[SI-LIST] Re: UltraCAD ESR and Bypass Capacitor Caculator

  • From: Raj Raghuram <raj.raghuram@xxxxxxxxxxxxxxx>
  • To: si-list@xxxxxxxxxxxxx
  • Date: Mon, 04 Aug 2003 08:02:35 -0700

Another opininon on this I have heard is that at higher frequencies, it 
is the decoupling on the package and IC that really matter. The package 
decap can have lower inductance and the IC decap almost negligible 
inductance. Unfortunately, the system level designer is often not privy 
to information on these decaps, especially the on-chip decap. All he has 
is the IBIS model which is made assuming ideal supplies. IC 
manufacturers need to provide some information about on-chip decoupling. 
Some opinions I have heard from designers on this are:

1. A profile of the current drawn by the IC is really needed for 
effective PDS design.

2. If there is on-chip cap deliberately added, this needs to be 
mentioned somewhere in the data or IBIS model.

3. Often there are a number of gates not switching. Each one has some 
capacitance between power and ground. The sum of these can be a 
significant source of decoupling, whether intended as such or not.

4. Issues such as whether on chip core supply is the same as the IO, 
etc. become important if on-chip decoupling is to considered.

Ray Anderson wrote:

>With all due respect to the UMR authors, I think the location
>of a decap on a set of power planes DOES matter.
>
>There is an inherent time delay associated with the current flow
>from the location where a decap is placed to the location where the 
>chip to bypassed is located.
>
>The decaps job is to provide current to the load quickly enough
>that a voltage collapse on the planes doesn't occur. If the decap
>is too far away from the current consuming load it is possible
>that load's need for a pulse of current will be over before the 
>current can get there.
>
>At low frequencies you can place a decap just about anywhere on a plane
>and it will function just fine. At higher frequencies the decap must
>be closer to the device to be bypassed. The concept of "effective bypassing
>radius" says that a decap must be within say 1/6 to 1/10 wavelength
>at the frequency of interest to be effective. (the exact number is
>debatable). The frequency of interest is the resonant frequency of
>the mounted decap (where the caps ESR is lowest), and the distance is 
>a function of the speed of propagation with a given dielectric between
>the planes (around 180ps/inch for FR4).
>
>You need to consider the role each component of a PDS (power distribution
>system) plays in maintaining a low impedance across a wide bandwidth.
>A typical PDS is composed of a VRM, bulk capacitors, ceramic decaps, and
>the power planes and last but not least the package housing the silicon.
>
>The VRM provides a low impedance from DC to perhaps a few hundred kHz,
>the bulk capacitors contribute their impedance profile from a few 
>hundred kHz up to about a MHz. From a Mhz to maybe 120MHz or so the
>ceramic decoupling caps provide low impedance. Above that frequency
>the distributed capacitance provided by the planes provides the requisite
>low impedance to the PDS. When the individual impedance profiles of
>each of the constituent parts are superimposed, the resulting composite
>profile provides a low impedance over the full frequency range that the
>PDS was designed for. Note that above the package resonance frequency
>(which usually is from about 50 to 100 MHz for typical packages) decoupling
>on the board is ineffective and the necessary decoupling must be provided
>for either in the package or on the silicon. However, at frequencies
>above the package resonance frequency the discrete decaps and distributed
>plane capacitance can be most useful for EMI purposes.
>
>To summarize, at low frequencies, the decoupling capacitors can be placed
>almost anywhere withing reason and will be effective. As the frequency
>of interest increases, the allowable distance decreases. As an example,
>at 1 MHz the wavelength (on FR4) is around 214 meters. So lambda/10 is
>around 21 meters. Hence, if your decap which is effective at 1 MHz is
>within 21 meters of the consuming device, then the decoupling will be
>effective. Conversely, if you are trying to decouple at device with
>a decap that is effective at say 100 MHz then the wavelength of interest
>is about 1.4 meters. Lambda/10 is then .14 meters (approx 5.5 inches).
>In this case if the decap is appreciably more than 5.5 inches from the
>currewnt consumer then the delay will be such that the decap will not be able 
>to 
>provide current quickly enough to be effective.
>
>-Ray Anderson
>Sun Microsystems Inc.
>
>
>Lee Ritchey wrote: 
>  
>
>>>If you remember the UMR paper on power bus decoupling, it made a clear case
>>>that the location of the capacitors is relatively unimportant.
>>>      
>>>
>
>
>Abe Riazi wrote: 
>  
>
>>To be fair, it should be added that different recommendations ( from above) 
>>    
>>
>exist in literature
>  
>
>>regarding optimum location of the decoupling caps.
>>
>>Two examples follow:
>>
>>1.  Stephen H.Hall et al., " High-Speed Digital System Design........", on 
>>page 
>>    
>>
>247 it is stated:
>  
>
>>" Figure 10.10 suggests not only that the local decoupling capacitor value 
>>    
>>
>should be chosen
>  
>
>>appropriately but that the capacitor is placed physically close to the 
>>    
>>
>component to minimize
>  
>
>>the loop area ".
>>
>>2. A technical paper by R. Chen, "Where to Place Decoupling Capacitors?..."  
>>    
>>
>available on
>  
>
>>www.sigrity.com web site, implies that the best location for decoupling caps 
>>    
>>
>needs to
>  
>
>>be ascertained with aid of electromagnetic field simulation.
>>
>>Best Regards,
>>
>>Abe Riazi
>>ServerWorks
>>    
>>
>
>------------------------------------------------------------------
>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
>  
>
>  
>

-- 
Raj Raghuram
Berkeley Design Automation (http://berkeley-da.com)
2902 Stender Way,
Santa Clara, CA-95054
PH: (408)-496-6600 ext.203
Cell: (408)-390-7614
EMAIL: raj.raghuram@xxxxxxxxxxxxxxx



------------------------------------------------------------------
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: