Transcript presentation

optimiSE
from data to knowledge
optimiSE Gmbh
München – Bruchsal
Germany
Werner-von-Siemens-Straße 47a
D-76646 Bruchsal
Telefon +49-7251- 930 667/668
Fax
+49-7251- 930 666
Email: [email protected]
Internet: www.optimiSE.de
SETNET, London, 13 September 2001
The Cost of Test
1
What reduTec accomplishes…
optimiSE
from data to knowledge
reduTec
or how to cut down on your test
cost budget
 reduTec analyses test data of complex electronic
products with respect to underlying redundancies
 reduTec regroups tests into “necessary ones” and
those without information gain
Benefits
 Controlled dismissal of redundant tests of intact devices
for high yield product lines
 Earliest possible detection of damages to shorten test
cycles of defect devices for low yield products
SETNET, London, 13 September 2001
The Cost of Test
2
reduTec optimises
optimiSE
your test management
from data to knowledge
Exampel: Chip-Testing
 Input
 real-valued (acceptance intervals) or binary (pass /
fail) intact / defect data from frontend / backend
tests
raw wafer
wafer fabrication
processed wafer
 optional: extra process data, e.g. chip location
frontend
 Methods
 deterministic, product-independent mathematical
wafer probe
procedure without arbitrary statistical assumptions
assembled chip
 neuro-fuzzy-system for test data mining and
process monitoring
sawing & assembly
backend
tested device
final test
 Output
 identification of redundancies for high yield
products
 optimised test sequences w.r.t. first failure out for
low yield product lines
time
 yield learning and process control
SETNET, London, 13 September 2001
The Cost of Test
3
optimiSE
from data to knowledge
What means redundant test
in the context of reduTec ?
A test Tk is redundant with respect to a given series of tests {Tj} jk,j=1,...,N, if
the information gain obeys I k N    for some user-defined  > 0. Here
I k N  : W N   Vk
where W N  and V k represent the information vectors corresponding to the
test subsets {Tj}j=1,...,N and {Tj}jk,j=1,...,N respectively. In most cases, the
suitable metric is the Euclidean distance.
Applies to intact devices, where
 tests resulting in large deviations from the ideal value and/or large variations around
the actual mean value of the test series carry information about the control of the
production process – they are “essential”
 tests which are “well-passed” by all devices are superfluous and potentially redundant
and defect objects, where
 binary tests are ordered with respect to “first stop on failure” and probes without information
gain can be neglected
SETNET, London, 13 September 2001
The Cost of Test
4
Test ranking by reduTec –
intact objects
optimiSE
from data to knowledge
reduTec assigns quantitative, weighted redundancy factors to each test :
300
615,2
250
weighted
information gain
200
obsolete tests
(for instance)
150
100
50
0
) )
0
I D 8 1 --V IG r:0 I G S1 S1 3A EX I A IL 4:0 :3 :0 :0 :0 EX :0 :0 A2 :0 S :0 S :0 C V) :0 FP :1 1) :0 :3 x: p
E (M4 (F1 32 -V ste _V_ TP PV 2S O_ p_T _TO p_ AN G2 CB OL T_ CG _OL IS OL t ES OL t CP OL _AC SA N1 _E C1 ) TC p_ 1 p_ 2 ma la m
R
A
t
F
(
_
c
_
_
A
C
O
D
V
)
n
Y
(
M
)
O
S
S
6
6
1
u
u
p
p
T ( o o / I_ V
U HD H L N a n
p p _ TC In n _I _ DI 4 1 IV 5 1 sa 0 sa 2 sa 5 LY B N L
XT S S D -E _m _I In In p _ n_ _I i 1 IT _ 6 H t 8 A V 4 _ V 4 _ V 4 t R ISC IA_ p R (M n d Gr Gr sat _5_
G tG g_ g_ V C
4
4
FI S_ N_ G_ VB I ST An An_ n_ n_In In p 1 A An ig 0 IN RU 3 2S sa 3 _(IS 4_
a
D
C
m
9
D
4
4
A
A
V
O
4
_
A _
C
Vs 47 8 8 cl a 6 S Vsa al o al o _4_ 0 IG
38 39
_( 3_
N C DI 4 I B 01 2 0 A n 0.0 02 4 32 1 C
41
43
4 V 1
n n C 4
-K AP s t- 21 7.1 4 0. 4 0.1 0.1 .08 0 A 28 9 0. 30
38 7 2
63
35
TM
4
1 5 52 7 1 A 2 A IG 5
2
5
0
M
0
2
2
A
3
.
hk hk- Te
2 2 6 0
2
T
50
3
C
5 55 3 3
2 27
.
A
C
e
5
_
7
5
2
n _ g
38
7.
Pi in id
0 0 P Br
36
1
2
2
1_ _ _
20 20 1 2_4
20
SETNET, London, 13 September 2001
* digital engine control unit
excerpt from
(400+) test sequence*
The Cost of Test
5
optimiSE
from data to knowledge
Test ranking by reduTec –
stop on first failure
tests
defect
devices
potentially redundant
tests (white)
test analysis protocol
essential tests (black)
information
gain
SETNET, London, 13 September 2001
optimal test sequence
w.r.t. first failure out
original pass / fail
test sequence
The Cost of Test
6
optimiSE
from data to knowledge
Sequencing with reduTec
- basics for the binary case
test #
object #
19 single tests to detect
deficiencies
reduTec
14 single detection tests
1 globally redundant test
SETNET, London, 13 September 2001
The Cost of Test
7
optimiSE
Estimation of test cost savings
from data to knowledge
0,31 €
0,31 €
Materials
Personell
- 33%
- 16%
0,26 €
0,20 €
Overhead
Capex (Handler)
Capex (Tester)
Tester
Parallelity
Price Tag
Investment
T5365
64
0,9 Mio. €
Devices under
test (DUT) for
amortization
SETNET, London, 13 September 2001
T5581H
128
1,9 Mio. €
1,0 Mio. €
10 Mio. DUT
T5365
10% savings under
80% correlation
with test time reduction
reduTec
50-100 T €
50-100 T €
0,3-0,8 Mio. DUT
The Cost of Test
8
optimiSE
from data to knowledge
Further developments within
reduTec
Test Data Mining
 Automated test- and object qualification based on user-specific
expert system
 Online neuro-fuzzy-system association rules
 Online analytic processing (OLAP)
 Test and know-how data bases for improved process control and
test management
 from data to knowledge
SETNET, London, 13 September 2001
The Cost of Test
9