[SI-LIST] Re: surface roughness
- From: Scott McMorrow <scott@xxxxxxxxxxxxx>
- To: "Loyer, Jeff" <jeff.loyer@xxxxxxxxx>
- Date: Tue, 29 Nov 2011 19:06:09 -0500
Jeff
I will grant you that a physics based model will give you insight into
the microscopic properties of the material. However, your company is
one of the few in the Known Universe that has the ability to impact
change at this level with the material and fabrication suppliers.
99.999% of us do not have that clout, or the ability to perform the
necessary measurements to characterize the copper. Thus, we need to
either obtain the information from the suppliers, which as we all know
is almost guaranteed to be wrong, or we need to perform a fit to
measured macroscopic data.
regards,
Scott
Scott McMorrow
Teraspeed Consulting Group LLC
121 North River Drive
Narragansett, RI 02882
(401) 284-1827 Business
(401) 284-1840 Fax
http://www.teraspeed.com
Teraspeed® is the registered service mark of
Teraspeed Consulting Group LLC
On 11/29/2011 6:44 PM, Scott McMorrow wrote:
> Answers
>
> Scott McMorrow
> Teraspeed Consulting Group LLC
> 121 North River Drive
> Narragansett, RI 02882
> (401) 284-1827 Business
> (401) 284-1840 Fax
>
> http://www.teraspeed.com
>
> Teraspeed® is the registered service mark of
> Teraspeed Consulting Group LLC
>
>
> On 11/29/2011 4:00 PM, Loyer, Jeff wrote:
>> Hi Scott,
>> Can you help me clarify where the disconnect is? Here are my
>> thoughts on the questions raised by this thread, and some request for
>> clarification:
>>
>> * It seems you are using a model based on empirical measurements of
>> the trace(s) insertion loss. You allude to separating the dielectric
>> and conductor losses, though I'm not sure what method you're using.
>> Can you clarify exactly what you're proposing?
> Fitting to de-embedded GMS-parameters as we have published with Yuriy
> Shlepnev of Simberian in multiple papers.
>>
>> * You imply that, for your "just fabricated package substrate", you
>> have no information on the surface profile, thus must be tweaking
>> some variables until your model matches measurement. (Am I missing
>> something?) Do you have any more information on your dielectric
>> properties, or are those adjusted to fit measured data also?
> How would I obtain "more data". I have de-embedded measurements to 40
> GHz of typical trace structures.
>>
>> * I agree that in many cases (perhaps most, if not all, those you
>> encounter) the Hammerstadt-type analysis can work. But, it's not
>> based on physics, and gives you no real insight into what is causing
>> the increase in insertion loss. In many cases, merely increasing the
>> loss tangent of the dielectric will give you an adequate fit to
>> measured data. What's the advantage of one over the other
>> (Hammerstadt vs. Df adjustment vs. some other "fudge factor")?
> Who cares whether it is based on physics. We are talking about
> macroscopic behavior, not microscopic behavior. Averaging over
> minuted fractions of a wavelength is a wonderful thing. As for the
> "it gives no real insight" that's just silly. The insight it provides
> is that the losses are due to surface roughness.
>
> Loss tangent will most definitely NOT give you a correct fit to
> measured data, since phase will be incorrect. If we are using a causal
> model, such as Djordjevic-Sarkar, as dielectric loss increases,
> effective delay decreases. This is contrary to what actually
> happens. If you are looking to fit broadband models to measured
> results, phase needs to also be correct. Thus, the dielectric model
> must be correct. This is why I Have been asking about phase. Not for
> the copper modeling, but to be sure that the baseline dielectric
> models correctly account for phase prior to comparison of the surface
> roughness models.
>>
>> * I agree that, if you aren't interested in the physics, a
>> Hammerstadt-type model should be adequate for most modeling. But, as
>> has been demonstrated, there are cases where it can't be made to
>> work. Perhaps the disconnect comes because you are working w/
>> smoother copper than other folks have to contend with. For instance,
>> most domestic (U.S.A.) vendors use RTF as a matter of course, which
>> is significantly smoother than the standard copper commonly used in
>> Asia. Might this be a significant difference in our experiences?
> I've had zero issue in using a Hammerstadt-type model, for any work
> I've done, domestic or asian. However, for the cases where the copper
> is that rough, then pretty much all bets are off for any sort of high
> performance work, anyway. Just the statistical variation of the
> copper is going to be huge.
>>
>> I'm going to refrain from commenting on the dielectric issues, since
>> I don't believe those are germane to this particular discussion.
>> They may be worthy of a thread in themselves.
> They absolutely are. Unless the dielectric is modeled correctly, you
> cannot be sure that the adjustments due to surface roughness are correct.
>>
>> Personally, I got involved in the work of Professor Huray, Femi, and
>> others because we were searching for a means of reducing insertion
>> loss due to copper roughness and needed to understand the physics
>> behind the loss mechanism in order to be able to do that. For
>> instance, there were some hints that roughness in one axis might give
>> less loss than in another (which turned out to be true).
>> Unfortunately, I've concluded (so far) that the same features that
>> give you the best mechanical strength (maximum surface area) give you
>> the worst loss. I don't see any cost-effective adjustments to the
>> copper roughness that would maintain mechanical strength while
>> reducing loss. We will have to use smoother copper to reduce losses,
>> and find other methods of maintaining mechanical strength w/ that
>> smooth copper, if necessary. (I'd love to hear others' opinions on that)
>>
>> On the other hand, it hasn't been wasted effort. The Huray model has
>> built up my engineering intuition into what copper texture features
>> would contribute most to loss, thus giving me some ability to predict
>> effects when I look at a cross section. And now I know more about
>> what I didn't know.
>>
>> As I've said before, I believe we are just beginning to understand
>> all the aspects of copper roughness; I appreciate you sharing your
>> insights.
> Interesting that you don't have the same curiosity regarding
> dielectrics. Copper is actually a much simpler problem.
>>
>> Jeff Loyer
>>
>>
>> -----Original Message-----
>> From: si-list-bounce@xxxxxxxxxxxxx
>> [mailto:si-list-bounce@xxxxxxxxxxxxx] On Behalf Of Scott McMorrow
>> Sent: Monday, November 28, 2011 4:38 PM
>> To: Oluwafemi, Olufemi B; silist
>> Subject: [SI-LIST] Re: surface roughness
>>
>> Femi
>>
>> So, what would be the huray model parameters for my just fabricated
>> package substrate, for which I have no reliable information on the
>> surface profile? In all the cases I have run across thus far, I can
>> use a surface impedance condition described by a Hammerstadt-type of
>> surface roughness coefficient and adjust the fit to de-embedded
>> measured data.
>>
>> I wonder why we are not all worried about physics-based dielectric
>> models that incorporate temperature and humidity, also? My comment
>> regarding phase had little to do with surface roughness. It was
>> directed towards accurate modeling of the processed dielectric
>> properties prior to testing of the Huray model fit.
>>
>> best regards
>>
>> Scott
>>
>> Scott McMorrow
>> Teraspeed Consulting Group LLC
>> 121 North River Drive
>> Narragansett, RI 02882
>> (401) 284-1827 Business
>> (401) 284-1840 Fax
>>
>> http://www.teraspeed.com
>>
>> Teraspeed(r) is the registered service mark of Teraspeed Consulting
>> Group LLC
>>
>>
>> On 11/28/2011 1:34 PM, Oluwafemi, Olufemi B wrote:
>>> All,
>>>
>>> Maybe it's going to take adding the alphabets A thru z as
>>> coefficients behind Hammerstad equation, before the SI community
>>> realizes that the equation is fundamentally flawed.
>>>
>>> Here are some reasons why the equation is flawed:
>>>
>>> 1) It is an empirical fit to Morgan's data that was published in 1948.
>>> There is no physics to it. "Effect of surface roughness on eddy
>>> current sources at microwave frequencies" Journal of Applied Physics
>>> V.20 pp 352-362, 1948
>>>
>>> 2) The ratio you get from the equation is going to saturate at 2; no
>>> matter how rough the copper is.
>>>
>>> 3) The only input to Hammerstad equation is rms value. As you all
>>> know, it is possible to have a rough and relatively smooth copper
>>> have the same rms value. Thus Hammerstad equation will give you the
>>> same ratio for both.
>>>
>>> Apart from providing an equation to use, what Huray et all presented
>>> at DesignCon 2010, was to explain the physics of surface roughness
>>> from an EM perspective.
>>> The presentation also debunked the myth that says, current flows in
>>> and out of the rough structures.
>>>
>>> As a community, I think we need to start going back to the physics
>>> of the problem, rather than, rules of thumb or equations that have
>>> been passed down that might not work. We can't keep patching...
>>> I'm not trying to convince anyone to use the Huray Model, but to
>>> look at the physics of the problem.
>>>
>>> By the way the Huray model correlates in phase as well. Please check
>>> pg 83-84 'Surface Roughness and Its Impact on System Power Losses'
>>> Olufemi Oluwafemi, ProQuest, Ann Arbor , MI.
>>>
>>> My two cents,
>>>
>>> Thanks,
>>>
>>> Femi Oluwafemi
>>>
>>> -----Original Message-----
>>> From: si-list-bounce@xxxxxxxxxxxxx
>>> [mailto:si-list-bounce@xxxxxxxxxxxxx] On Behalf Of Scott McMorrow
>>> Sent: Tuesday, November 22, 2011 4:32 PM
>>> To: si-list@xxxxxxxxxxxxx
>>> Subject: [SI-LIST] Re: surface roughness
>>>
>>> All
>>>
>>> I'd like to be clear. Any simulation model for dielectric and copper
>>> surface roughness modeling must be a good match to measured data in
>>> amplitude and PHASE to be an accurate model. It is clear to me that
>>> the papers presented on the Huray method have thus far not shown any
>>> phase correlation data.
>>>
>>> regards,
>>>
>>> Scott
>>>
>>> Scott McMorrow
>>> Teraspeed Consulting Group LLC
>>> 121 North River Drive
>>> Narragansett, RI 02882
>>> (401) 284-1827 Business
>>> (401) 284-1840 Fax
>>>
>>> http://www.teraspeed.com
>>>
>>> Teraspeed(r) is the registered service mark of Teraspeed Consulting
>>> Group LLC
>>>
>>>
>>> On 11/20/2011 1:59 PM, Yuriy Shlepnev wrote:
>>>> Hi Naga,
>>>>
>>>> As Jeff mentioned the theory of conductor roughness is still evolving
>>>> research area and there is a lot to do here. However, this may be
>>>> important only for the completeness of the theories. For practical
>>>> purpose we can use heuristic models that capture the macroscopic
>>>> observable effect of the roughness (similar to how we characterize
>>>> dielectrics). Practically any roughness theory produces a roughness
>>>> correction coefficient, that can be used to adjust surface impedance
>>>> of a conductor surface locally for accurate electromagnetic analysis
>>>> or to adjust skin-effect matrix extracted with a static field solver
>>>> (in case if less accurate static field solver is used for the
>>>> analysis). In both cases the parameters of a roughness model can be
>>>> identified by fitting the parameters of the correction coefficients
>>>> to measured data (reflection-less generalized modal S-parameters are
>>>> best suitable for such fitting). See a brief overview of roughness
>>>> correction coefficients in Y. Shlepnev, C. Nwachukwu, "Roughness
>>>> characterization for interconnect analysis". - Proc. of the 2011 IEEE
>>>> International Symposium on Electromagnetic Compatibility, Long Beach,
>>>> CA, USA, August, 2011, p. 518-523
>>>> - paper is available at
>>>> http://ieeexplore.ieee.org/search/freesrchabstract.jsp?tp=&arnumber=6
>>>> 038367&
>>>> openedRefinements%3D*%26filter%3DAND%28NOT%284283010803%29%29%26searc
>>>> hField%
>>>> 3DSearch+All%26queryText%3Dshlepnev
>>>> Sorry for the long link - you can just search for authors names at
>>>> ieeeexplore.org, and paper and presentation are also available at
>>>> www.kb.simberian.com (in publications and presentations sections,
>>>> registration is required).
>>>> This paper proposes a simple 2-parameter Modified Hammerstad
>>>> Correction Coefficient (MHCC) as an extension of 1-parameter
>>>> Hammerstad model used in microwave IC applications for decades. The
>>>> paper shows how to identify 2 parameters of MHCC by fitting
>>>> GMS-parameters, assuming that the parameters of a broadband
>>>> dielectric model are identified separately. The model shows excellent
>>>> correlation with measurements up to 50 GHz. Parameters in Huray's
>>>> snowball model can be also identified with GMS-parameters with
>>>> minimal knowledge about the surface structure - no expensive
>>>> conductor surface investigation is required for such identification.
>>>> Note, that the identified Huray's model produces practically the same
>>>> results as MHCC - we will show that in our oncoming paper at
>>>> DesignCon 2013. Both models are capturing well the effect of
>>>> additional absorption by surface due to increase of the surface area.
>>>>
>>>> Considering the roughness effect on PCB or packaging interconnects, I
>>>> can draw a parallel here with the wideband Debye dielectric model
>>>> (also known as Djordjevic-Sarkar or as Swenson-Dermer) that produces
>>>> very good correlation in analysis of interconnects in PCB and
>>>> packaging applications over ultra-wide frequency band. The model
>>>> describes a dielectric with a continuous spectrum over a wide
>>>> frequency band. It correlates well with the measured data and is
>>>> used in practically all SI software nowadays. Though, as engineers we
>>>> do not care much about the internal structure of the composite
>>>> dielectrics that produces behavior captured by the model - all we
>>>> need is 2 or 3 model parameters that characterize a particular
>>>> dielectric. A multi-pole Debye model has more parameters can be used
>>>> to describe almost any dielectric used in PCB/Packaging and we do not
>>>> care what structure produces those poles - all we need are just
>>>> values of the poles and residues that produce good correlation
>>>> within the spectrum bandwidth of our signal.
>>>>
>>>> Considering the effect of roughness on degradation of signal in
>>>> PCB/packaging interconnects, rough surface leads to increase of
>>>> attenuation (insertion loss) and increase of inductance at high
>>>> frequencies due to simple increase of the total surface of the
>>>> conductor. Both MHCC or Huray's snowball models capture this increase
>>>> in loss and inductance if applied appropriately in the
>>>> electromagnetic analysis. Conductor surface impedance adjustment for
>>>> instance produces causal models that take into account current
>>>> distribution within a conductor due to roughness on just one side of
>>>> plane or strip or due to additional plating layers for instance. Note
>>>> that both low and high profile copper roughness lead to a substantial
>>>> increase in attenuation above 10 GHz. In addition roughness also
>>>> increases the capacitance of conductor - that is clearly visible in
>>>> cases with large roughness and thin dielectric layers - see the paper
>>>> and presentation cited above. The capacitive effect of roughness due
>>>> to surface singularity is practically frequency-independent and was
>>>> reported in IC applications a few yeas ago - see for instance this
>>>> paper and references there: A. Albina at al., Impact of the
>>>> surface roughness on the electrical capacitance, Microelectron. J.
>>>> 37 (2006) 752-758.
>>>>
>>>> Finally, if you need a software that correlates with the measured
>>>> data up to
>>>> 50 GHz, it has to have at least wideband and multi-pole Debye models
>>>> for dielectrics in addition a selection of advanced models for
>>>> roughness (MHCC or Huray's snowball models for instance) that can be
>>>> easily identified without expensive investigation of the surface
>>>> structure. Classical Hammerstad model has low accuracy and not
>>>> usable above 3-5 GHz in general.
>>>> The software has to be validated with measurement over the bandwidth
>>>> of your signal. Though, the software is only a part of the story
>>>> here. In addition you have to establish a procedure to identify the
>>>> parameters of both dielectric and conductor models. Without the
>>>> rigorous material parameters identification it is "garbage in -
>>>> garbage out" no matter how advanced is the analysis. Models that
>>>> describe dielectrics and roughness over a wide frequency band from
>>>> MHz to 50 GHz are typically not available from manufacturers at this
>>>> time and have to be identified. Companies like Teraspeed Consulting
>>>> Group for instance can help you to establish such material
>>>> identification procedure.
>>>>
>>>> Best regards,
>>>> Yuriy Shlepnev
>>>> www.simberian.com
>>>>
>>>>
>>>> -----Original Message-----
>>>> From: si-list-bounce@xxxxxxxxxxxxx
>>>> [mailto:si-list-bounce@xxxxxxxxxxxxx] On Behalf Of
>>>> nagachander.sing@xxxxxxxxxxxxxxxxxxxx
>>>> Sent: Friday, November 18, 2011 11:35 PM
>>>> To: si-list
>>>> Subject: [SI-LIST] surface roughness
>>>>
>>>> Hi all,
>>>> What is the best tool to simulate surface roughness when simulating
>>>> for the differential S-parameters for striplines and microstripline
>>>> in IC packages?
>>>>
>>>> I am interested in simulating till 50GHz.
>>>>
>>>> What kind of effects are you seeing with surface roughness? i mean
>>>> things like does your insertion loss increase? what abt diff return
>>>> loss?
>>>>
>>>> Any good papers to understand the problems in detail...
>>>>
>>>> Any feedback will be appreciated
>>>>
>>>> Thanks,
>>>> Naga
>>>>
>>>> ------------------------------------------------------------------
>>>> 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 technical documents are available at:
>>>> http://www.si-list.net
>>>>
>>>> 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 technical documents are available at:
>>>> http://www.si-list.net
>>>>
>>>> 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 technical documents are available at:
>>> http://www.si-list.net
>>>
>>> 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 technical documents are available at:
>>> http://www.si-list.net
>>>
>>> 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 technical documents are available at:
>> http://www.si-list.net
>>
>> 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 technical documents are available at:
http://www.si-list.net
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:
