on effective tsv repair for 3d-stacked ics
Post on 23-Feb-2016
121 Views
Preview:
DESCRIPTION
TRANSCRIPT
l i a b l eh k C o m p u t i n gL a b o r a t o r y
On Effective TSV Repair for 3D-Stacked ICs
Li Jiang†, Qiang Xu† and Bill Eklow§
†CUhk REliable Computing LaboratoryDepartment of Computer Science & Engineering
The Chinese University of Hong Kong
§ Cisco, CA,US
Outline Introduction
Motivation
TSV Redundancy Architecture
Routing Heuristic for Timing Consideration
Discussion & Conclusion
3D Product and To appear
3D stacked DRAM Package PCB
TSV DRAM
I/O BufferRD/WR
CMOS Image Sensor
Interposer based 2.5D FPGA Interposer
PackageTSV
PFGA Die
Memory on Processor More TSV
More Complicated
Requires manufacturing yield to be commercially viable
The Impact of 3D Stacking on Yield
Stack Yield Loss
Misalignment Impurity Open Short Leak & Delaminating
Void & Break
Assembly Yield Loss
Leveraged by KGD test and D2W stacking
Clustered TSV Defects
Assembly Yield is dramatically affected by TSV clustered faults
Source: IMECBond pad short Unsuccessful fill
TSV Repair Schemes: Neighboring Repair
1
2
1
N
1 Spare TSVN TSV Chain
2 Spare TSVs4 Signal TSVs
Redundancy Ratio
Signal-Switching Signal-Shifting
Source: Kang, SAMSUNG Source: H.H-S.Lee, GATech
M Spare TSV rows N x N TSV grid
Crossbar
Source: Loi, U. Bologna
M
N
Motivation
Signal-ShiftingSignal-Switching
Random
faults
Clustered TSV faults
Due toSurface
Roughness, Wafer bow, alignment
error
Redundancy Ratio: 1/2
CrossbarTo overcome: Repair faulty TSV from redundancy far apart
Motivation
Repair with TSV far-apart by topology
mapping
High repairing flexibility
Router based TSV grid
Reduce the cost of spare
TSVsReduce the
router complexity
Problem: TSV redundancy architecture and repair algorithms
Router Based TSV Redundancy
Successively Signal Rerouting• TSV Grid and Signal Routing Infrastructure
Redundant TSV
Signal TSV
SwitchSignal
Repair faulty TSV with nearest good TSV, and continue until a redundant TSV is used
M+N Spare TSVsM X N TSV Grid
M+NMxN
Switch DesignDirection of Rerouting
North to south, West to east (2 direction)Bypassing signal
Allow multi-hop signal reroutingMore Complex Design for more
routabilityTSV
Signal
East
North
South
West
Initial
TSV
Signal
East
North
South
West
Lend TSV to North
TSV
Signal
East
North
South
West
Lend TSV to North, bypass signal from west
Rerouting SchemeEdge Disjoint Paths Problem Maximum Flow
Method
W
TSV
Sig
E
N
S
Repair ChannelRepair Path
Problem Formulation
• Maximum Flow Method with 1 edge capability • Find Repair Path in Flow Graph (edge disjoint)
• Timing Constraint Length Constraint • Decision making in flow graph, affecting following
solution• Transfer the problem by finding Repair Channel
• Length Bonded Maximum Flow (NP-Hard)
a
b
min
c
b-min
a-min
c-min
0
Heuristic
Diagonal Direction GroupingBounded BFS Search (Length Bound & Maximal
Hops)
i=11
2
3
45
1 2 3 4 5
i=2
1
2
3
45
1 2 3 4 5i=2 C1,3
C1,4 C2,3
C1,5C2,4
C2,4C3,3
C2,2
C2,3 C3,2
C2,4 C3,3
C3,3 C4,2
i=3
1
2
3
45
1 2 3 4 5 i=3 C1,4
C1,5 C2,4
C2,5 C3,4
C2,3
C2,4 C3,3
C2,5 C3,4 C3,4 C4,3
C3,2
C3,3
C3,4 C4,3
C4,2
C4,3 C5,2
i=4 i=5
1
2
3
45
1 2 3 4 5
Maximal Hops = 2
Experiment Setup
Shifting: 2:1Switching: 4:2Crossbar: 8:2router: 4x4:8,8x8:16
Vary TSV Number: 1000 ~ 10000 ~ 100000
Fault Injection: • Poisson Distribution varying failure rate• Compound Poisson Distribution varying cluster
effect
Timing Constraint: • Assuming equal distances between
neighboring TSVs• Length constraint: 3 – 1 times of the distance
Comparison
Experimental ResultsCompound Poisson Distribution with Fixed TSV Failure Rate as 0.5%Alpha: Clustering Effect 0.4~3
1000 TSV
0.950000000000001
0.960000000000001
0.970000000000001
0.980000000000001
0.990000000000001
1shift2:1 cross8:2 switch4:2 router4x4 router8x8
10000 TSV
0.6
0.7
0.800000000000001
0.900000000000001
1shift2:1 cross8:2 switch4:2 router4x4 router8x8
Experimental ResultsCompound Poisson Distribution with Fixed TSV Failure Rate as 0.5%Alpha: Clustering Effect 0.4~3
100000 TSV
0
0.2
0.4
0.6
0.8
1shift2:1 cross8:2 switch4:2 router4x4 router8x8
100000 TSV
0.930.940.950.960.970.980.99
1
Experimental Results
1000 TSV
0.997
0.998
0.999
14x4(3T) 4x4(2T) 4x4(1T) 8x8(3T) 8x8(2T) 8x8(1T)
Conclusion
Cost Effective and scalable Solution to effectively repair clustered TSV faults.
From the cost perspective:• Limited extra Muxes and wires
• To achieve the same TSV yield, the required redundant TSVs with the proposed repair scheme is much less than existing solutions
l i a b l eh k C o m p u t i n gL a b o r a t o r y
Thank you for your attention !
top related