[SI-LIST] Re: resend - Specctraquest model: mounted inductance
- From: Larry Smith <larry.smith@xxxxxxx>
- To: bart.bouma@xxxxxxxxx
- Date: Thu, 05 Jun 2003 21:20:36 -0700
Bart - thanks for setting me straight on the X2Y cap. The X2Y publication
has much more detailed information than what they were giving out a few
years ago. Perhaps the patents went through..? I can see now that the
two reference terminals can be treated as a "feed through" capacitor. The
other two terminals are isolated capacitors to the "through" terminals.
There are several ways to hook up this capacitor, which will give several
impedance responses. The measured responses are shown in figure 14 and the
simulated responses are shown in figure 16. Note that one of the responses
is low and flat over a large frequency range, indicating a lot of insertion
loss. That will be true if hooked up in that configuration, but no DC current
will flow through in that case.
Another way to hook up the X2Y cap is with both independent capacitor
terminals to the Vdd plane and both reference terminals to the Gnd plane.
In this configuration, it behaves just like a normal decoupling capacitor.
I believe this is the way in which you intend to use it. It has an
inductance advantage over a normal 2 terminal cap in that there are 4 paths
for the current to go in and out. Actually, you can do almost the same thing
with a two terminal capacitor if you use 2 vias per cap pad, 4 vias in all.
This makes a lot of sense if you have a reverse geometry (i.e. 0306), low
inductance
capacitor. The intrinsic ESL and ESR of an 0306 cap is about 1/3 that of an
0603 cap.
Unless you pay very close attention to your vias and pads, most
of the inductance is in the mount and very little is in the capacitor itself.
Several capacitor manufactures make capacitors with 8 terminals. There are just
two sets of plates with 4 terminals per set of plates, arranged in a
checkerboard pattern of + and -. It has even less inductance than the
previously mentioned caps. The downside is that you have to put 8 vias under
the cap and solder the small structures. If you already have blind vias, this
might
not be too bad, but 8 through hole vias will really chew up wiring channels.
Anyway, the total mounted inductance of a capacitor is a stronger function
of via and pad design that it is of cap design. If you want to decrease mounted
inductance, add more vias.
You have a nice 50 Ohm fixture for measurement of capacitors. Yes, I would call
that a transfer impedance measurement. But, you are measuring
the capacitor in a 50 Ohm environment. The power plane environment where
the capacitors are normally used is more like 1 Ohm and capacitors
behave differently. When caps are mounted with low inductance vias and
pads to PCB power planes that are near the
surface of the PCB and separated by thin dielectric (2 to 4 mils), they
exhibit very interesting resonances. The ESR goes up and the intrinsic
inductance
goes down as the frequency increases past series resonance. This is a very nice
property because the anti-resonance (parallel resonance with another capacitor
or power plane) is greatly diminished by this mechanism. Please see our
ectc_2001.pdf
and ectc_2002_caps.pdf papers for more details.
http://groups.yahoo.com/group/si-list/files/Signal%20Integrity%20Documents/Published%20SI%20Papers%20from%20Sun/
regards,
Larry Smith
Sun Microsystems
Bart Bouma wrote:
>
> Larry,
> thanks for your very detailed answer. I like the Transfer Impedance
> approach.
> You are right when mentioning that the pcb will have 4 vias (Vdd1, Vdd2
> and 2 gnd vias)
> I will concentrate on the X2Y: it's the most interesting three (or 4 )
> terminal capacitor.
> I'm afraid that your idea on how the model of the X2Y looks like is more
> or less based on the feeedthrough capacitor.
> The X2Y is not a feedthrough capacitor, e.g. it has no DC-connection between
> the the "terminations" Vdd1 and
> Vdd2.
> It should be used as a bypass or shunt component, providing a low
> impedance path to ground over a broad frequency range.
> Compare this to what you wrote:
> "If on the other hand, we obtained -120 dB "transfer impedance" by
> keeping the series L and R small and by having a very low
> capacitive impedance (high capacitance), we have a DUT that may be able
> to conduct the 100 amps from vdd1 to vdd2".
> The good insertion loss figures that the X2Y shows are not because it has
> large LLL's, but because of a very low impedance of the capacitor to
> ground, and the X2Y is very well capable of passing by large currents. The
> X2Y doesn't carry DC-current.
>
> The X2Y described simply: the X2Y capacitor consists of 2 capacitors in
> one package. These two caps can be assumed as connected in series, and
> have a common centre node (the 2 ground connections) formed by two
> side-terminations (so from the outside, it looks like a feedthru, but it
> behaves quite differently).
> This is the key to the X2Y's low inductance: currents flow in opposite
> directions and magnetic fields cancel.
> This has been shown with careful placement of 2 standard capacitors in a
> paper by University of Missouri-Rolla and even in a patent by Dell.
> The X2Y is in fact the same, but optimized and integrating the 2 caps in
> one shielded package.
>
> A third capacitor is formed by the 2 above mentioned capacitors in series.
> This results in 2 line to ground capacitors (2 Y-caps) and 1 line to line
> capacitor (1 X-cap).
> In other terms: 1 cap between Vdd1 and Vdd2, and 2 caps placed between
> Vdd1 & Gnd1, and Vdd2 & Gnd2.
>
> vdd1 o
> L
> R
> ---
> gnd1---------gnd2
> .-.
> L
> R
> vdd2 o
>
> To be complete: here is a link to a paper that shows the internal
> structure and the X2Y model:
> http://www.x2y.com/cube/x2y.nsf/(files)/InternalModel060303.pdf/$FILE/InternalModel060303.pdf
>
> We have been carrying out S21 Insertion Loss measurements on a small pcb,
> this 2-sided pcb had one large ground plane at the bottom, the top-side
> consists of a 50 Ohm microstrip line, with two smaller ground planes at
> each side of it. Multiple vias conected the top-ground planes with the
> bottom ground plane.
> The pcb was mounted in a Wiltron Universal fixture, and the X2Y soldered
> on the 50 Ohm microstrip line and grounded at the ground planes.
> See for a more detailed document on this the link below.
> Port 1 and port 2 connected to the same ground plane and to the same 50
> Ohm line, but not at the same position.
> Parallel to this the X2Y, which is located between the 50 Ohm line and
> ground.
> The X2Y was measured in 2 modes: 1 Y-cap only (your Vdd1 and Gnd1
> terminals open) and both Y-caps connected.
> You wrote:
> " If you want to obtain a model for the capacitor to be used in SQPI
> simulation, you would have to make a "transfer impedance" measurement by
> connecting both port 1 and port 2 of the VNA to the vdd2
> and gnd2 terminals of the DUT and leave the vdd1 and gnd1 terminals open. "
> I think that what I described is comparable to your "transfer Impedance"
> measurement. Do you agree?.
> See also the link below.
> ? But how to differentiate the intrinsic inductance from the mounting
> inductance ?
>
> Re. to S11 and S22: when a well defined fixture is used, I believe that
> these two parameters can be measured quite well upto high frequencies.
> Condition: a well done calibration.
> But I am aware that by the limitations of the Vector Network Analyzer low
> impedances, or high impedances (i.e. far away from the 50Ohm of the
> system), can't be measured in this way. I think the limit lies at aprox
> 100 milliOhm or at -48dB.
>
> Larry, one last issue: you mentioned folowing:
> " This measurement will be far different than the "insertion loss"
> measurements given by the 3 terminal cap manufactures. One measurement
> has to do with power going "through" a component, the other measurement
> has to do with power getting "past" a component on highly conductive
> power planes."
>
> The "through" measurement refer to feedthrough capacitors, I assume.
> But is this "past" measurement not identical to your transfer measurement?
>
> When using higly conductive power planes, one can assume that both ports
> are connected to the same power planes.
> Or do I overlook something?
> I think I do understand your S21 measurement and how you extract the Z21
> transfer impedance from the S21 data.
> I think that the measurement method I described, i.e. the X2Y on a small
> pcb, is identical to your transfer measurement.
> What do you think?
> We don't carry out the S21 to Z21 conversion. Just showing the S21
> Insertion Loss data.
>
> Here a link to measurements we did on the X2Y mounted on this pcb: it
> shows clearly the better performance of the X2Y over standard MLCCs.
> The X2Y performs at least 10dB better - at frequencies beyond its
> resonance frequency - than 2 standard MLCCs connected in parallel.
> http://www.x2y.com/cube/x2y.nsf/(files)/X2YMLCC.pdf/$FILE/X2YMLCC.pdf
>
> Thanks again,
> kind regards,
> Bart Bouma
> appl. eng.
> www.yageo.com
>
> =============================================================================
>
> Bart - To directly answer your question about mounting inductance, I
> believe you will want to treat the mount as if the capacitor is a 4
> terminal device. As I understand it, the PCB will have 4 vias with the
> two end vias going to vdd1 and vdd2 and the two middle vias going to gnd
> on a PCB that has a stackup something like this:
>
> pads
> vdd2
> gnd
> vdd1
> more layers...
>
> Since there are 4 vias and 4 pads, an inductance matrix involving 4
> conductors will handle the situation. An EM solver can be used to find
> the inductance matrix. Measurement of this matrix will be difficult.
>
> This brings up some questions about how many terminals the "3 terminal
> capacitor" has? What is the topology of the circuit model that will be
> simulated? Several years ago, I asked similar questions to the
> manufactures of the X2Y capacitor but did not get any clear answers.
> They basically told me not to worry about what is inside the caps and
> just look at the measurements. That is when I got really worried...
>
> To go any further in this discussion, we need to clearly define some
> terminology. For this discussion, I would like to distinguish between
> "insertion loss" and "transfer impedance". People use both of these
> terms when referring to an S21 measurement but I would interpret these
> very differently, even though the magnitude and phase may be
> identical. First, we need a circuit topology.
>
> I don't know what is inside of an X2Y package, but I will speculate
> with the following circuit diagram (hope this works out...):
>
> vdd1 o--LLL-----LLL-----LLL-----LLL-----LLL-----o vdd2
> | | | |
> --- --- --- ---
> .-. .-. .-. .-.
> | | | |
> gnd1 o--LLL-----LLL-----LLL-----LLL-----LLL-----o gnd2
>
> Please consider each inductor (LLL) to also have a resistance. I
> believe that if you attached an ohm meter to the vdd1 and vdd2
> terminals, current would flow through the part with some resistance.
> The gnd1 and gnd2 terminals would behave the same way. You would
> measure a capacitance between either of the vdd and gnd terminals.
> (Somebody please correct me if I am wrong about this..) The above
> diagram behaves in this way. Let's say that the VRM (power source) is
> attached to vdd1 and the load (power sink) is attached to vdd2. Gnd1
> and Gnd2 will end up being attached to the single gnd plane through two
> inductive vias. Vdd1 and Vdd2 are connected to different power planes
> through two more vias, 4 inductive connections in all.
>
> Now, back to "insertion loss" and "transfer impedance". You might make
> a full 2 port measurement of this device by connecting port 1 of a VNA
> to vdd1 and gnd1; and connect port 2 to vdd2 and gnd2. Gnd1 and gnd2
> may even be shorted together if the measurement is made with the DUT
> mounted on a PCB. The S21 measurement will tell us how much of the
> wave from port 1 made it to port 2. If this is a good DUT, we will
> probably see something like -60dB for S21 in some frequency range.
> This is the traditional meaning of insertion loss. Not much of the
> signal got through. Note that we could get large insertion loss by two
> different means: we could make the L and R series impedance very large,
> or we could make the capacitive admittance very large (capacitive
> impedance very small).
>
> In fact, we could make the LLL's high inductance (and high resistance,
> nearly an open circuit) so that there was little conductivity from vdd1
> to vdd2 and have extreme insertion loss, perhaps -120 dB. But should
> this be interpreted as a low impedance power distribution system (PDS)
> that is capable of bringing 100 amps from vdd1 to vdd2? Absolutely not.
>
> If on the other hand, we obtained -120 dB "transfer impedance" by
> keeping keeping the series L and R small and by having a very low
> capacitive impedance (high capacitance), we have a DUT that may be able
> to conduct the 100 amps from vdd1 to vdd2. From the S21 measurement,
> you cannot tell what you have. If you could make clean S11 and S22
> measurements on the DUT, you could figure out how the DUT will perform
> in a PDS. But S11 and S22 measurements below about 60 dB and above 10
> MHz are very inaccurate because the inductance of the fixture dominates
> and it is very difficult to to calibrate or compensate it out. Yet this
> is the frequency and impedance range where power distribution is
> very important.
>
> Therefor, the engineers at Sun Microsystems have been using the phrase
> "transfer impedance" when we refer to the S21 measurement that involves
> port 1 and port 2 of the VNA connected to the same conductors. A good
> example of this is when we take power plane measurements with port 1
> and port 2 each connected to vdd and gnd of the same power planes, but
> at different locations. An ohm meter would tell you that port 1 and
> port 2 are connected in parallel, but at high frequency, this is not
> the case. Istvan Novak has discussed a good way to obtain the "self
> impedance" (S11 at one point on the power planes) by probing the same
> Vdd and Ground vias with port 1 on top and port 2 on the bottom of the
> PCB. The probe inductance does not hurt you in this "transfer
> impedance" measurement. It is a lot like 4 port kelvin probes for DC
> measurements.
>
> In a similar way, I often solder port 1 and port 2 of the VNA to a two
> terminal decoupling capacitor to obtain the impedance vs frequency.
> With PDS and capacitor measurements made in this way, it is possible to
> obtain accurate S21 readings of -80 dB at more than 100 MHz (transfer
> impedance), which is far beyond the published accuracy of the
> instrumentation. But, this should not be interpreted as insertion
> loss, if you understand my distinction between the two terms. Istvan
> has published several papers at Design Conn on the measurment of low
> impedances at high frequencies. We really want to know the S11
> (actually the Z11) for the PDS but that is too difficult to measure, so
> we measure S21, convert it to Z21 and call it a transfer impedance.
> Subtle, but important.
>
> Now, back to the 3 terminal capacitor... Be very careful about
> interpreting the insertion loss measured on a 3 terminal capacitor as
> being the low impedance presented to a power distribution system. It
> is not. If you want to obtain a model for the capacitor to be used in
> SQPI simulation, you would have to make a "transfer impedance"
> measurement by connecting both port 1 and port 2 of the VNA to the vdd2
> and gnd2 terminals of the DUT and leave the vdd1 and gnd1 terminals
> open. This measurement will be far different than the "insertion loss"
> measurements given by the 3 terminal cap manufactures. One measurement
> has to do with power going "through" a component, the other measurement
> has to do with power getting "past" a component on highly conductive
> power planes.
>
> Essentially, the 3 terminal capacitor isolates the vdd1 node from noise
> on the vdd2 node, and vice versa. This may be very useful for
> providing clean power to analog circuits such as a PLL. But we have to
> be very careful in the interpretation of the S21 measurements for PDS
> components. This discussion is closely related to the discussion that
> we had on ports and terminals several weeks ago under subject thread
> "N-port model limitations in simulators"
>
> regards,
> Larry Smith
> Sun Microsystems
>
> > Larry,
> > Thanks for your answer.
> > Yes, we are looking for mounting inductance of a decoupling capacitor
> for
> > use in the SQPI tool.
> > As you know, models for this require inductance values for the part
> itself
> > (intrinsic inductance) and inductance associated with mounting.
> > You wrote: " .. measure the loop inductance on an existing product or
> test
> board ...", I agree, this can be done quite well on a two terminal device
> > (standard MLCC).
> > But how to measure this for a three terminal device, like an feedthru or
> a
> > X2Y-capacitor?
> > Especially the latter: the X2Y consists of 3 embedded capacitors and
> forms
> > actually 3 current loops.
> >
> > Following describes how we think how to attach a three-terminal device,
> > like the X2Y.
> > The ideal situation would be that both end-terminations - as a standard
> > two terminal capacitor has - are attached to two powerplanes, e.g. as
> you
> > mentioned Vdd and Vdd2. The third terminal, which actually consists of
> two
> > side connections (internally connected to each other) must be connected
> to
> > a ground plane in between with at least 2 via's. One at each side of the
>
> > part..
> > Stackup: Vdd - Gnd - Vdd2 with equal distances between the planes.
> >
> > There will be a power flow through the feedthru capacitor, which will
> > result in some power loss due to its series resistance.
> > But not in case of the X2Y, which is attached in bypass (shunt).
> > In case of the X2Y, measurements have shown that a very low impedance
> over
> > a broad frequency range can be reached, due too its low intrinsic ESL,
> and
> > low ESR (as presented at CARTS2002 USA & Europe).
> > The intrinsic inductance of the part itself can be extracted very well
> by
> > a VNA 2-port measurement (as described in some SUN-papers).
> > The part is mounted in a low inductance fixture, designed for
> > three-terminal devices and is de-embedded by a TRL-calibration.
> >
> > Now back to the mounting inductance:
> > almost needless to mention, but to fully employ the low inductance
> > character of e.g. the X2Y, mounting should be paid attention too.
> > How to measure/determine/extract/estimate the mounting inductance for
> > these three terminal devices.
> > Can I assume the mounting inductance to be half the value of a standard
> > capacitor with the same size?
> > For from mounting point of view the two halves of e.g. the X2Y are
> > paralleled. Or is this too simple?
> > Do you have any additional suggestion?
> >
> > best regards,
> > Bart Bouma
> > www.yageo.com
> >
> >
> >
> > Bart - Are you looking for the mounting inductance associated with a
> > decoupling capacitor to be used in the Spectra Quest Power Integrity
> > tool? SQPI has a fast henry type of EM solver that will estimate the
> > loop inductance for the Vdd/Gnd terminals. Or, you can do as we do and
> > measure the loop inductance on an existing product or test board and
> > insert the inductance manually.
> >
> > I must confess, I don't really know what to do with 3 terminal
> > capacitors for the decoupling application. Where do you attach the 3
> > terminals? Two of the terminals will be Vdd and Gnd, but where does
> > the third terminal attach to the system? The mounting inductance
> > for your 3 terminal device will be highly dependent on where the
> > current is flowing (current and return current).
> >
> > Does power flow through the 3 terminal device? For example, do you
> > hook it to Vdd, Gnd and Vdd2? Is Vdd2 a separate power plane in the
> > PCB stackup? A 3 terminal capacitor used in this way may be very
> > useful for a sensitive PLL or similar low power application. But the
> > power consumer will be downstream of the relatively high series
> > impedance of the 3 terminal device. It may be possible to meet a
> > target impedance of several ohms using the 3 terminal device, but it
> > will be very difficult to build a 1 mOhm power distribution system
> > up to the 100 MHz range using 3 terminal capacitors.
> >
> > regards,
> > Larry Smith
> > Sun Microsystems
> >
> > >
> > > Hi all,
> > > I'm resending this post on Specctraquest mounted inductance. Until now
>
> > no
> > > reactions received on this topic.
> > > Is there no one on the si-list who can help a vendor to generate
> > > Specctraquest models?
> > > I read so often that models are not available from vendors, now is
> your
> > > chance!
> > > But I need some help. Please!
> > >
> > > Bart
> > > www.yageo.com
>
> > > ================ original post
> ========================================
> > >
> > > Hi all,
> > >
> > > I am starting to develop Specctraquest models for three-terminal
> > > capacitors, e.g. a feedthru capacitor.
> > > Does anybody know how to extract, or how to determine, the mounted
> > > inductance of such a three-terminal capacitor?
> > > Maybe someone already did this?
> > >
> > > For a two terminal device the loop-inductance can be determined quite
> > > easily, but for a three terminal capacitor I could not find a single
> > > reference.
> > > Of course it all depends on substrate thickness, vias and layout, but
> is
> >
> > > there a general method how to calculate the mounted inductance or
> > > loop-inductance for such a device?
> > > Any information is welcome!
> > >
> > > regards,
> > > Bart Bouma
> > > www.yageo.com
> > >
>
> ------------------------------------------------------------------
> To unsubscribe from si-list:
> si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
>
> or to administer your membership from a web page, go to:
> //www.freelists.org/webpage/si-list
>
> For help:
> si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
>
> List archives are viewable at:
> //www.freelists.org/archives/si-list
> or at our remote archives:
> http://groups.yahoo.com/group/si-list/messages
> Old (prior to June 6, 2001) list archives are viewable at:
> http://www.qsl.net/wb6tpu
>
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List archives are viewable at:
//www.freelists.org/archives/si-list
or at our remote archives:
http://groups.yahoo.com/group/si-list/messages
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
Other related posts:
