Modeling Active Components for Fast and Accurate Simulation
1. PB 1990-2009 AN 02
DWN & DWV
MODELING OF ACTIVE COMPONENTS
The fast growth of signal integrity equivalents, as those utilized in the
and EMC problems of digital transmission line simulators, are The modeling approach shown in
systems requires very accurate not sufficient to describe the this application note is based on the
modeling techniques of active unpredictable dynamic effects of primitives offered by the DWN
components, especially regarding real devices. simulator and allows the designer to
their dynamic behavior at high
speed. On the other hand, the
growing complexity of systems
imposes severe goals regarding the Ctdr
coax
simulation speed. As known the
two requirements of speed and Z0 DUT
accuracy are opposite, so that the
R
traditional approach to these Vbias vcc
problems leads to unsatisfying TDR gnd
ground plane
results. SPICE models, for
example, are not suitable for the Fig. 1: Measurement set-up for TDR characterization.
simulation of circuits with more
than few hundred elements, A behavioral or mixed electrical- perform very accurate electrical
because the simulation time rises behavioral approach is much more simulations with a speed at least
prohibitively with circuit effective to face real-world two orders of magnitude greater
complexity and convergence situations, where the effects of than other commercial products1.
problems could occur. active parts, lossless or lossy This modeling procedure is simple
Furthermore, the results often don't interconnections, EMC constraints, and the data can be collected from
reflect the real situations. In fact, and electrical, logical and timing several sources. In particular, it is
the parasitic effects introduced by issues must be taken possible to extract model
the discontinuities of the package simultaneously into account. parameters from datasheet, from
and the pin bouncing (caused by analog simulations (SPICE, ELDO,
simultaneous output switching) DWN MODELING etc.) or, better, directly from
can often invalidate the results. On APPROACH TDR (Time Domain
the other hand, simple circuital Reflectometer) measurements.
1.00 # Thanks to this fast experimental
#RHO
D approach, the model can take all the
0.80 #
non-linearities of the I/O cells of
0.60 # the device into account,
C
0.40 #
B
0.20 #
A
-0.00 #
-0.20 #
-0.40 #
-0.60 #
-0.80 #
-1.00 #
1DWN uses a very fast DSP engine that
0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80 2.00 2.20 2.40 2.60 assures high speed without convergence
TIME[nS] problems.
Fig. 2: TDR response for a CMOS input
HDT Copyright 1990
2. PB 1990-2009
behavior of the input circuitry (the
STF
1 Tin Vout reactive effects of the input gates in
2
1 this particular case). The final value
Z0,Td of the transient C (point D) is the
0 +
Bin -R 0 5 Vin reflection coefficient determined by
E1
the input ohmic resistance of the
device under test in parallel with
Fig. 3: Simple CMOS input model including logic level shift. the biasing resistor R. It is possible
to compensate the error introduced
by the resistor R with the insertion
as well as its dynamic behavior. the package must be kept as short
of a negative resistor -R in the
The currents flowing through the as possible.
model description, as shown in
model are simulated with accuracy The DC biasing of the input is
Fig.3. Sweeping the bias voltage
and can be used for ground obtained via a variable supply
within the possible operation range
bouncing noise analysis and EMI (Vbias). The resistor R is chosen as
(0V - 5V) there is no practical
verifications. Furthermore, these high as possible, in order to provide
deviation of the behavior shown in
TDR-based models are typically a high-impedance path for the
Fig.2, so that a simple linear model
is suitable for this situation. The
.SUBCKT INCMOS 1 2 choice is a mixed TLM
* TLM model of package (Transmission Line Modeling) and
TIN 1 0 3 0 Z0=75 TD=200PS BTM (Behavioral Time Modeling)
* input dynamic behavior
approach. In particular the package
BIN 3 0 S11=PWL(0 0.2 30PS -0.6 0.6NS -0.5 1.5NS 0.7 2.5NS
+ 0.85 5NS 1.0) Z0=50 TD=0 effect can be modeled as a short
* voltage shifter ("0" -> 0V, "1"->1V) transmission line, while the
E1 2 0 4 0 PWL(0V 0 2.4V 0 2.6V 1 5V 1) behavior of the active input is
.ENDS INCMOS directly modeled by a PWL one-
Fig. 4: Simple CMOS input model description (DWN syntax). port scattering element (Bin)3. The
separation of package effects
wideband2 and are suitable for reflectometer, but low enough to allows the user to simulate the
EMC analysis of even slow bias the input at the desired level. behavior of internal input node so
components. A typical TDR response is shown that it is possible to extrapolate
in Fig.2 where it is possible to
SIMPLE CMOS INPUT
10 VDD
This section is dedicated to the Bdvdd
modeling of a CMOS input without ASvdd
protection diodes. The model is
ST F
based on experimental Pvdd Vout
reflectometer characterizations of 1 T in 1 2
the input stage of the device. The
Z0,Td 0
test-fixture is shown in Fig.1. The Vin +
0 5
TDR output pulse (typically a Bin Pgnd
E1
voltage step with 250mV amplitude
Bdgnd
and 25ps rising edge) is fed to the ASgnd
input pin of DUT package via a
DC-block capacitor and a semi rigid 20 GND
50 coax cable. The realization of
Fig. 5: CMOS input model including protection diodes and supply pins.
the test-fixture must take all high-
speed issues into account, so that identify four sections: the peak device responses with other
ground and Vdd supply planes must identified by the label A is the packages. If this is not required, a
be provided and the interconnection parasitic effect due to the launch whole BTM model can include the
of the device to the fixture must cable where it is connected to the package effects.
reflect that of real operation. The device under test.
connection of the launch cable to The section B is related to package 3 The PWL (PieceWise Linear) fitting of
and package-die bond effects. The Bin is easily extracted from the actual
2 Current TDR analysis offers band up to section C shows the dynamic behavior using the MCS (Model Capture
40 GHz. System) facility of DWV.
3. PB 1990-2009
0.60 #
clamping diodes for protection
0.40 # against electrical discharge, so an
Iclamp=0.2MA accurate model must take both their
0.20 #
non-linear and dynamic effects into
-0.00 # account, as shown in Fig.5. In this
situation it is necessary to introduce
Iclamp=1MA
-0.20 #
the effect of power and ground pins
-0.40 #
because during the clamping action
the diode current flows through
Iclamp=5MA
-0.60 #
them.
-0.80 # Iclamp=20MA
-1.00 #
0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 18.00 20.00 V (V) I(mA)
TIME[nS]
0.00 0.0
a) 0.50 0.1
0.60 #
0.55 0.3
0.40 #
0.60 0.9
0.20 #
0.65 1.8
Iclamp=0.5MA
0.68 3.8
-0.00 #
0.70 10.7
-0.20 #
Iclamp=1MA 0.72 25.4
Tab.1: Example of static charac-
-0.40 # teristic of clamping diode.
-0.60 #
Iclamp=5MA Often each diode has a specific
static and dynamic behavior.
-0.80 #
Iclamp=20MA
Sometimes the static characteristic
is available from the datasheet with
-1.00 #
0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80 2.00
a format similar to tab.1. Biasing
TIME[nS] the device at 5.7V (on the DUT
b) pin) it is possible to test the
Fig. 6: Reflectometer response of protection diodes versus clamping current: a) VDD diode, clamping effect toward the supply,
b) GND diode. while a bias voltage of -0.7V
tests of the ground clamping diode.
In order to simulate the I/O timing By changing the current in
properties of the device it is CMOS INPUT WITH clamping conditions, it is possible
possible to include in the model the PROTECTION DIODES. to collect a family of reflectometer
input static transfer characteristics responses for each protection diode.
of the device with output values Usually, a CMOS input has two Fig.6 shows an example of these
shifted to the logical state "0" and
"1". In this way the model is 0.60 #
directly interfaceable to a core 0.40 #
block able to model the timing logic
behavior of the component. Fig.4 0.20 #
shows DWN net list of the model
-0.00 #
composed by the Bin block and the
transmission line that models the -0.20 #
package contribution. The level
-0.40 #
shifter completes the model. DWN
uses a SPICE-like syntax and this -0.60 #
model can become a DWN sub
-0.80 #
circuit.
Upon the model is completed, it is -1.00 #
possible to validate it through a 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80 2.00
TIME[nS]
simulation of the measure set-up Fig. 7: PWL fitting of a clamping diode reflectometer response: the last sample is
that has been used during the extrapolated to -1000m .
characterization (see App.A).
4. PB 1990-2009
10 VDD internal available) or directly from V-I
DT F measurements and used for Pvdd
Vout E1 and Pgnd (non-linear resistors)
descriptions;
t Pvdd - plot the TDR dynamic behavior
of both protection diodes in
ST F
sw1 clamping condition (with a suitable
Vout 2
1 value of clamping current4). The
sw2
Z0,T d bottom value of the response,
Bout
Vin always higher than -1000m (for
DT F Pgnd typical CMOS protection diodes
Vout the value ranges from -700m to
E2
-900m ), is the reflection
t
coefficient related to the ohmic
20 GND internal
resistance of the diode. Because
this resistance has been already
Fig. 8: CMOS output behavioral (BTM) model.
taken into account by the non-
linear resistor Pvdd and Pgnd, it is
graphs for a CMOS EPROM in DIL - extract the package and Bin
necessary to compensate this effect
package. It is interesting to observe descriptions as already explained
during the PWL extraction with the
the different behavior of the two
related utility MCS of DWV, as
diodes. R( ) shown in Fig.7.
The ground diode shows a very fast
1M The dynamic behavior of the VDD
response so that low impedance sw2 sw1 and GND path through the supply
levels are reached in about a 1K pins is easily modeled in a
nanosecond. On the contrary, the
behavioral way. This can be
Vdd diode shows a slow transient 1
accomplished by means of the two
(several nanoseconds long) before
2.5 5 S-parameter blocks Bdvdd and
it reaches low impedance levels. Vcontrol Bdgnd obtained as PWL fittings of
This slow behavior obviously has Fig. 9: Static characteristics of the two actual TDR behaviors. The two
significant impact in the clamp switches.
series adaptors ASvdd and ASgnd
action, so that, if a voltage
are used to connect the blocks in
overshoot due to reflections occurs, for the model of Fig.3;
series to the supply paths. In fact
it will be effectively clamped only
after the delay observed in the TDR 0.60 #
characterization. This kind of
situation is very common and 0.40 #
difficult to forecast. This modeling 0.20 #
1MA
approach, based on TDR
measurement, is the best way to -0.00 #
pinpoint these effects. Furthermore, 2MA
during the modeling, the user gets -0.20 #
also a lot of information, not 5MA
-0.40 #
supplied by the manufacturer,
concerning the "quality" of the -0.60 #
10MA
component he is going to use.
20MA
To complete the model shown in -0.80 #
Fig.5 the following procedure is
suggested: -1.00 #
0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80 2.00
TIME[nS]
Fig. 10: TDR responses for an ECL output versus biasing current.
- determine the static input the series adaptor connects the one-
characteristic of the DUT in normal
and clamping conditions. The data
can be extracted from datasheet (if 4 It is suggested to increase the clamping
current up to its maximum limits.
5. PB 1990-2009
performed by two switches
Pout AS Tout 2
DTF controlled by the output voltage
Vout STF Vout Z0,Td itself. The two switches,
1
Eout Bout implemented by voltage controlled
Vin t
+ resistors, have low impedance when
closed in order to not affect the
20 GND
output impedance value. Fig. 9
Fig. 11: ECL output behavioral model
shows suggested static transfer
characteristics that could be used to
control the switch resistance. The
port element placed at its third port pin bouncing phenomena, as it will current flowing from VDD to the
between the two nodes be discussed later. load during the 0->1 transition and
corresponding to its remaining from the load to GND during the 1-
ports. CMOS OUTPUT. >0 transition is accurately modeled.
The TDR dynamic behavior of the Because the switches transitions are
diodes in clamping conditions Typically, the outputs of a CMOS not abrupt, also the current feed
takes into account the driver act as linear resistor until a through between power and ground
contributions of all the elements determined output current level is during the transitions is modeled, a
present along the signal path and, reached. Increasing the current typical effect of CMOS circuitry.
in particular, the package effect of level, the output enters the The two dynamic characteristics are
the input pin, the diode saturation region, where its modeled directly from
capacitance, the on-chip power and resistance grows. measurements of the output voltage
ground rails and the bonding and Other non-linear effects are transient with the driver unloaded,
package effects of the power pins. introduced by the output clamp using PWL fitting. The non-linear
If the diode is a "good diode" (that diodes. The proposed model resistors represent the static
means that the low impedance accepts logic levels (0,1) as input characteristics of the output stage in
condition is achieved in less than 1 and the first STF block (Static both normal and clamping
ns), its intrinsic dynamic effect is Transfer Function) shifts the conditions.
negligible compared with the incoming (logic levels) signal to the The Bout block models the dynamic
behavior of the output in normal
operating condition.
VDD AS
A description of the package
completes the model.
Bvdd out1
in1 IN OUT
ECL OUTPUT
in2 IN OUT out2
Timing/logic
CORE
An ECL device presents a strong
IN OUT
(logic levels) non-linearity of the output
resistance at low current levels. In
fact, the ECL output is
inm IN OUT outn
Bgnd implemented by a bipolar emitter
follower that has low output
GND
resistance (normally less than 10
AS
at 10mA biasing) for normal
Fig. 12: Complete device model including logic/timing behavior and simultaneous operating current. During the 1-> 0
switching noise effect.
transition there are situations in
power and ground distribution desired output levels. The two which the output transistor goes
effects, so that the blocks Bdvdd dynamic transfer functions (DTF) near cut-off and its output
and Bdgnd can be obtained from model the actual behavior of the impedance greatly increases. Fig.10
the measured dynamic behavior of output waveform without load. shows the reflectometer response of
the power pins5. This model The dynamic behavior of the an ECL output with different
allows accurate simulation of the clamping diodes is taken into biasing current for both "1" and "0"
account by the dynamic model of logic states. It is possible to point
the power pins, as discussed in the out the sensitivity of the ohmic
previous section. The switching resistance (given by the right-end
5 Extracted from TDR characterizations of between the two logical states is steady-state responses) versus the
the power pins. current.
6. PB 1990-2009
The dynamic behavior is usually output drivers affects the input times (~1 hour). The wide-
the same for both "0" and "1" logic stage operation. bandwidth models are also suitable
states6 as well as the static output The core sections model the for EMC analysis and in very high-
characteristics. The model structure internal logical functions of the speed applications involving the
is shown in Fig.11. device which can be represented by use of new packaging and
The static transfer function logical functions (AND, OR, etc). interconnection technologies, like
translates the signal from the logic The logic behavior can be modeled the MCMs (Multi-Chip Modules).
input levels to ECL output levels. in DWN using voltage controlled
The dynamic transfer function has switches. APPENDIX A
been represented by a PWL fitting With a correct introduction of
of the output waveform (driver delays along the input, core, or This section describes the
unloaded). The non-linear resistor output sections, the macro model validation of the device model that
is described using the V-I static
characteristic of the output. The
E1=Vcs Ctdr sw1
Bout and package models are
Model
extracted with the same procedure +
1 under
described in the previous sections. test
The pin #20 can be connected td=50ns sw2
directly to GND or to other +
25mV/25ps
circuitry taking pin bouncing Vtdr Cs=1pF
Bias
effects into account.
55ns t
PIN BOUNCING EFFECT Fig. 13: Simulation scheme for model validation.
All the models proposed for input allows accurate electrical, timing has been created using the
and output sections of the device and logic simulation at the same behavioral approach. The model is
are suitable for simulations that time. validated through a simulation of
take the simultaneously switching the same measure set-up that has
noise into account. The model of a CONCLUSION been used for device
complex I/O section of a device characterization. The model is
with package effects is shown in The effects of high-speed operation validated by direct comparison of
Fig.12. of active devices are very complex the simulated waveforms with the
All the I/O models are coupled by and involve several phenomena. actual ones. A very fast step (25ps)
the dynamic behavior of the power The only way to get accurate simulates the TDR pulse. Due to the
and ground rails on chip, by the models is through experimental decoupling capacitor connected to
bonding wires and the package characterization that can take all the the TDR, the initial transient during
itself. The Bdvdd and Bdgnd effects into account, pinpointing the simulation could be very long.
elements are used once and replace unusual operation modes that often In fact, the TDR step must be only
all the dynamic models of the occur. DWN & DWV allow direct activated after the initial transient is
clamping diodes for both input and utilization of the experimental exhausted and when the device is
output models (with the exception measures to build very accurate and biased at the proper operating point.
of "slow" diodes, as previously reliable models without resorting to Fig.13 shows a scheme that can be
mentioned). The power and ground circuital equivalent. used for the simulation of TDR
coupling allows a good simulation In this way the user is able to build responses. At time 0 the switch sw1
of the pin bouncing effect in both up his own library of DWN models is open and the switch sw2 is
normal and clamping operating of active and passive components. closed. The bias generator can be a
conditions. It is interesting to point These models can be utilized for current generator with very high
out that the macro model (that is fast and accurate simulation of internal impedance or a voltage
described as a sub circuit using the digital systems both in the pre- generator with an internal resistor
SPICE-like syntax of DWN) layout and in the post-layout phase greater than 10k . The circuit is
couples the input section of the of the design. In order to automate designed in order to use a current
device with the output one. As a this type of analysis, DWN & generator for current greater than
consequence, the noise due to the DWV are interfaced to the most 100µA (for example validating
popular CAD/CAE environments clamping diodes) and a voltage
through a specific tool (PRESTO) generator for current lower than
6 In fact, a unique transistor is active for allowing accurate and exhaustive 100µA (for example testing
both the logical states. simulation of PCB boards in short
7. PB 1990-2009
elements with very high input
impedance). The values of the
current or voltage bias can be
positive or negative (source or
sink). After 50ns the device is
biased at the correct operating point
and the generator E1 copies the
voltage present at the pin of the
model. At this time, the two
switches change condition and a
step of 25mV amplitude and 25ps
rise time is fed to the model. The
amplitude of the signal at the test
point 1 multiplied by 80 (because
of the 25mV step) gives the
reflectometer response scaled
between -1000m and +1000m in
order to provide a direct
comparison to the TDR measure.