[SI-LIST] Re: SSO pushout, ground bounce definition

  • From: Larry Smith <ldsmith@xxxxxxxxxxxxxxxxxx>
  • To: ldsmith@xxxxxxxxxxxxxxxxxx, si-list@xxxxxxxxxxxxx,Pat_Diao@xxxxxxxx
  • Date: Fri, 19 Apr 2002 09:30:49 -0700 (PDT)

Pat - Different EM extractors use different techniques for finding
partial inductance and mutuals.  A long time ago in their 3D extractor,
Ansoft used to sum up the energy in all of the tetrahedron when 1 amp
is forced through a conductor.  Energy is half of inductance times the
current squared, W=L*I^2/2.  From the energy and 1 amp, they found
inductance.  I think the equations that you have quoted below are
equivalent, but perhaps easier computationally.

Lot's of 2d extractors find the capacitance between structures and
calculate inductance from velocity=sqrt(1/LC).  Velocity is known from
the dielectric constant and the speed of light.  As you can see, these
extractors do not calculate L directly but infer it from some other

Other 3d extractors descritize conductors up into lots of little
rectangular bars.  Grover published inductances for simple bar
geometries long ago, I think the book has been out of print for more
than 50 years.  Fundamentally, these inductance extractors are based on
Grover's calculations for inductance.

There are probably a dozen different methods for EM extraction.  To
answer your question, yes the several different techniques may obtain
different partial inductances for the same geometry problem.  The
example I gave several emails ago was a bit extreme, I don't think the
extracted partial inductances will differ by a factor of 10.  But even
if they did, they could all produce the same loop inductance and
therefor the same results in a circuit simulator.

Larry Smith
Sun Microsystems

> From: Pat Diao <Pat_Diao@xxxxxxxx>
> To: "'Larry Smith'" <ldsmith@xxxxxxxxxxxxxxxxxx>, si-list@xxxxxxxxxxxxx
> Subject: RE: [SI-LIST] Re: SSO pushout, ground bounce definition
> Date: Wed, 17 Apr 2002 17:19:53 -0700
> MIME-Version: 1.0
> Hi Larry,
> Thanks for the message.  The issue regarding inductance is becoming clear in
> definition, but still confusing in application.  I found the following in
> Ansoft tech note:
> "AC Inductance: 
> All current in AC analysis are surface currents.  To take this into account,
> the system computes the surface magnetic field, H, such that 
> H = ...
> H dot n =0
> At high frequencies, the magnetic field is tangential to the surface of a
> good conductor. After solving for the magnetic field, the system computes
> the surface current density K:
> K = n x H
> and the inductance is found using the relationship:
> L = (integral over S)(A dot K dV) "
> I think the calculation is sound for partial L. Frankly this is the only way
> I can see to calculate the partial L. Then it come back to the question why
> different extractors may come up with different partial L?  Do they use
> radically different ways to figure out Partial L?
> Regards,
> Pat
> -----Original Message-----
> From: Larry Smith [mailto:ldsmith@xxxxxxxxxxxxxxxxxx]
> Sent: Wednesday, April 17, 2002 9:46 AM
> To: ldsmith@xxxxxxxxxxxxxxxxxx; si-list@xxxxxxxxxxxxx; Pat_Diao@xxxxxxxx
> Subject: Re: [SI-LIST] Re: SSO pushout, ground bounce definition
> Pat - thanks for the comment.
> With almost every electromagnetic extractor you have to be thinking
> about where the return current is when you set up the problem.  Where
> is the reference node?  This is especially true with Ansoft as they
> make it very easy for you to magically source current somewhere in the
> problem and then magically sink current somewhere else.  Unless you try
> really hard to complete the return path, you will obtain a partial
> inductance with all the associated problems discussed in previous
> emails.  
> To get valid circuit simulation results, you have to apply Vdd and
> Ground currents to the ports that you assumed when set up the
> extraction problem.  Said another way, before you run the extraction
> problem, you have to anticipate where your return currents will flow in
> your simulation.  I have seen many cases where SI engineers do Ansoft
> extraction for an SSN problem and the only nodes present are signals
> and grounds.  If they leave out the Vdd path, they have left out half
> of the return current path.  ...On the bright side, their simulation
> results will be within 50% of being correct.
> It has been several years since I actively worked with Ansoft, but I
> often advise engineers that do.  For power distribution analysis, I
> usually tell them to short the Vdd and ground structures on one side of
> the problem, probably at the PCB power planes.  Then, force and sink
> current in close proximity to each other, probably on some chip or
> chip-connect structures.  That way, current is forced to run in a
> loop.  Take the loop inductance number that you get from that
> extraction, divide by 2 and call it the "half loop inductance".  In the
> circuit simulation, assign a half loop inductance to Vdd and another
> half loop inductance to ground.  I have no idea what the partial or
> mutual inductances are for that problem, nor do I care.  They don't
> mean anything anyway.  
> I see a lot of guys draw up the ground structures and extract a partial
> inductance.  Then they draw up the Vdd structures and extract a partial
> inductance.  Then they combine the the two partial inductances in the
> same circuit simulation problem (with no mutuals) and obtain very
> incorrect results.  To get correct results, you have to set up the
> extraction problem the way you are going to simulate it and simulate
> the extracted matrix the way you set it up in extraction.  User
> beware!
> For the SSN problem, you almost have to draw the Vdd, Ground and signal
> structures all at the same time.  You then need input and output ports
> for each power, ground and signal structure.  The path that return
> current takes through the structure depends greatly upon the data
> patterns given to the drivers in circuit simulation.
> Your second question has to do with the sign polarity on the mutual
> inductance.  Many extractors will place a minus sign in front of the
> mutual inductance, in that case, you want to add it.  This question is
> closely related to where you put the dots on a transformer.  From a
> physical perspective, if currents are running parallel (the same
> direction) as each other, the flux and mutuals are going to add to the
> loop inductance.  If currents are running opposite directions to each
> other, the flux and mutuals are going to subtract from the loop
> inductance.
> regards,
> Larry Smith
> Sun Microsystems
> > Delivered-To: si-list@xxxxxxxxxxxxx
> > From: Pat Diao <Pat_Diao@xxxxxxxx>
> > To: "'ldsmith@xxxxxxxxxxxxxxxxxx'" <ldsmith@xxxxxxxxxxxxxxxxxx>, 
> si-list@xxxxxxxxxxxxx
> > Subject: [SI-LIST] Re: SSO pushout, ground bounce definition
> > Date: Tue, 16 Apr 2002 16:51:09 -0700
> > MIME-Version: 1.0
> > Content-Transfer-Encoding: 8bit
> > X-archive-position: 2675
> > X-ecartis-version: Ecartis v1.0.0
> > X-original-sender: Pat_Diao@xxxxxxxx
> > X-list: si-list
> > 
> > 
> > Hi Larry,
> > 
> > Very nice writing about the partial and loop inductance.  I agree that a
> > partial inductance is meaningless without defining the loop.  However,
> there
> > are two practical questions remaining:
> > 
> > 1.  Most of the RLC extractors, including Ansoft Q3D which I use, only
> > require users to define the source and sink locations on the net.  You can
> > draw a ground plane but it is not required, and sometimes there is no
> ground
> > plane nearby.  It then calculates the inductance of the net, which is of
> > course the "partial inductance".  But the question is, where is the other
> > portion of the "loop"?  There could be a hundred different ways to define
> > the other portion of the loop.  Then will we have a hundred different
> > inductance values for the net? 
> > 
> > 2.  On the loop inductance, there is no question about 
> > L(loop) = L1 + L2 - 2*M12
> > But occasionally one can see 
> > L(loop) = L1 + L2 + 2*M12 
> > Is this a totally wrong equation or the mutual inductance can be of
> > different signs?
> > 
> > Thanks,
> > Pat 
> > 
> > 
> > 
> > -----Original Message-----
> > From: Larry Smith [mailto:ldsmith@xxxxxxxxxxxxxxxxxx]
> > Sent: Tuesday, April 16, 2002 1:56 PM
> > To: si-list@xxxxxxxxxxxxx
> > Subject: [SI-LIST] Re: SSO pushout, ground bounce definition
> > 
> > 
> > 
> > Hmmm.  Looks like my table separated by tabs did not make it... 
> > I'll try it again.  The table in the text has been replaced.
> > 
> > ------------- Begin Forwarded Message -------------
> > 
> > Ege - I don't believe there is any contridiction between non-unique
> > inductance matricies and unique system voltages.  Partial inductances
> > are valid as long as they are defined in the context of a loop.
> > Partial inductances by themselves have no meaning.  But in an
> > inductance matrix, where they are combined with all partials and
> > mutuals in a loop, the extracted values (not unique) can be used to
> > find unique voltages and currents within that loop.
> > 
> > Take the example of two connector pins, one used for a signal and the
> > other used for ground.  Loop inductance in this case is L1+L2-2*M12
> > (the sum of the self inductances minus two times the mutual inductance
> > between them).  The following partial and mutual inductances might be
> > obtained from two different electromagnetic extractors:
> > 
> > extractor   L1      L2      M12     Loop L   units
> > ---------   -----   -----   -----   ------   -----
> > 1           10      10      9        2       nH
> > 2           2       2       1        2       nH
> > 
> > Which extractor got the right answer?  They both did.  Partial and
> > mutual inductances are meaningless outside the context of a loop.  Both
> > sets of extracted data give the same loop inductances and will lead to
> > unique voltages and currents when the loop is simulated in a circuit
> > analysis tool, even though the extracted partial and mutual inductances
> > are not unique. =20
> > 
> > I see a lot of engineers try to use partial inductance (alone) in a
> > circuit simulation.  This is at best misleading and may give flat out
> > wrong answers in the simulation.  I don't think you will find any
> > disagreement between Al Ruehli, Bryan Young or any of the other experts
> > in this matter (please comment if I have misspoken).
> > 
> > Now concerning measurements, simulation, SI and EMI...  For SI
> > purposes, it only makes sense to measure voltages in very local areas
> > with a probe that has a very short ground lead.  When we measure a
> > signal with respect to (WRT) local ground, we have measured a
> > "difference" voltage.  This is a differential measurement, even if it
> > is done with a single ended probe.  We have not measured the signal WRT
> > spice node 0, the center of the earth, or any other place in the
> > system.  If a simulation is set up correctly, the inductance matrix
> > from either extractor 1 or 2 above will deliver an inductance matrix
> > that can be simulated and match the hardware measurement of a signal
> > WRT it's local ground.
> > 
> > As Raymond Chen has stated in points one and two below, it does not
> > make sense to measure anything other voltage WRT local ground.  There
> > are at least two problems with measurements that are not made WRT local
> > ground.  If we try to measure across several inches (significant
> > portion of a wavelength) we are trying to measure across a time delay.
> > What meaning is there in voltage measured across time?  If we try to
> > measure across a big inductance (i.e. connector pin), magnetic flux
> > will penetrate the loop involving the inductance, probe and ground
> > lead.  The size of the measurement loop will determine the magnetic
> > flux "captured" by that loop and lead to non unique measurements which
> > depend on where the ground lead is positioned WRT the inductance.  With
> > probes, the only legitimate thing to measure is the difference between
> > a node and it's local reference point, hopefully in an area where the
> > time varying magnetic field is not significant.
> > 
> > Even though we don't have a good way to measure "ground bounce" across
> > a connector, I believe a circuit simulator with the inductance matrix
> > from either extractor 1 or 2 above is capable of simulating the correct
> > voltage across the ground pin of a connector (and getting an identical
> > voltage solution from either matrix).  If 10 signals crossed the
> > connector and all switched the same way at the same time, there would
> > be a huge (almost Vdd) voltage across the single ground pin.  This is
> > real and is what we call the SSN or SSO problem.  It can cause flipped
> > bits and EMI radiation as one local ground gets perturbed WRT to the
> > other.  Simulations of "ground bounce" are probably more trustworthy
> > than measurement, assuming that all partial inductances and mutual
> > inductances are given in the context of a loop.  It is difficult to
> > measure ground bounce directly.  The best indicators of ground bounce
> > are funny waveforms at the far end of a quiet line and EMI radiation.
> > 
> > regards,
> > Larry Smith
> > Sun Microsystems
> > 

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:

For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field

List archives are viewable at:     
or at our remote archives:
Old (prior to June 6, 2001) list archives are viewable at:

Other related posts: