[SI-LIST] Re: DC Resistance calculation in 2.5D solver

  • From: hanymhfahmy@xxxxxxxxx
  • To: bradb@xxxxxxxxxxx,cristian.gozzi@xxxxxxxxx,"'Tesla'" <emcesd@xxxxxxx>
  • Date: Thu, 17 May 2012 22:49:57 +0000

Hi Tesla. 
My experience for busses like DDR w SSTL technology that signals switch around 
VddQ/2 as a DC value, then extracted S-parameters from EM field solvers should 
solve @ DC operating VddQ/2 accurately as the signal shape due to reflections 
on the memory channel especially ISI effects r greatly dependent on accurate DC 
solution of the S-parameters. 

One of the big benefits of MOM from EEsof is that it solves the fields @ DC n 
not extrapolate to DC. 

U can also choose between RF modes n uwave mods while assuring accurate DC 
solution that is critical to such interfaces as memory DDR, GDDR n LPDDR where 
switching happens starting from a DC level such as VddQ/2 or VddQ. 

  Hope this helps. 

Hany Fahmy
Agilent Technologies. 
Sent via BlackBerry offered by Mobinil

-----Original Message-----
From: "Brad Brim" <bradb@xxxxxxxxxxx>
Sender: si-list-bounce@xxxxxxxxxxxxx
Date: Thu, 17 May 2012 11:25:55 
To: <cristian.gozzi@xxxxxxxxx>; 'Tesla'<emcesd@xxxxxxx>
Reply-To: bradb@xxxxxxxxxxx
Cc: <si-list@xxxxxxxxxxxxx>
Subject: [SI-LIST] Re: DC Resistance calculation in 2.5D solver

hello all,

Life is easier if the EM solver natively includes both AC and DC solutions
and applies the DC as an "anchor" for the frequency-swept results. This
capability is available in PowerSI.

AC solvers consider the full set of coupled Maxwell's equations for both RL
and GC effects, where in the DC case Maxwell's equations decouple into two
independent set of equations. To solve for RL you apply only the
magnetostatic equations. The DC solution is much easier and better
conditioned numerically. Typical AC solver formulations can not solve at
exactly DC and are ill conditioned numerically at "low" frequency. Consider
that in an AC solver you are simultaneously dealing with numerical values
proportional to (j*omega*L) and 1/(j*omega*C), or the inverse of each. As
you approach DC the numerics become ill conditioned as 1/(omega*omega). The
frequency at which the numerical instability becomes significant varies and
can be controlled to some extent by formulation and matrix solvers. This
lower frequency of stability is also dependent on the design being
simulated, the mesh size/quality and a few other factors not often under
user control. If you are not careful, then your low frequency results may
have significant error due to this instability. This behavior defines
conflicting goals. You want to simulate as low in frequency as possible to
enable better extrapolation to DC by a circuit simulator later, but you want
to stop at a high enough frequency to maintain accuracy in the AC
formulation. For each design the low frequency stability point can vary; and
it can vary with the mesh applied. For some simulators the low frequency
bound can be as high as 100MHz, for others 1kHz or even less.

A related but somewhat independent issue is the current distribution within
metals at low frequencies - as mentioned earlier in this thread. This can be
addressed in various solvers, but in all cases it implies a larger and more
complex solution. For 3D solvers it can be addressed with volumetric meshing
of metals - a pursuit for the brave of heart who have large compute
resources available. For planar (2.5D) solvers it is often accomplished by
considering independent currents on the top/bottom of a metal. For parallel
plate (2D) solvers it is often addressed by considering field penetration
through metals. All of these improve accuracy at low frequency, some waste
effort at high frequency, and some can even cause issues at high
frequencies. For example, in a 3D solver volumetric meshing is only required
for metals thinner than about 3 skin depths, so the current distribution
doesn't vary rapidly and a coarse mesh is often adequate. However, at high
frequencies the coarse mesh is not adequate to represent the rapid decay of
current in the metal. Therefore, it is often not adequate to use the same
mesh over a frequency range spanning bulk to skin current flow regions.

For capacitors (especially blocking caps) I would simulate the design with
extra port(s) and then insert the cap model in the subsequent circuit
simulation. It is likely your circuit simulator has more options to control
the frequency extrapolation for your cap model. To extrapolate cap vendor
S-parameters to low frequency you might want to use the special knowledge
you have - that it's a capacitor. A fairly simple RLC equivalent circuit
fitted to the data may work very well for you. Some people insist on using
S-parameters, assuming they are inherently more accurate than a lumped
model. However, you should consider the manufacturing variations that *will*
occur prior to adopting this assumption too firmly. If your cap S-parameters
are showing higher order resonances you may also want to ask yourself how
much variation these will experience when the device is applied in your
design vs the test board and mounting structure applied for the measurement.

best regards,
 -Brad
  Sigrity


> -----Original Message-----
> From: si-list-bounce@xxxxxxxxxxxxx 
> [mailto:si-list-bounce@xxxxxxxxxxxxx] On Behalf Of Cristian Gozzi
> Sent: Thursday, May 17, 2012 12:02 AM
> To: Tesla
> Cc: Istvan Novak; long.0.yang@xxxxxxxxx; weirsi@xxxxxxxxxx; 
> rtatikola@xxxxxxxxx; si-list@xxxxxxxxxxxxx
> Subject: [SI-LIST] Re: DC Resistance calculation in 2.5D solver
> 
> Hi Tesla
> with SIwave you can create multiple concatenated frequency sweeps.
> Since AC solver is not optimized for DC extraction, as I told 
> you it extrapolates DC value
> 
> to let this extrapolation works well, I suggest to create 
> first sweep with dense step points close to DC (let me say, 
> for instance from 0 Hz to 1 MHz, with 100 points for decade, 
> log sweep) than use another sweep from 1MHz to Fmax
> 
> NOTE: you can also try to compare DC value extracted from 
> S-parameter with this such sweep with same port setup and IR 
> simulation in SIwave and get the worth of its accuracy!
> 
> Be aware that usually this kind of non uniform frequency 
> sweep could be difficult for spice time domain solver, since 
> it use convolution method and usually is better to have dense 
> and uniform step frequency sweep
> 
> I'm not an expert of ADS, last time I used it was to many 
> years ago, so I cannot help you so much on that...
> 
> I'm using Nexxim, within Ansys DesignerSI, and there are 
> different options/algorithms to extrapolate DC point from an 
> S-parameters model that start from 1MHz
> 
> One suggestion, I understood that the purpose of SI -list is 
> to drive users on specific SI & PI application challenges 
> that it may involve simulators, measurement or both
> 
> but if you want to talk deeply inside specific software 
> feature, setup and tricks I strongly recommend you to contact 
> your local reference AE from Ansys, Sigrity and Agilent
> 
> all of them have a great AE team to support and assist you on 
> all these topics
> 
> if you need more info, you can contact me off-line
> 
> Regards
> 
> Cristian
> SI & PI Specialist
> Technoprobe
> 
> 2012/5/17 Tesla <emcesd@xxxxxxx>
> 
> > Hi,
> >
> > Thanks for your kindly advice.
> > If i want to get the S-Parameter of DC to Fmax for time domain 
> > simulation from some 2.5 field solver. Could i use the 1. IR drop 
> > calculated value(DC) and FEM high frequecny(1M to 1GHz) to 
> combine a 
> > DC to 1GHz S parameter.
> > 2 Use the FEM high frequecny(1M to 1GHz) only, let the time-domain 
> > simulator do the extrapolation work.
> >
> > it may be a old and discussed many times topic.
> >
> > i do a little experiment in ADS of Agilent, It seems that 
> ADS did not 
> > do very well in low frequency extrapolation for S-parameter(Maybe i 
> > miss something).
> > I use the S-Parameter of a 0.1uF capacitor from vendor. 
> Suppose it has 
> > 0.3MHz to 1GHz data of S-Parameter. Then i use the dataset 
> of vendor 
> > to calculate S-Parameter of DC to 1GHz. The simulator simply give 
> > results that the value below 0.3MHz all equal to the value 
> at 0.3MHz. 
> > obviously the result do not make sense.
> >
> > Are simulator not suitable for S-parameter extrapolation?
> >
> > Thanks.
> >
> >
> >
> >
> >
> >
> > At 2012-05-16 20:40:41,"Istvan Novak" 
> <istvan.novak@xxxxxxxxxxx> wrote:
> > >Hi,
> > >
> > >I wont comment on the specific tools, but will give you 
> some generic ideas.
> > >
> > >We many times dont realize that DC resistance calculations can be 
> > >almost as tricky as the high-frequency computations.  We usually 
> > >assume
> > >(wrongly) that at DC the current density is uniform in the 
> conductor 
> > >cross section, but except of a few hypothetical cases, it is not.  
> > >This is why, even at DC, the correct answer needs careful volume 
> > >meshing, to make sure that the different current density 
> values are 
> > >captured properly throughout the conductor volume, 
> including the end connections
> > >leading to the observation points.   Most tools have knobs 
> for you to
> > >turn on some of the key parameters (you may be surprised 
> to see that 
> > >once you start turning those knobs, you get different answers from 
> > >the same tool to the same DUT).
> > >
> > >So when you compare results from different tools, you may 
> want to check:
> > >- how the meshing is done and get them as close to be 
> similar/same as 
> > >possible
> > >- how the connection is assumed
> > >
> > >For this second item, one hint: we can not use point connection, 
> > >because for zero cross section area the current density and the 
> > >equivalent resistance would be infinite: we have to use finite 
> > >connecting cross section area.  So first you have to find 
> out how the 
> > >connections are assumed in the two tools and then make 
> sure that they 
> > >are as close/similar as possible.  If these key elements 
> are the same 
> > >or close, we can then expect similar results.
> > >
> > >Regards,
> > >
> > >Istvan Novak
> > >Oracle
> > >
> > >
> > >On 5/16/2012 7:52 AM, Tesla wrote:
> > >> Hi, Experts
> > >> In 2.5 field solver(eg: Sigrity or SIwave), if i want to 
> get DC resistance of interconnect, i use the two method:
> > >>
> > >> 1 Use FEM to calcute from DC to Fmax Hz, use the DC s 
> parameter to 
> > >> get the S parameter
> > >>
> > >> 2 Use IR drop in the analysis to get the DC resistance
> > >>
> > >> but the two method give two different DC resistance 
> value, Which one i should trust?
> > >>
> > >> Thanks.
> > >>
> > >

------------------------------------------------------------------
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 forum  is accessible at:
               http://tech.groups.yahoo.com/group/si-list

List archives are viewable at:     
                //www.freelists.org/archives/si-list
 
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 forum  is accessible at:
               http://tech.groups.yahoo.com/group/si-list

List archives are viewable at:     
                //www.freelists.org/archives/si-list
 
Old (prior to June 6, 2001) list archives are viewable at:
                http://www.qsl.net/wb6tpu
  

Other related posts: