[SI-LIST] Re: Best PDS impedance for package damping

  • From: steve weir <weirsp@xxxxxxxxxx>
  • To: Larry Smith <Larry.Smith@xxxxxxx>
  • Date: Fri, 13 Feb 2004 12:52:19 -0800

Larry, since we can do it, so can the IC mfg's.  I don't really care if 
they fail to appreciate the genius of Archimedes, if they need something 
other than R => 0, I feel very strongly it is their obligation to not only 
specify it but to insure that they remain within a reasonably bounded range 
as they adjust their manufacturing.  As a customer I find it totally 
unreasonable that I should have to potentially requalify my PDS everytime 
the IC manufacturer does something with their die or package.

I agree that you can perform a sweep on boards to check go/no-go.  However, 
I don't really know the value of a statistical sample unless we are 
guaranteed a test sample each time any of the reels are reloaded.  And if 
the board doesn't work, then it gets really ugly.  Do I scrap an assembly 
possibly costing thousands of dollars?  Or do I go through the expense of 
removing all of the decoupling caps and start over?  It isn't pretty.

A colleague of mine had once suggested dummy sites for each value 
capacitor.  That fell apart when mfg pointed out the problem with reel 
changes.  One compromise might be to run two assemblies, one that is just 
the capacitor coupon, and one the real boards.  Anytime a reel change 
occurs one of the coupon boards would have to be run.  I have never pursued 
this to see how much the CM's would balk.


At 12:37 PM 2/13/2004 -0800, Larry Smith wrote:
>Steve - I completely agree that IC manufacturers should tell the board 
>integrators about the inductance, capacitance and the transient current 
>properties of their product.  Unfortunately, they usually don't know.  It 
>is easy to get capacitance information on a few gates or inductance 
>information on a few pins or vias.  But it is difficult to obtain the 
>aggregate (equivalent) inductance and capacitance that appears to the PCB 
>PDS.  Transient current is often more difficult than inductance or 
>capacitance and depends on the code that is running at the time.
>Another way to get this information is to measure it yourself.  By 
>mounting a processor or major ASIC on a bare fab, it is not difficult to 
>find the equivalent inductance, capacitance and resistance of the 
>component with a VNA S21 measurement.  Istvan and I have documented the 
>techniques for doing this in various papers.  Be sure and bias the 
>component with VDD because the capacitance across the device's power 
>terminals changes greatly between zero bias and normal bias.  By doing 
>this, you have measured the chip/package resonance frequency (sometimes 
>called cut-off) for the component on your board.  This gives you great 
>insight into what you need to do at the PCB level to provide power to the 
>part.  It also tells you what you cannot do (the frequency beyond which 
>the packaged chip is on it's own).
>Your other issue has to do with the ability to verify that the PCB 
>assembly house has actually placed the decoupling capacitors you 
>specified.  This is a tough one.  One sure test is to measure the 
>impedance vs frequency of a sample of assembled boards and see if it meets 
>the impedance that it was designed to meet.  You cannot tell if individual 
>components are correct but you can tell if the PDS is going to be able to 
>meet the power requirements of the power consumers, which is what you 
>really wanted to know anyway.
>Larry Smith
>Sun Microsystems
>steve weir wrote:
>>Larry agreed, the problem is that the IC mfgs don't tell us what their 
>>networks look like and the resonant frequency shifts with the external L, 
>>which in this case is the mounted capacitor array.  So, we are stuck 
>>shooting in the dark.  So, do we throw a bunch of cost at a problem that 
>>may or may not be there, and where the cost may or may not yield a 
>>satisfactory solution?
>>As to the significant work that you have done on closely spaced 
>>resonances I have followed this with great interest, but have some 
>>exceptions to it, most notably the difficulty auditing the manufactured 
>>It is certainly verifiiable that using closely spaced resonances the 
>>phase is bound closer to 0, and it is equally verifiable that this can 
>>yield fewer components to cross a target Z than would be necessary by the 
>>inductance alone.  When you are SUN and buy some 100's millions of 
>>capacitors per year, there is probably little penalty to the multitude of 
>>values, and I suppose you have worked out the manufacturing procedures so 
>>that they are either auditable or accidents don't happen or both.  But, I 
>>do not see that as a general case applicable to small and medium size 
>>At 10:30 AM 2/13/2004 -0800, Larry Smith wrote:
>>>(I changed the subject to better reflect the content..)
>>>Craig, Istvan, Steve - "the big V" associated with discrete capacitor
>>>impedance does not cause ASICs to resonate at the die but it can
>>>certainly make it a lot worse.  There will always be some level of
>>>inductance in the mounted package that will resonate with the
>>>capacitance in the die (and/or capacitance on the package).  Depending
>>>on how the PCB capacitors are chosen, their inductance  may (and
>>>usually does) add to that of the package and mount .  This will
>>>increase the Q of the resonant circuit and lower the resonant
>>>frequency (bad things).
>>>If PCB capacitors are chosen such that they present a resistive
>>>impedance rather than inductive impedance at the chip/package resonant
>>>frequency, they provide damping and reduce the Q of the resonant
>>>circuit rather than increase it.  This is the best you can hope for.
>>>The optimum situation is to have the PCB present the target impedance
>>>at zero phase (resistance) to the resonant circuit.  This presents the
>>>maximum damping resistance with acceptable voltage regulation (IR drop
>>>in the PDS resistance).
>>>Sure, it would be nice to have specified ESR capacitors to help us
>>>create the optimum impedance.  But I work in a components group.
>>>There is no way we will pay for the added costs, the testing, the
>>>inventory control or deal with the number of unique P/N capacitors
>>>that this would required if there were specified ESR options for each
>>>capacitor value.
>>>On the other hand, the ESR of ceramic capacitors available today is
>>>very predictable.  For a particular value of capacitor, we find about
>>>20% ESR variation over the several major capacitor suppliers.  We can
>>>live with that!  By choosing from a menu of 2 dozen or so capacitors
>>>(three per decade over 5 decades of values) it is easy to present any
>>>impedance you want to the mounted ASIC including a nice flat profile
>>>with frequency.  In every case that we have evaluated, this has
>>>resulted in the lowest component cost but does create added
>>>manufacturing complexity with more reels of components around the
>>>"pick and place" equipment.  Optimize it any way you want!  I know
>>>what I am going to do.
>>>Larry Smith
>>>Sun Microsystems
>>>Craig Twardy wrote:
>>> > Steve, Istvan;
>>> > "the big V" can also cause ASICs to resonate at the die.
>>> > Usually the significant lossy component (dampening) is the ESR of 
>>> "the big
>>> > V".
>>> > When this gets small, the ASIC package can become underdamped.
>>> > "the big V" does not have to coincident exactly with the natural resonant
>>> > frequency of the ASIC Package.
>>> > It just needs to be somewhere close (within an octave or two can be
>>> > sufficient).
>>> >
>>> > When this happens the noise seen at the PCB is usually low.
>>> >
>>> > I guess this goes back to a previous thread; without knowing the ASIC
>>> > package
>>> > And die components, power supply decoupling is a crap shoot.
>>> >
>>> >
>>> >
>>> > Craig
>>> >
>>> > -----Original Message-----
>>> > From: Istvan NOVAK [mailto:istvan.novak@xxxxxxxxxxxxxxxx]
>>> > Sent: February 13, 2004 8:27 AM
>>> > To: weirsp@xxxxxxxxxx
>>> > Cc: si-list@xxxxxxxxxxxxx
>>> > Subject: [SI-LIST] Re: Stack up for EMI reduction,plane resonance and 
>>> u-str
>>> > ip radiation etc etc
>>> >
>>> >
>>> > Steve,
>>> >
>>> > Well said.  Given the fact that there is still hardly any bypass 
>>> capacitor
>>> > on the market where the designer would have a known range for its ESR,
>>> > selecting the largest value cap in a ceramic case style, and creating a
>>> > single deep V seems to be a good working compromise.
>>> >
>>> > There are two penalties associated with this solution.  At low 
>>> frequencies,
>>> > where the V shape interfaces with the impedance of bigger capacitors, we
>>> > will have to pay a factor of two either in the inductance of the 
>>> bigger caps
>>> > (need twice as many) or in the capacitance of the ceramic caps 
>>> creating the
>>> > deep V (if we selected the biggest capacitance in the case style, 
>>> this also
>>> > means we need twice as many).  There is a similar but more severe 
>>> penalty at
>>> > high frequencies, where these ceramic capacitors interface with the 
>>> planes,
>>> > let it be thin dielectric -:) or thick dielectric.  To sufficiently 
>>> suppress
>>> > the capacitor-plane resonance and the first few plane modal 
>>> resonances, the
>>> > cumulative inductance of the parts has to be several times less, 
>>> which means
>>> > correspondingly more parts.  But I agree that given the circumstances 
>>> this
>>> > is a safe working solution.
>>> >
>>> > I hope sooner than later the industry will demand bypass capacitors with
>>> > specified ESR values (with +- tolerance) where the nominal ESR value 
>>> can be
>>> > selected from a list, similar to nominal voltage, material, etc.
>>> >
>>> > Regards,
>>> >
>>> > Istvan
>>> >
>>> >
>>> >
>>> > ----- Original Message -----
>>> > From: "steve weir" <weirsp@xxxxxxxxxx>
>>> > To: "Bart Bouma" <bart.bouma@xxxxxxxxx>
>>> > Cc: <si-list@xxxxxxxxxxxxx>; <si-list-bounce@xxxxxxxxxxxxx>;
>>> > <zhang_kun@xxxxxxxxxx>
>>> > Sent: Friday, February 13, 2004 4:47 AM
>>> > Subject: [SI-LIST] Re: Stack up for EMI reduction,plane resonance and 
>>> u-str
>>> > ip radiation etc etc
>>> >
>>> >
>>> >
>>> >>Bart, I don't know why people fear that big "V".  Capacitors by the
>>> >>decade are something that I oppose.  I have seen people, including
>>> >>respected consultants mess up capacitors by the decade and blow
>>> >>impedance targets by a factor of 3:1 or more.  In the meantime, no
>>> >>parts were saved. There is nothing wrong with an impedance lower than
>>> >>target, and the capacitor count is driven by the requisite inductance
>>> >>to meet the HF intercept.  Take the same qty of capacitors using
>>> >>decade spacing, and just substitute the larger value for all of them
>>> >>and the impedance plot is
>>> >
>>> > still
>>> >
>>> >>very well behaved, and the phase doesn't go all over creation.
>>> >>
>>> >>The only argument that anyone could ever try and make for smaller
>>> >>value capacitors that makes any sense to me is the higher ESR of the
>>> >>small values, provided it is high enough to get close to Ztarget that
>>> >>will help damp anti resonance with the planes.  In that case, I can
>>> >>see clear to two values of ceramic caps properly chosen, but not by
>>> >>the decade.  But, I
>>> >
>>> > have
>>> >
>>> >>yet to see any author who advocates multiple values of MLCCs advocate
>>> >>on the basis of bringing up the ESR.  It has always been based on this
>>> >>folklore surrounding some perceived need for a flat impedance curve,
>>> >>that many then blow due to antiresonance.
>>> >>
>>> >>Regards,
>>> >>
>>> >>
>>> >>Steve.
>>>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 technical documents are available at:
>>>                 http://www.si-list.org
>>>List archives are viewable at:
>>>                 http://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:

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

List technical documents are available at:

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

Other related posts: