Transcript Fault-Tolerant Computing Dealing with Low-Level Impairments

Fault-Tolerant Computing
Dealing with
Low-Level
Impairments
Oct. 2006
Defect Avoidance and Circumvention
Slide 1
About This Presentation
This presentation has been prepared for the graduate
course ECE 257A (Fault-Tolerant Computing) by
Behrooz Parhami, Professor of Electrical and Computer
Engineering at University of California, Santa Barbara.
The material contained herein can be used freely in
classroom teaching or any other educational setting.
Unauthorized uses are prohibited. © Behrooz Parhami
Oct. 2006
Edition
Released
First
Oct. 2006
Revised
Defect Avoidance and Circumvention
Revised
Slide 2
Defect Avoidance and Circumvention
Oct. 2006
Defect Avoidance and Circumvention
Slide 3
Oct. 2006
Defect Avoidance and Circumvention
Slide 4
Multilevel Model
Component
Logic
Defective
Legend:
Legned:
Service
Result
Oct. 2006
Low-Level
Impaired
Faulty
Initial
Entry
Entry
Information
System
Ideal
Erroneous
Mid-Level
Impaired
Deviation
Malfunctioning
Remedy
Degraded
Tolerance
Failed
Defect Avoidance and Circumvention
High-Level
Impaired
Slide 5
The Manufacturing Process for an IC Part
Blank wafer
with defects
30-60 cm
Silicon
crystal
ingot
Slicer
15-30
cm
Patterned wafer
Processing:
20-30 steps
x x
x x x
x
x
x x
x x
(100s of simple or scores
of complex processors)
0.2 cm
Dicer
Die
~1 cm
Oct. 2006
Die
tester
Good
die
Microchip
or other part
Mounting
Part
tester
Usable
part
to ship
~1 cm
Defect Avoidance and Circumvention
Slide 6
Effect of Die Size on Yield
Shown are some
random defects;
there are also
bulk or clustered
defects that affect
a large region
120 dies, 109 good
26 dies, 15 good
The dramatic decrease in yield with larger dies
Die yield =def (Number of good dies) / (Total number of dies)
Die yield = Wafer yield  [1 + (Defect density  Die area) / a]–a
Die cost = (Cost of wafer) / (Total number of dies  Die yield)
= (Cost of wafer)  (Die area / Wafer area) / (Die yield)
The parameter a ranges from 3 to 4 for modern CMOS processes
Oct. 2006
Defect Avoidance and Circumvention
Slide 7
Effects of Yield on Testing and Part Reliability
Die yield =assume 50%
Out of 2,000,000 dies manufactured, 1,000,000 are defective
To achieve the goal of 100 defects per million (DPM) in parts shipped,
we must catch 999,900 of the 1,000,000 defective parts
Therefore, we need a test coverage of 99.99%
Oct. 2006
Defect Avoidance and Circumvention
Slide 8
Examples of Random Defects in ICs
Resistive open due to unfilled via
[R. Madge et al., IEEE D&T, 2003]
Oct. 2006
Defect Avoidance and Circumvention
Particle embedded
between layers
Slide 9
Defect Modeling
Extra-material defects are
modeled as circular areas
Pinhole defects are tiny
breaches in the dielectric
From: http://www.see.ed.ac.uk/research/IMNS/papers/IEE_SMT95_Yield/IEEAbstract.html
Oct. 2006
Defect Avoidance and Circumvention
Slide 10
Sensitivity of Layouts to Defects
Extra material
Latent
defect
Missing material
Killer
defect
VLSI layout must be done with defect
patterns and their impacts in mind
A balance must be struck with regard
to sensitivity to different defect types
Actual photo of a
missing-material defect
http://www.midasvision.com/v3.htm
Oct. 2006
Defect Avoidance and Circumvention
Slide 11
The Bathtub Curve
Many components fail early on because of residual or latent defects
Components may also wear out due to aging (less so for electronics)
In between the two high-mortality regions lies the useful life period
Primarily due to latent defects
Failure
rate
Infant
mortality
Useful life
(low, constant failure rate)
End-of-life
wearout
Mechanical
Electronic
l
Time
Oct. 2006
Defect Avoidance and Circumvention
Slide 12
Survival Probability of Electronic Components
Percent of parts still working
Infant mortality
No significant
wear-out
From:
Oct. 2006
Time in years
http://www.weibull.com/hotwire/issue21/hottopics21.htm
Defect Avoidance and Circumvention
Slide 13
Burn-in and
stress tests
are done in
accelerated
form
Difficult to
perform on
complex and
delicate ICs
without
damaging
good parts
Percent of parts still working
Burn-In and Stress Testing
Expensive
“ovens” are
required
From:
Oct. 2006
Time in years
http://www.weibull.com/hotwire/issue21/hottopics21.htm
Defect Avoidance and Circumvention
Slide 14
Defect Avoidance vs. Circumvention
Defect Avoidance
Defect awareness in design, particularly layout and routing
Extensive quality control during the manufacturing process
Comprehensive screening, including burn-in and stress tests
Defect Circumvention (Removal)
Built-in dynamic redundancy on the die or wafer
Identification of defective parts (visual inspection, testing, association)
Bypassing or reconfiguration via embedded switches
Defect Circumvention (Tolerance)
Built-in static redundancy on the die or wafer
Identification of defective parts (external test or self-test)
Adjustment or tuning of redundant structures
Oct. 2006
Defect Avoidance and Circumvention
Slide 15
Defect Bypassing via Reconfiguration
Works best when the system on die has regular, repetitive structure:
Memory
FPGA
Multicore chip
CMP (chip multiprocessor)
Irregular (random) logic implies greater redundancy due to replication:
Replicated structures must not be close to each other
They should not be very far either (wiring/switching overhead)
Oct. 2006
Defect Avoidance and Circumvention
Slide 16
Defects in Memory Arrays
Defect circumvention (removal)
Provide several extra (spare) rows and/or columns
Route external connections to defect-free rows and columns
Defect circumvention (tolerance)
Error-correcting code
Methods in use since the 1970s;
e.g., IBM’s defect-tolerant chip
Peripheral reconfiguration elements
Spare rows
Defective row
Defective
column
Memory
array
Spare columns
Somewhat more complex with
both spare rows and columns
(still combinational, though)
Spare rows
Spare columns
With m rows and s spares,
can model as m-out-of-(m + s)
Memory
array
Modeling with coded scheme
to be discussed at the info level
Oct. 2006
Defect Avoidance and Circumvention
Slide 17
Yield Improvement in Memory Arrays
Example of IBM’s experimental 16 Mb memory chip
Combines the use of spare rows/columns in memory arrays with ECC
Four quadrants,
each with
16 spare rows &
24 spare columns
ECC corrects
any single error
via 9 check bits
(137 data bits)
Bits assigned to
the same word
are separated
by 8 bit positions
80
ECC and
spares
60
ECC
only
40
20
Spares
only
0
Oct. 2006
Yield
100
0
Avg. number of failing cells per chip
1000
2000
Defect Avoidance and Circumvention
3000
4000
Slide 18
Defects in FPGAs
Defect circumvention (removal)
Provide several extra (spare) CLBs, I/O blocks, and connections
Route external connections to8-input
available blocks
I/O blocks
ANDs
Defect circumvention (tolerance)
Not applicable
CLB
CLB
CLB
CLB
01
Mu x
C
Q
FF
D
Q
Mu x
01
(a) Portion of PAL with storable output
Oct. 2006
Defect Avoidance and Circumvention
Configurable
logic block
Programmable
connections
(b) Generic structure of an FPGA
Slide 19
Defects in Multicore Chips or CMPs
Defect circumvention (removal)
Similar to FPGAs, except that processors are the replacement entities
Interprocessor interconnection
network is the main challenge
Will discuss the switching
and reconfiguration aspects
in more detail when we get
to the malfunction level in
our multilevel model
Oct. 2006
Defect Avoidance and Circumvention
Slide 20
Circumventing Defects in Processor Arrays
Oct. 2006
Defect Avoidance and Circumvention
Slide 21
Defect Tolerance Schemes for Linear Arrays
Bypassed
Test
I/O
P0
P1
Test
Spare or
Defective
P2
I/O
P3
A linear array with a spare processor and reconfiguration switches
P0
P1
Spare or
Defective
P2
Mux
P3
A linear array with a spare processor and embedded switching
Oct. 2006
Defect Avoidance and Circumvention
Slide 22
Defect Tolerance in 2D Arrays
Two types of reconfiguration switching for 2D arrays
Assumption: A defective unit can be bypassed in its row/column by
means of a separate switching mechanism (not shown)
Oct. 2006
Defect Avoidance and Circumvention
Slide 23
A Reconfiguration Scheme for 2D Arrays
Spare
Row
Spare Column
A 5  5 working array salvaged
from a 6  6 redundant mesh
through reconfiguration switching
Oct. 2006
Seven defective processors
in a 5  5 array and their
associated compensation paths
Defect Avoidance and Circumvention
Slide 24
Limits of Reconfigurability
No compensation path
exists for this faulty node
A set of three defective nodes, one of which cannot be
accommodated by the compensation-path method.
Extension: We can go beyond the 3-defect limit by providing
spare rows on top and bottom and spare columns on either side
Oct. 2006
Defect Avoidance and Circumvention
Slide 25