Demystifying Measurements
from HBM and CDM ESD Testers
(How to Solve ESD Problems by
Thinking in Pictures)
Tim Maloney
Intel Corp.
IEW Seminar
Outline
Tim Maloney, Intel Corp. 2
•Quick review of math background
•Complex impedances, Laplace Transforms, s-domain •HBM as a 2-pole circuit
•Extracting RC and network from waveform •Measurement aside: Transformer effects •CDM as a 2-pole circuit
•Network from Ipk, Qfp, Qtotal and V0
•Spark rise time and scope response may add poles •Oscilloscope bandwidth limits and effects on waveforms
•Measurement aside: 2-pole oscilloscope model •Other ESD examples: CCDM, TLP
•Antenna response to CDM events •ESD chair antenna
•Appendices:
Flow Chart of Outline
Tim Maloney, Intel Corp. 3
Math review
HBM 2009-11
Antennas 2011-13
Thermal 2013-15
IEC, Chairs
2015 TLP, CCDM
2012
CDM 2013-14
Math Review
Tim Maloney, Intel Corp. 4
Double Exponential (e.g., HBM) F(s) f(t) Differentiate: s d/dt
Integrate: 1/s ∫ Unit step: 1/s
Impulse: 1 ↑
Laplace Transform and inverse:
Gaussian Convolution
Tim Maloney, Intel Corp. 7
⇔
“flip and slide”*
) exp(
)
( 2
2
f
t t
f
τ
− =
2 2
2
*g f g
f τ τ
τ = +
Source: www.wolfram.com
Convolution in Excel
Tim Maloney, Intel Corp. 8
Suppose you have two functions with identical time steps and want to convolve them—these could be, for example, a digital record of a scope trace and a calculated filter impulse response in the time domain.
Once you have the filter impulse response, there’s no reason to sweat FFTs, number of points in powers of 2 (don’t you hate that?), etc., you can install free “Excellaneous” VB macros, at
http://www.bowdoin.edu/~rdelevie/excellaneous/#downloads , into Excel and use one of several convolution macros.
Macrobundle12 is the latest, appearing as a Word file.
This is what was done for modeling oscilloscope response in a 2011 EOS/ESD paper,
Flow Chart of Outline
Tim Maloney, Intel Corp. 9
Math review
HBM 2009-11
Antennas 2011-13
Thermal 2013-15
IEC, Chairs
2015 TLP, CCDM
2012
CDM 2013-14
Two-pole HBM (or CDM) Network
,
1
)
(
2+
+
=
RCs
LCs
Cs
s
Y
)
(
)
(
),
1
(
)
(
0 0a
s
s
aV
s
V
e
V
t
V
at+
=
−
=
−C
L
R
So
)
)(
)(
(
)
(
0c
s
b
s
a
s
abc
CV
s
I
+
+
+
=
and
c
b
RC
=
1
+
1
(Vieta, 1579)
Slide 11
Elmore Theorem (1948)
∫
∞ −=
0dt
e
t
h
s
H
(
)
(
)
st∫
∞⋅
⋅
⋅
+
−
+
−
=
0 3 3 2 26
2
1
st
s
t
s
t
dt
t
h
(
)[
]
∑
∞∫
= ∞−
=
0 01
k k k kdt
t
h
t
s
k
!
(
)
.
)
(
A waveform
h(t)
transforms and expands in powers of t:
0
thmoment (k=0) is integral (total charge Q
t
for HBM)
Waveform Moments
∫
∞
=
0
0
I
t
dt
m
(
)
= charge Q
= centroid (time)
= Elmore Delay
= 2
ndmoment
For a current function,
I(s)=
m
0(
1+m
1s+m
2s
2+…+m
n
s
n)
∫
∫
∞ ∞=
−
0 0 1dt
t
I
dt
t
tI
m
)
(
)
(
∫
∫
∞ ∞=
0 0 2 22
I
t
dt
dt
t
I
t
m
)
(
!
)
(
Slide 13
Simple extraction of any moment
Start with a network function (impulse response),
N(s)=
a
0+a
1s+a
2s
2+…+a
n
s
nStep response (HBM, CDM, TLP) is
M(s)=
a
0/s+a
1+a
2s+…+a
ns
n-1DC offset (
a
0/s)
gives
a
0,
remove to
get
L(s)=
a
1+a
2s+…+a
ns
n-1Now integrate to get
K(s)=
a
1/s+a
2+…+a
ns
n-2DC offset gives a
1.
Successive
Integration*
Tim Maloney, Intel Corp. 14
) ) 2 ) (( ) ( 1 ( ) 2 1 )( 1 ( ) ( 2 2 2 0 2 2 2
0 ≈ − + + + − − − +
+ + +
+
= CV RC s RC LC RC s
s s LCs RCs CV s
I e e e e
e e τ τ τ τ τ τ
0.5 1.0 1.5 2.0
2 2 4 6 8
Expand I(s) around s=0, using
Now integrate, i.e., multiply by 1/s, to get charge Q(s):
) ) 2 ) (( ) ( 1 ( ) ( ) ( 2 2
0 − + + + − − − +
≈
= RC RC LC RC s
s CV s s I s Q e e e e τ τ τ τ
This is primarily a step of height CV0 , the total charge Q
0.5 1.0 1.5 2.0
2.5 Now suppose we normalize, remove the
step and integrate again… current charge + + + + = − 3 2 1 1 1 x x x x
Time Constant from Integration
Tim Maloney, Intel Corp. 15
+ − − − + + + − ≈ − ) 2 ) (( ) ( 1 ) (
1 2 2
0 e e e e RC LC RC s RC s CV s Q s τ τ τ τ
Again, suppose we normalize, remove the step and integrate again…
This is a (negative) step that measures RC+τe (=A1-A2+A3 areas):
0.5 1.0 1.5 2.0
0.2 0.4 0.6 0.8 1.0 1.2 A1 A2 A3 Q(t)/CV0 nsec
We now have C from V0 and integrating I(t) to get Q. We get R from the 2nd
integration if we can estimate
About Successive Integration
Tim Maloney, Intel Corp. 16
0.5 1.0 1.5 2.0
2 2 4 6 8 nsec amps
Centroid, 267 psec
RC+τe is the centroid (“balance point”) of the original current waveform
Remove the time constant, continue the integration and you get an expression including inductance L. This can be repeated to the limits of measurement accuracy.
+ − − − + + + − ≈ − ) 2 ) (( ) ( 1 ) (
1 2 2
Slide 17
Obtaining true RC time constant
RC=A/Q
t, the Elmore Delay of the integrated HBM current
t t
Q
Q
RC
∫
∫
∞
−
=
0 0dt
d
τ
τ
h
t
)
(
Current Transformers for HBM Waveforms
)
)(
(
)
(
b
s
a
s
s
s
T
+
+
=
Transfer function:
where
a
= 1/
τ
xf(lo-freq cutoff)
b
= 1/
τ
hf(hi-freq cutoff )
s
=
σ
+ j
ω
(complex frequency)
Zero at
s
= 0 guarantees
eventual zero integral
Slide 19
Tektronix CT1, CT2 Current Probes*
*figures used with permission from Tektronix, Inc.
1 MHz
10 KHz
Current Probe Step Response
•
For CT1,
τ
xf≈
6.35
µ
sec
•
For CT2,
τ
xf> 100
µ
sec
•
This has considerable effect as “real” waveform transforms
to measured waveform
– Impulse response is derivative of step response
– Convolution means eventual negative values measured
V
t
τ
xfV
t
τ
xfstep
impulse
Slide 21
HBM 0
Ω
Waveform, CT1 Convolution
∫
∞
−
=
0
1
(
t
)
HBM
(
τ
)
CT
1
(
t
τ
)
d
τ
HBM
CT impHBM waveform, calculated
0 1 2 3 4 5 6 7
0 100 200 300 400 500 600 700 800 900
time, ns c u rr e n t, a rb u n it s
HBM and CT1 convolution
-0.02 -0.01 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
660 680 700 720 740 760 780
time, ns c u rr e n t, a rb u n it s CT1 HBM pure HBM
Calculation shows zero crossing of measured waveform
Slide 22
4-pole HBM model, with socket cap C
2
C
hbC
1R
hbL
1C
2R
1When R
1= 500
Ω
,
,
1
)
1
(
)
(
4 500 4 3 500 3 2 500 2 500 1 1 0 500s
b
s
b
s
b
s
b
s
C
R
C
V
s
I
hb hb− − − −
+
+
+
+
+
=
b
1-500=R
hb(C
hb+C
1)+R
1(C
hb+C
2),
Slide 23
0 0.5 1 1.5 2 2.5 3 3.5
0 100 200 300 400 500
time, nsec
Cu
rre
n
t,
A
4 kV 0
Ω
HBM Waveform with CT1
t
d= 126 nsec
decay constant t
dis out of spec
CT2 removes droop and t
dis in spec
36.8% I
pkCu
rre
n
t,
Effects of Low-Frequency Cutoff
CT2 and CT1, 500 ohm
-0.01 -0.005 0 0.005 0.01 0.015 0.02 0.025 0.03
760 780 800 820 840 860 880 900 920
time, nsec
cu
rr
en
t,
A
CT1
CT2
CT2 provides a more accurate waveform tail
Slide 25
0
Ω
and 500
Ω
HBM waveforms
)
C
(C
R
C
R
C
R
b
C
R
2 hb
1 hb
hb 1
hb 500
1 500
hb hb
+
+
=
−
=
=
−
τ
τ
0Socket Capacitance C
2
, from Waveforms
hb
C
R
C
=
−
−
500 0 500
2
τ
τ
Tester
C
hb, pF
R
hb, k
Ω
τ
0, nsec
τ
500, nsec
C
2, pF
MK-4
114
1.412
161
225
13.9
Zapmaster
512
101
1.445
146
222
51
Zapmaster
Slide 27
Designing for Reduced Socket Capacitance
C1
Chb
Rhb
Z0
R1
V0
Zin
Most tester “socket” cap is distributed, can benefit
from high Z
0Could raise Z
0with ferrites…
2 1
2 1
2 0 2
1
1
)
1
(
1
1
C
s
R
R
Z
sC
R
Z
Y
in
in
=
≈
+
−
=
+
α
α
, reduces effective cap
2 1 0
C
L
Slide 28
Using Distributed Resistance in Tester
α
C
2Z0’, R
a
Y
inR
1R
1 RaTake some of
R
hband
distribute in-line
+
′
−
+
+
+
≈
)
(
1
1
1 1 2 0 1 1 21
R
R
R
Z
R
R
R
sC
R
R
Y
a a a in 2 ' 0C
L
Z
=
tα
, reduces effective cap
For
α
<0.5, need
2 1 2
0 2
2
Z
R
R
a+
′
>
If R
1= 500
Ω
,
Slide 29
Additional Discussions in the Paper
•
Expression for full loop HBM current I
full(s) for 4-pole
model
•
Expansion of H(s) transfer functions
•
Pole-zero sum rules
– From Vieta, 1579
– Use sum rule to calculate effect of inductance L
1on 0
Ω
HBM waveform rise time
•
These and other concepts adopted from the signal
integrity field
HBM Paper Conclusions
•
Current probe transformers examined
– Tek CT2 is more agreeable for HBM time
decay spec, little droop due to low-f cutoff
• CT1 still ok for most spec-related measurements
• CT2 is now called out in JS-001 HBM spec (2012)
– Tek CT2 also works better for integrals,
centroids needed for HBM circuit model
elements (accurate asymptotic properties)
•
4-pole HBM model with socket cap re-examined
– Analysis: Time constants (centroids) of 0
Ω
and 500
Ω
waveforms readily yield socket cap
value C
2Slide 31
HBM Paper Conclusions, cont’d
•
Synthesis: Design strategies for reducing tester C
2– “Socket” cap is mostly distributed along line
– Analytical model points to lower effective C
2:
• Effective line Z
0can be raised
– Ferrite loading
• Part of R
hb(e.g., 300 out of 1500
Ω
) can be distributed
along that line
• Effective cap can thus be reduced by half or more
More HBM Trickery (and pictures)
See Appendix II, IEW11 presentation
•
HBM double exp: What are the poles?
•
Short
τ
(a few nsec) from I(t) rise time
•
Long
τ
(about 150 nsec) from
integrated current Q(t) rise time
•
Adjustments for 3-pole HBM
Flow Chart of Outline
Tim Maloney, Intel Corp. 33
Math review
HBM 2009-11
Antennas 2011-13
Thermal 2013-15
IEC, Chairs
2015 TLP, CCDM
2012
CDM 2013-14
Slide 34
CDM Tester simulates event
Vf
Cg 300 MΩ
Cf Cfrg
1 ohm disk resistor here dielectric to
field plate
top gnd plane
. *
+ +
+ =
Cfrg Cf
Cfrg Cf
Cg Cf
Cg Cf Vf
Qimm
Immediate charge packet is
3 capacitors collapse to one equivalent “fast” cap
Simplified CDM Network
Slide 35
2
1
)
(
LCs
RCs
Cs
s
Y
+
+
=
2 0
1 ) ( ) ( )
(
LCs RCs
CV s
Y s V s
I
+ +
= =
R
L
C
V(s)
≈
V
0/s
•
Examine step response of this network
•
R is spark resistance, ~25-30 ohms
2-pole First Peak charge, Q
fp
First half-cycle is what causes damage
2 4 6 8 10
0.1 0.1 0.2 0.3 0.4 0.5
2 4 6 8 10
0.2 0.4 0.6 0.8 1.0 1.2
Qfp (area)
Qfp (16% overshoot)
amps
nC
t, nsec
t, nsec Example for D=0.5, RC0=1 nsec, C0V0=1 nC:
1
),
1
exp(
1
2 0 0<
−
−
+
=
D
D
D
V
C
Q
fpπ
) 1 ( )( 0 0 2
s s s V C s Q + + = 2 0 0 1 ) ( s s V C s I + + =
See Appendix A, esd13
0 0
2
LC
RC
D
=
2-pole I
max
= f(R
eq
,D)
Slide 37
•
For D>1
•
For D<1
))
1
(
tanh
1
exp(
2
1 22 0 max
D
D
D
D
D
R
V
I
eq−
−
−
=
−))
1
(
tan
1
exp(
2
1 22 0 max
D
D
D
D
D
R
V
I
eq−
−
−
=
−Also maps to
package size
I
maxtrend
Lower C
0
lower
D, lower I
max 0 0.2 0.4 0.6 0.8 10 1 2 3 4 5
g
(D)
D -- Damping factor
Imax = g(D) *V0/Req
See Appendix B, esd13
0 0
2-pole RLC calculation
Qa
Qfp V0 Imax
C0
Leq D
Cmax
Req
derived
quantity
measured
quantity
0 0
2
LC
RC
D
=
0 0
V
Empirical Fit for C
0
vs. Target Size
•
Metal targets (JEDEC, P4, P6) represent a worst
case for equivalent capacitance of package area
•
Tester inductance and spark resistance give the
rest of the circuit model
Slide 39 y = 0.6696x
0 5 10 15 20 25 30
0 10 20 30 40 50
C
o
, p
F
Sqrt(Area), mm
Co vs. Target Size
250V 500V
Linear (500V)
τ
= R
eq
*C
0
is linear vs. C
0
small target
large target
Leq=2.39±0.6nH
Leq=4.26±0.4nH
•
Dielectric thickness variation, 14-59 mils, several
companies
•
Linear Req*C
0happens repeatedly
•
JEDEC targets, no ferrites; note low inductance Leq
•
20.6
Ω
slope, 26 psec offset
More
τ
= R
eq
*C
0
linearity
Slide 41 y = 20.562x + 68.704
0 100 200 300 400 500 600
0 5 10 15 20 25
ta u = R e q *C o , p s e c Co, pF Orion2, 250V
Leq=10.33±1.49 nH
y = 27.798x + 8.8219
0 100 200 300 400 500 600 700 800
0 5 10 15 20 25 30
ta u = R e q *C o , p s e c Co, pF Orion2, 500V
Leq=12.03±2 nH
•JEDEC tester with ferrite (higher Leq)
•JEDEC coins plus P4, P6 targets
20.6 Ω, 68.7 psec offset
τ
= R
eq
*C
0
linearity for 7 metal targets
y = 27.482x + 56.849
0 100 200 300 400 500 600 700 800
0 5 10 15 20 25 30
ta
u
=
R
e
q
*C
o
, p
s
e
c
Co, pF
JEDEC, 500V
•JEDEC tester with ferrite, 500V
•27.5
Ω
, 56.8 psec offset
Air Spark + 25 ohms
Slide 43
Leq=11.62±1.16 nH
•
Terminate CDM2 test head with SMA tee, 50
Ω
in each branch
•
50
Ω
termination and 50
Ω
scope channel; 25
Ω
instead of 1
Ω
•
Test like regular CDM, with air spark in series
•
No ferrites but extra Leq seems due to 50 ohm mismatch at top
of test head
•
Req = 48.9
Ω
; 25
Ω
plus air spark
Example of 2-pole RLC fit
•
500V, 25
Ω
+ air spark, 4.57pF small target
•
I
maxand Q
fpmatch are guaranteed by method
2-pole RLC fit for large JEDEC target
•
JEDEC tester with ferrite, 500V
•
RLC fit is 28.6
Ω
, 11.7 nH, 16.3 pF
Slide 45
-4 -2 0 2 4 6 8 10 12
0 1 2 3 4
cu
rr
en
t, A
time, nsec
Large JEDEC target, 500V
measured 2-pole fit
What can we do so far?
•
With these methods, we can readily:
•
Estimate equivalent cap C
0from package size
•
Estimate R
eqfrom slope-intercept of
τ
= R
eqC
0as
measured by metal targets
•
Estimate L
eqfor ferrite/no ferrite situation
•
Therefore, generate a complete set of worst-case
CDM waveforms for a given product situation
– Focus on I
maxand Q
fpat test voltage V
0I
max
vs. Q
fp
, 25 ohm + air spark
Slide 47
sm
lg
P4 P6
•Come close to green (JEDEC) point at intermediate voltage for
each target, closest approach being a least squares fit
•JEDEC legacy result expected when I
max, Q
fpmatch
•Data and analysis for 50
Ω
and 25
Ω
CCDM/CDM2 could be even
better. See III.5 of text, esd13 paper.
RLC element trend: Compression of D
•
Why is the damping factor D compressed into a
mid-range for so many measurements?
Slide 48
(a)
0 0.2 0.4 0.6 0.8 1 1.2
0 2 4 6 8 10 12 14
da
m
pi
ng
fa
ct
or
, D
Co, pF
D vs. Co, dielectric thickness variation
(b)
0 0
2
LC
RC
D
=
Damping factor and EM field dissipation
•
D=1/
√
2 maximizes field energy burn rate of spark,
thus minimizing the time integral, A
e(=
√
2)
•
With optimized D(t) it is possible to beat
√
2…
Slide 49 0 0
2
LC
RC
D
=
0 0.1 0.2 0.3 0.4 0.5 0.6 0.70 1 2 3 4 5 6
Po
w
er
time units
I²R vs. time
D=0.5 D=0.707 D=1
A
p D=0.5 D=1 (a) 0 0.2 0.4 0.6 0.8 1 1.20 1 2 3 4 5
fr ac tio n o f E time units
Remaining Field Energy vs. t
D=0.707
(b)
A
eSee Appendix C, esd13
0
LC
Why you should read the Appendices
•
Appendix A:
First Peak Charge (Q
fp)
– Take generalized i(t) expression for RLC,
integrate first cycle, find that Q
fp/C
0V
0= f(D)
•
Appendix B:
Peak Current (I
max)
– Differentiate i(t) to find that I
maxis V
0/R
eqtimes a
unique monotonic function g(D)
•
Appendix C:
Maximum Burn Rate of Field Energy
– Find i
2(t)R
eq
=P(t), convert to P(s), find that field
collapse time is
τ
=D+1/2D, min at D=1/
√
2
Conclusions
•
RLC calculation scheme for CDM waveforms
– Used 4 measured quantities, V
0, Q
fp, Q
a, I
max– Spreadsheet extraction of RLC
D to fit all 4
quantities precisely
•
Linear trends allow RLC prediction:
– C
0goes as
√
Area of target/package
–
τ
= R
eq*C
0vs. C
0is linear (slope-intercept)
• Fits JEDEC, air spark + 25
Ω
, ferrite-free tests
•
JEDEC legacy conditions (I
max,Q
fp) approximated
– With air spark + 25
Ω
, possibly better with CCDM
•
Damping factor D stays near maximum burn rate
Tim Maloney, Intel Corp. 52
Simplified CDM Network (again)
2
1
)
(
LCs
RCs
Cs
s
Y
+
+
=
Examine step response of this network
R is spark resistance, ~25-30 ohms
V(s)=Vo/s
2 0
1 ) ( ) ( )
(
LCs RCs
CV s
Y s V s
I
+ +
Tim Maloney, Intel Corp. 53
2-pole CDM solutions
•
The poles are
•
where the damping factor, and
•
the characteristic frequency
•
Note only two independent variables for an
RLC waveform profile
•
Thus the same waveform results with
–
L1=
α
L, C1=C/
α
, R1=
α
R,
α
>0
–
many ways to get the same profile
[
21
]
2 ,
1
=
−
D
±
D
−
P
ω
LC
RC
D
2
=
LC
1
=
ω
54
Definitions of CDM Tester Parameters
Tim Maloney, Intel Corp. d2-fplt
(effective)
±V
d1-fplt (probe)
upper gnd plate (Area=Afplt)
field plate
t-fplt
(dielectric thickness)
target dia
targ-th
probe radius = a
L and C Model Parameters
Tim Maloney, Intel Corp. 55
Textbook references for C (e.g., Kaiser):
“Our” Bob Renninger for L, 1991 EOS/ESD:
d=d1, probe length
2a
Measurements of Voltages on IC Leads
Slide 56
• Device held by vacuum in a Discharge Test Fixture simulating
an automated handler pick and place mechanism
• Device held at a distance of 2.5 mm from ground
• Device leads charged to a known voltage
Discharge
Test Fixture
Discharge Target
•
Made with FR-4 double-sided PC board material
•
SMA connector on bottom
•
50-ohm cable to LeCroy oscilloscope with 50-ohm input.
•
Current calculation based on 50 ohms
Slide 57
Top
Bottom
Waveform Analysis
•
I
max, Q
fp, and Q
aare calculated from
oscilloscope data, while V
0is the initial
charging voltage
•
C
0V
0= fC
0* V
0/f where f = 1.5 for
charging at 2.5 mm from ground
•
Previously described computation
methods solve for R,L, and C
Slide 58
Test fixture discharge for 500V, 64-pin device.
Qa
Qfp
Vo
Imax
Co
Leq D
Cmax
Req
derived quantity measured quantity
LC
RC
D
2
=
Time (500 psec/div) Pulse Height
2.27 Amps
Blind Alleys and Dubious Ventures
in CDM Modeling
Tim Maloney, Intel Corp. 59
•Method of Moments for fixture capacitance
•Parallel plate and fringing field estimates work well enough (but we’re glad MoM was done at least once) •Plasma physics modeling of spark
•Too many unknown quantities; may as well use linearized models with Rspark and τspark
•1990s publications can help with the linearization •Field and ground plates are an open-circuited radial
transmission line but does it really have any ¼-wave effects? •Modeling hasn’t shown that the device feels much
•Maybe try a shorted package trace stub in conjunction… •Device could feel resonances that may not be
CDM Summary
Tim Maloney, Intel Corp. 60
• CDM tester configuration and RLC equivalent • RLC circuit extraction from waveforms
• Exact matching of charge and peak current • Remarkable result: RC = R0C + τ0
• Describes how R varies with capacitance • Trends in damping factor D
• Estimating L and C from hardware properties
Flow Chart of Outline
Tim Maloney, Intel Corp. 61
Math review
HBM 2009-11
Antennas 2011-13
Thermal 2013-15
IEC, Chairs
2015 TLP, CCDM
2012
CDM 2013-14
The Problem
Same CDM pulse on fast and slow oscilloscopes
Filter models needed for measurement channel
-4 -2 0 2 4 6 8 10
0 500 1000 1500 2000 2500
c
u
rr
e
n
t,
A
m
p
s
time, psec
500V small JEDEC target
8 GHz Tek scope
1 GHz Tek scope
Use LTI* System Theory
Slide 63
*LTI=Linear, Time-Independent
Generalized transfer function:
Expansion in powers of s:
Use to model voltage step and scope
m>n
Tim Maloney, Intel Corp. 64
Treats Gaussian, 1-pole, and 2-pole response functions:
2 2 D for 2 is ; 2 2 1 2 2 1 2 0 3 4 2 2 0
3 = − + ⋅ − + = π =
ω π
ω
dB
dB D D D f
f 2-pole bandwidth: Gaussian bandwidth: πτ 2 177 . 1 3dB =
f rise time τ10-90%=1.812τ
Bandwidth-Rise Time Product
Tim Maloney, Intel Corp. 65
for Gaussian and for 2-pole with 0.6<D<1. For pseudo-Gaussian, D=0.707.
3396 .
0 % 90 10
3dB ⋅
τ
− ≅f
Convolved 10-90% rise times of Gaussian,
pseudo-Gaussian add in quadrature:
2 2 2
1 2
12
τ
τ
τ
= +0 0
% 90 10
2 2D
RC ; 509 . 1 2
2 3396 . 0
ω ω
π
τ − = ⋅ RC = RC = =
Waveform Analysis
Tim Maloney, Intel Corp. 66
C
L
R
+ + = 2 1 ) ( 2 2 0 s s s V s V e e τ τ Recall thatτ10-90%=1.509τe for a pseudo-Gaussian ) 2 1 )( 1 ( ) ( ) ( ) ( 2 2 2 0 s s LCs RCs CV s Y s V s I e e τ τ + + + + = = Thus
Now has spark rise time. Add a scope filter (2 more poles, pseudo-Gaussian) if scope speed is an issue
Waveform and Peak Current
Tim Maloney, Intel Corp. 67
Total charge Q(nC) for voltage V
84 psec rise time 3 GHz scope RLC series circuit
In Mathematica*, Ipeak can be extracted with a 1-line command:
For quadratic terms, s (σ+jω) in GHz and coefficients in nsec and nsec2
In this case, 60 mil diel thickness gives Ipk=5.495 amps and Q=0.8509 nC for 500V
*Mathematica is licensed software; some progress is possible with free I.L.T. app as cited in references.
+ + + + + + = ) 1 )( 0028 . 075 . 1 )( 0016 . 056 . 1 ( )
( 2 0 2 2
LCs RCs s s s s CV t
1-Whoa! What did he just do with
that I(t) function?
Tim Maloney, Intel Corp. 68
Introduce 84 psec spark rise time:
3 GHz scope response:
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0 1.0 nsec
0.2 0.4 0.6 0.8 1.0 2 4 6 8 1.0 nsec Scope modeling: https://sites.google.com/site/esdpubs/ documents/esd11.pdf + + ⇒ + + ⇒ 2 0 2 0 0 2 0 2 0 0 0 2 1 2 1 ω ω ω ω s s s V s s D s V s V 2 0 2 0 2 1 1 ω ω s s+ +
Gaussian Convolution
Slide 69
Resulting Width:
Bessel-Thomson filtering CDM RLC
Tim Maloney, Intel 70 Using FFPA book values L=6 nH, Csm=5 pF, Clg=15.76 pF,
Calculated ratio Rlg=0.9, Rsm=0.75. Curve is nearly insensitive to D when D<1
Measured ratio Rlg=0.899, Rsm=0.74 (5 or 6 samples) J-lg
J-sm
Ratio=Ipk1GHz/Ipk8GHz
JEDEC CDM targets
noted
Presented at Sept 2013 CDM
TDMR version of esd13 (CDM models)
Tim Maloney, Intel Corp. 71 convolved
(a)
next slide
CDM I
max
multiplier for D
≈
0.707
Tim Maloney, Intel Corp. 72 (b)
Good estimate of peak current change
given scope bandwidth and FWHM, using 2-pole
pseudo-Gaussian scope model
Further notes on CDM waveform
xforms
FWHM adds in quadrature, as with rise time formula
from Gaussian math in 1999 M-S paper:
•
FWHM of f
0=1 GHz is 507 psec; scale from there
•
Q
fp(first peak charge) not affected much; often <1%
•
If FWHM > 1/f nsec, Imax multiplier
≈
1, no correction
•
Our CDM committee’s Orion2 data fits pretty well
•
Large target correction for 1 GHz is 10-13%
•
Small target should use 2-3 GHz scope & correct
•
Small target is at least 40% low for 1 GHz
4-pole Bessel-Thomson scope filter
Tim Maloney, Intel Corp. 74
)
1
)(
2
1
)(
2
1
(
)
(
2 2 02 2 02 2 2 01 2 01 1 0LCs
RCs
s
s
D
s
s
D
CV
s
I
+
+
+
+
+
+
=
ω
ω
ω
ω
For f3dB=1 GHz,
ω01=10.07439 GHz D1=0.620703
ω02=8.9862 GHz D2=0.957974
Scalable to any f3dB!
Flow Chart of Outline
Tim Maloney, Intel Corp. 75
Math review
HBM 2009-11
Antennas 2011-13
Thermal 2013-15
IEC, Chairs
2015 TLP, CCDM
2012
CDM 2013-14
TLP s-domain Model
Tim Maloney, Intel Corp.
Rd
Z0, τ
V0/s Zin of an open-circuited
transmission line of
impedance Z0, transit time
τ, is Z0 coth(τs)
Wave series with time step 2τ, round-trip transit time Simple square pulse if reflection coefficient = 0:
[
]
2 002 0 0 0 0 , ) 1 ( 1 ) ( ) coth( ) ( Z R Z R e e R Z s V s Z R s V s I d d s s d d + − = − − + = +
= −−
ρ
ρ
τ
ττ
[
]
kk d k t u k t u Z R V t
I
∑
τ τ ρCCDM in the s-domain
Tim Maloney, Intel Corp. 77
Z0
Z0, τ
V0/s
ZL CCDM ZL is (nearly)
sL +1/sC but could be anything for this general expression. ) sinh( ) cosh( ) sinh( ) cosh( ) ( ) sinh( ) cosh( ) sinh( ) cosh( 1 1 ) ( 0 0 0 0 0 0 0 s s s Z Z s Z Z s V s Z s Z s Z s Z Z s V s I L L L
L τ τ
τ τ τ τ τ τ + + ⋅ + = + + + ⋅ = s L L e Z Z s V s sZ V s s s s Z Z s V s sZ V τ τ τ τ τ τ τ 2 0 0 0 0 0 0 0 0 ) ( 1 ) coth( 1 ) sinh( ) cosh( ) sinh( ) cosh( ) ( 1 ) coth(
1 ⋅ −
+ + + ⋅ = + − ⋅ + + + ⋅ =
CCDM 2-pole Model
Tim Maloney, Intel Corp. 78
C
V(s)
≈
Vo/s
L
Z
0•The transmission line impedance Z0 replaces the spark resistance, aside from relay spark
•Z0=50 ohms is higher than usual air spark R, resulting in lower Ipeak for given V0, but stability is good
Flow Chart of Outline
Tim Maloney, Intel Corp. 79
Math review
HBM 2009-11
Antennas 2011-13
Thermal 2013-15
IEC, Chairs
2015 TLP, CCDM
2012
CDM 2013-14
Objectives (from esd12 & related pubs)
•
Assembly factory
in situ
static event monitoring
• Need for 3-D chip assembly with hi-speed I/O
•
Antenna and detector arrangement in factory
• Create CDM events at will using CDMES (event
simulator) with antenna and detector
•
Detector calibrated with a “standard” pulse
• Artificial, reproducible antenna-like pulse
•
Present theory of
Outline
•
CDM as a 2-pole circuit
• Add spark rise time, map to time domain
•
CDM as a source of dipole radiation
• Detect with monopole antenna
• s-domain expressions for everything
•
Measured antenna signal, agrees with theory
•
Artificial antenna signals with “monocycle” pulser
• Theory and experiment, compared favorably
• Calibration of MiniPulse detector with monocycle
pulser
Simplified CDM Network
•
Examine step response of this network
•
R is spark resistance, ~25-60 ohms
V(s)=Vo/s
C
L
CDM dipole radiation, monopole
antenna
Slide 83
Experimental arrangement of CDM electric dipole
initial source
p
and 6 mm coaxial antenna.
to 50
Ω
scope
p
15 cm
150 cm monopole antenna
•- - - - - -
Charge plate
Coax to 50 ohm scope
Ground plate
CDM Event Simulator (CDMES)
CDM pulse generator. Charge plate probe hits
pedestal and dipole collapses, with current pulse and
dipole radiation.
10 Meg
+V
+++++ +++++ - - - - - -
Simco-Ion CDMES Model
7” long. Voltage cable and signal coax cable shown.
10 Meg
Charge plate
Coax to 50 ohm scope +V
Ground plate
-Slide 87
Electric Dipole “equatorial” field
sinθ=1 at equator
In the s-domain,
s=σ+jω
But we know that
So
Solve in time domain with inverse Laplace Transform;
pole-zero expansion in natural frequencies
s=σ+jω
Calculated current and field
Slide 89
Calculated CDM current pulse,
1 nsec full scale.
E-field pulse E
θat 15 cm from CDM
current source; 1 nsec full scale.
Antenna Transfer Function
From Ref. 10, esd12
(2007 Trans. EMC)
Transfer function in terms of initial
dipole source p
Slide 91
Calculated antenna response to E-field,15
cm from CDM source, 1.5 nsec full scale
Adjust the current function slightly…
Slide 93
Measured current (top) and antenna response (bottom) to
E-field at 15 cm, using artificial CDM source; 2 nsec/division
-100V pulse -0.496 A
max
125 mV =Vp-p -166.4 pC
antenna, 15 cm current
1.664 pF
Antenna problem—ESD effects in chairs
Tim Maloney, Intel Corp. 94
Math review
HBM 2009-11
Antennas 2011-13
Thermal 2013-15
IEC, Chairs
2015 TLP, CCDM
2012
CDM 2013-14
Recent Vermillion-Smith paper*
95
500mV/div, 20 ns/div, 50 Ω antenna termination Antenna is 15 cm long. Suppose it is #18AWG wire.
With concepts in this presentation, you can estimate the initial static field (in kV/cm) at the antenna. The
field then collapses over the time scale shown. You can also prove that the “radiation” is quasi-static, with
very little signal due to 𝑝̇ and 𝑝̈.
ESD Publications
Tim Maloney, Intel Corp. 96
Download most from TJM ESD publications page:
https://sites.google.com/site/esdpubs/documents
HBM:
https://sites.google.com/site/esdpubs/documents/esd09.pdf https://sites.google.com/site/esdpubs/documents/esd10.pdf
https://sites.google.com/site/esdpubs/documents/MR13.pdf (nearly the same as esd10)
plus TJM, IEW 2011 (unpublished)
CDM:
https://sites.google.com/site/esdpubs/documents/esd13.pdf
https://sites.google.com/site/esdpubs/documents/tdmr14.pdf (augmented esd13) https://sites.google.com/site/esdpubs/documents/esd14.pdf (with Arnie Steinman)
https://sites.google.com/site/esdpubs/documents/Ipk-engnrg-2.xlsx (CDM tester modeling program, Excel file)
B. Atwood, et al., “Effect of Large Device Capacitance on FICDM Peak Current”, 5A.1, 2007 EOS/ESD Symposium, pp. 273-282.
Oscilloscope filter models:
https://sites.google.com/site/esdpubs/documents/esd11.pdf
Antennas:
https://sites.google.com/site/esdpubs/documents/pulsdipole11-emc.pdf https://sites.google.com/site/esdpubs/documents/esd12.pdf
https://sites.google.com/site/esdpubs/documents/emc13-proof.pdf (augmented esd12)
Publications, cont’d
Tim Maloney, Intel Corp. 97
Thermal, s-domain:
https://sites.google.com/site/esdpubs/documents/irps13.pdf
plus Industry Council EOS Whitepaper section (not yet published)
TLP, CCDM, s-domain:
2012 TJM Seattle ESDA Tutorial (unpublished)
Pre-work for 2012 Seattle Tutorial (with RLC net practice):
https://sites.google.com/site/esdpubs/documents/prework-tjm1.pdf
IEC, s-domain:
TJM Abstract, 2015 EOS/ESD Symposium:
Other References
Tim Maloney, Intel Corp. 98
• Free online inverse Laplace Transform applet:
http://www.eecircle.com/applets/007/ILaplace.html • Free “Excellaneous” VB macros for Excel, at
http://www.bowdoin.edu/~rdelevie/excellaneous/#downloads
• Rise time-bandwidth, quad addition, pseudo-Gaussians, etc.: C. Mittermayer and A. Steininger, “…Dynamic Errors for Rise Time Measurement with an Oscilloscope”, IEEE Trans. Instrumentation and Measurement", Vol. 48, pp. 1103-1107, December 1999.
• 1998 Lucent paper: R.E. Carey and L.F. DeChiaro, “…Physical Design Parameters in FICDM ESD Simulators …", 1998 EOS/ESD Symposium Proceedings, pp. 40-53.
•Y. Ismail, et al., “Equivalent Elmore Delay for RLC Trees”, ACM DAC, 1999.
•K. Verhaege, et al., “Analysis of HBM ESD Testers…”, EOS/ESD Symposium Proc., 1993, pp. 129-137.
Appendix I—Metal heating, thermal effects,
temp waveforms—from IRPS13
Tim Maloney, Intel Corp. 99
Math review
HBM 2009-11
Antennas 2011-13
Thermal 2013-15
IEC, Chairs
2015 TLP, CCDM
2012
CDM 2013-14
Metal Self-Heating Test Pattern
100
M5
t V RC x
V
∂ ∂ =
∂ ∂
2 2
heat flow for 1-D heat slab:
Cp sK Zth
ρ
1
=
s=σ+jω
K Cp s
th
ρ
γ = Electrical: Thermal:
Volts ⇒ °C, temperature (usually a ∆T from room T) Amps ⇒ Watts
Coulombs ⇒ Joules
Ohms ⇒ thermal impedance °C/W
Farads ⇒ Joules/°C
1 µm2 metal cross section; electrical current through M5
K=thermal conductivity Cp=heat capacity
ρ=mass density
Wires Embedded in ILD Oxide
101
M5
wm+g
…… ……
wm
tox hm
Cmetal
Z01, γoxtox
ZL Z02, γoxg/2
P(t)
Pattern C as shown
Pattern D is wider, gD=2gC
Thermal circuit model:
Note open circuit b.c. Thermal Ohm’s Law:
Thermal Feedback Model
102
+
Z(s)
α
T(t)P
0(t)
P
0(t)
P(s)Z(s)=T(s)
α
= metal tempco
T = temp (“voltage”)
P
0= I
2R
0
in (“current”)
))
(
1
(
)
(
t
R
0T
t
R
=
+
α
α = metal tempco=0.0025/°C for Cumetal resistance ) ( * )] ( ) ( [ ) ( * ) ( )
(t P0 t Z t T t P0 t Z t
T = +
α
[
]
) ( ) ( * ) ( ) ( 1 ) ( * ) ( ) ( 0 0 t T t Z t P t T t Z t P t T α − =or =
∫
− ⇔t s Z s P d Z t P t Z t P 0 0 0
0( )* ( ) ( τ) (τ) τ ( ) ( )
For “current"
source P0(t):
mixer
thermal Ohm’s Law
General Feedback Network
103
+
Z(s)
FB(t)
P
0(t)
Convolve
P(t)*Z(t)=T(t)
T(t)
Z(s)⇔Z(t)
For TLP:
(
)
20 0 2 0 0 0 50 ) ( + = = R R V P t P − + + + = 1 50 ) ( 1 ) ( 1 ) ( 2 0 0 0 R t T R t T P t FB α α
•Because of source resistance Zs=50Ω, TLP introduces negative
feedback, and when R(t)>50 Ω, it becomes net negative
•Current source produces
positive feedback
•Voltage source produces
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
0 50 100 150 200 250 300
Amp lit ud e nsec exp fit FEM data 118 . 0 432 . 0 142 . 0 )
(t = e−t/2.88 + e−t/71.4 + Z
Z(t) from Finite Element Modeling
differentiate and normalize
Step response gives
thermal
impulse response
Z(t)
104 0 100 200 300 400 500 600 700 800
0 200 400 600 800 1000
de lta -T, d eg C nsec
FEM raw data
8.44W step, M6 8.44W, M6-M7-M8
105
0 oC
430 oC
Close-up of metal after 200 nsec
FEM results
M4 M5 M6
Transmission Line Pulsing (TLP)
•
Transmission line pulsing generates brief, high current (several ampere)
pulses; same current/time scale as ESD
Setup:
•
Equivalent circuit
50 ohms Device
Scope
L
V 10 Meg
(Rdevice<50 Ω):
50 ohms
Device
Scope
Idevice=(V-Vdevice)/50
tpulse = 2L/c, c=20 cm/nsec
Apply to metal lines and use Cu tempco (α=0.0025) for in-situ T measurement
TLP data
107 AC
calculated from FB(t)
C e
t
T( ) ≈ 440(1− −t/58)
measured 0 2 4 6 8 10 12 14
0 50 100 150 200 250
W
atts
nsec
TLP Power, M5-C-60V
Apwr
measured
Pattern C, 60V
) 47 1 ( 09 . 4 39 . 8 ) ( s s W s W s P + + = ) 58 1 ( 440 ) ( s s C s T + =
Thermal Impedance
108 2 0 0 2 0 0 ) 50 ( R R V P += for TLP
2 0 1 2 1 0 2 1 1 1 1 2 1 2 1 ) ) ( ( 1 ) 1 )( ( ) ( s C C R R s C R R C R s C R R R R s Z + + + + + + =
•Z(s) from slide 10 has 2 poles and 1 zero; 5-element RC network •For TLP, temp should flatten out at Pfinal(R1+R2) = PfinalZ0 = Tfinal
P(t) ⇔ P(s) C0
C1
R1 R2
C
0= 1.1 nJ/
°
C (metal + oxide)
R
2= 32.7
°
C/W (oxide)
C
1= 20 nJ/
°
C
R
1= 2.52
°
C/W
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.91
0 50 100 150 200
Am
pl
itu
de
nsec
Z(t) from TLP data (°C/nJ)
) 00718 . 0 0185 . 0 ( 26 . 35 )
(t e t/31.6 e t/58
Z = − + −
109 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0 5 10 15 20
Am
pl
itu
de
nsec
Z(t) at short times
Dt model TLP data
Dt model adiabatic
Thermal Impulse Response
) 58 1 )( 6 . 31 1 ( 47 1 26 . 35 ) ( s s s s Z + + + =
from °C/W
thermal sheath for short time: Dt …… …… ) ( ) ( s P s T =
TLP on Patterns C and D, 60V, 1
µ
m
2cross-section
110
Pattern C width = X
T
final= 440
°
C (“volts”)
Z
0C= 35.26
°
C/W (“ohms”)
P
0= 8.24W, P
final=12.16W
A
C(normalized) = 58 nsec
-100 0 100 200 300 400 500
0 50 100 150 200
de lta -T, d eg C nsec M5-C-60V AC
Pattern D width = 1.48X (g
D=2g
C)
T
final= 225
°
C (“volts”)
Z
0D= 23.2
°
C/W (“ohms”)
P
0= 8.24W, P
final=10.99W
A
D(normalized) = 46 nsec
-50 0 50 100 150 200 250 300 350
0 50 100 150 200
de lta -T, d eg C nsec M5-D-60V AD
α= 0.0025
gC ….
Thermal Circuit Elements
111
P(t) ⇔ P(s) C0
C1
R1 R2
Pattern TLP Volts C0 R2 R1 C1 Tfinal
C 60 1.1 nF 32.7Ω 2.52Ω 20 nF 440 °C C 70 1.09 nF 33.9Ω 2.84Ω 21.7 nF 645 °C D 60 1.53 nF 21.9Ω 1.33Ω 29.5 nF 225 °C
from TLP data