1149.6
Scan: SAF, TF,
PD, Ram Seq
Ring
Osc
SerDes PRBS &
1149.6
PLL check
ATE Structural Test
Internal
Memory
BIST
Consequences of ATE and System
Test Differences
• Board functional tests run different data from
ATE BIST and structural testing
• Environmental conditions at system level can
cause fails on marginal devices
• Components marginal for timing, voltage
• ATE test escapes
– Hard to emulate functional environment with a
standalone chip
– Less than optimal structural/BIST coverage
12
*Component Supplier
ATE Test Escapes
system diags
board h/w or mfg
IC design
other
13
Diagnosis of ASIC Fails at System Test
Data from various DPPM Reduction Initiatives 2005-2017
*ATE test escapes
Components that
failed system test for
functional or IC BIST
test, and subsequently
found to have a test
gap at IC ATE test.
• Added coverage to existing test
• Added a new ATE test
• Ported a functional test from system to ATE
• Changed limits on existing Test
ATE Gaps - Breakdown by Test
14
Func
Memory
At-Speed Scan Other
Logic BIST
High speed serial data
Voltage
• ATE tests added
Heterogeneous Functions Drive Growth
15Based on ”IC Insights Report”, Sept 2014, 3DIn-cite.com
Increasing Levels of Integration
interoperability
component to component
(sub-)component
Ext mem
Ext mem
Ext mem
Using Components to Test the Board
• SerDes
• IOBIST
• Internal MBIST
• External MBIST
• Programmable BIST
for application
specific algorithms
• T/V Sense
• PLL Checkers
• Traffic
17
NPUA6
NPUA1
NPUA2
NPUA5
NPUA4
NPUA3
ASICB
ASICB
Ext mem
Ext mem
Ext mem
CPU
FPGA
Ext mem
Ext mem
PCIeswitch
QSFP QSFP QSFP
Ext mem
Ext mem
Ext mem
Ext mem
Gearbox
Gearbox
Gearbox
Gearbox
Gearbox
Gearbox
FabricASIC
FabricASIC
NPU A1-6
NPU A1-6
Reduced Structural Test Point Impact
• Test points reducing with denser routing
• Components BIST becomes critical for board testing*
• BIST can cover:- Board opens/shorts- Trace transitions, - High speed interconnect and SI
*iNEMI BIST Program http://community.inemi.org/bist
EDA Tool Portfolio Opportunities
• SLT addition just keeps the status quo!
• Component applications’ testing manifests
unique coverage gaps in ATE tests
• Denser silicon and heterogeneous integration
requires new types of defect modeling and ways
to isolate defect locations in real time
• Embedded and external memories continue to
be a challenge, with high application fall-out.
19
Connecting Suppliers & EMS Partners
Big Data Analytics CloudMachine to
Machine
End to End CollaborationAutomation
IoE and Adaptive Test to automate End to End Collaboration
ITC 2015, Bert Gab, End to End Adaptive Test, “From system to component and back”, invited paper
Test Data Analytics and Adaptive Test
EMS PartnerComponent Supplier
Big Data Analytics/Machine Learning to predict probability of part failing inSystem Test.
Improved Process/Optimized Test Data Templates
Adaptive Testing:Test only relevant portion of the population to catch 100% failures
System TestComponent Test
Faster Time to Quality
Faster Component FA
24/7Data Loop
Fab
Cloud Based Predictive Analytics
22
SystemTraffic
testICT
Functional Diagnostic
testParametric
Board BuildComponent Supplier
Componentdesign &
verification
WaferMfg & Test
Package/heterogeneous
/stack testSLT
Fab process data
Test
dat
a
componenttest dataData
gateway
Cloud servers
Big dataprediction analytics
Process and test data
Boardtest data
PredictionTest data
test area 2test area 1 test area 3 test area 4Wafer test Pkg test SLTtest
Adaptivetest engine
Feed forward
Unique Die id traceability
EMS PartnerComponent Suppliers
Closing Remarks
• There are gaps in ATE tests - structural tests do not
exercise chips in the same way the application does.
• SLT provides an application test to weed out final defects.
• Actionable data at SLT required to keep closing gap from
system to ATE.
• EDA must keep up with increasing complexities of
designs, integrations, quality requirements.
• Cooperation and bi-directional info exchange across the
entire supply chain is a must.
• Standards for data exchange formats and hand-off
criteria between supply chain partners need to be agreed.