[SI-LIST] Re: distributed packaging model
- From: "Perry Qu" <perry.qu@xxxxxxxxxxx>
- To: Ozgur Misman <omism@xxxxxxxxx>
- Date: Thu, 08 Nov 2001 13:04:07 -0500
Hi! Ozgur:
Thank you for your reply. I have some bad experience with lumped RLC type of
model in a 200-300 ps rise time differential pair application. The package is a
big size ceramic one. The stub length on the package itself approach 1", and
ceramic material makes it longer electrically. We see some strange ripples on
the rising/falling edge when measured at the BGA balls and we suspect packaging
is the problem. A lumped RLC type of simulation does not reflect this. We then
request a distributed model from the vendor and we are lucky to get a model
including details such as solder balls, traces, vias, coupling between traces,
etc. With this detailed model, we are able to obtain good correlation with
measurement.
Since then I became very careful about packaging parasitics whenever I have
fast rise time signals. And by playing around with spice simulation, it appears
to me that packaging stub could cause problems at a slower rise time for
single-ended application.
I agree with you that the packaging model should be divided into sections
representing traces on packaging substrate, which can be simulated as
transmission line, and other discontinuety such as vias, wire bond, solder
balls, etc., which can be simulated using lumped RLCs. The question is, how
many vendor will actually do that ?
Regards
Perry
Ozgur Misman wrote:
> Hi Perry,
>
> 1. divide lumped RLC into sections of Pi/T circuit and interconnect them
> together.
>
> Package interconnect is typically electrically small, unless the package
> size is large and the risetimes are very fast ( 100 psec or so.) the lumped
> approach works fine. Especially with chip scale or near chip scale packages
> gaining dominance it is true that a lumped model should suffice for most
> applications. So as you see as risetimes get faster packages shrink as
> well.:)
> However there are instances where a lumped model is not sufficient such
> gigabit transceivers where the data transfer rates are quite high (OC-48,
> OC 768 applications). I can tell you that extracting the lumped parameters
> and then just segmenting it into 4-5 ( as determined by the bandwith
> requirement ) is a shortcut and works very well for uniform transmission
> lines only!. However, package interconnect structures are very complex with
> wirebond, trace, via, solderball making the signal transmission path.
> Typically, a good lumped model should extract the wirebond and the rest of
> the net seperately. So modeling the wirebonds as lumped and the rest of the
> net (trace via and solderball ) lumped should be good enough ( at leat you
> are capturing the most obvious impedance discontinuity). Then if we need
> segmentation you can do it on the trace portion. My experience is that
> this works quite well. I agree that you are not capturing the impedance
> discontinuity at the via, and solderball, however that is fine most of the
> time.
> We are constantly keeping an eye on commercial simulators and actually in
> the process of buying one to provide distributed models.
> Even the commercial simulators however do not capture each impedance
> discontinuity. They simply compare the risetime (and therefore the bandwith
> required) and compare that to the physical length and decide how many
> segments is required from the model. then analyze each segment seperately,
> in each segment there maybe more than one discontinuity so you may not be
> able to capture it.
>
> Do you relly need that level of detail is another question, though!
>
> 2. instead of RLC, use a transmission line with impedance of Z
> =sqrt(L/C) and delay calcuated from length of packaging trace and L/C.
>
> Well this is the actually same as your question number one. Unless that Z
> reflects all the discontinuities ( multiple Z segments) this is comparable
> to the lumped model. so the same arguments apply.
>
> I think if you would like a very detailed job, you should ask a model for
> each discontinuity,
> a lumped model for the wire, a transmission line (Z,Td) or a distributed
> R,L,C (enough segments) model for the trace , a lumped model for the via
> and another lumped model for the solderball and then built eqivalent
> circuit manually if you have time for it:)
>
> Regards,
> Ozgur Misman
> Amkor Technology
>
> "Perry Qu" <perry.qu@xxxxxxxxxxx>@freelists.org on 11/08/2001 09:03:11 AM
>
> Please respond to perry.qu@xxxxxxxxxxx
>
> Sent by: si-list-bounce@xxxxxxxxxxxxx
>
> To: si-list <si-list@xxxxxxxxxxxxx>
> cc:
>
> Subject: [SI-LIST] distributed packaging model
>
> Hi!
>
> I'm aware that distributed packaging model should be used for rise time
> signal simulation. However, most of the packaging model we got from
> vendors are in terms of lumped RLC value, e..g, those listed in IBIS
> model. When I asked for distributed model, I got two suggestions from
> vendors as an easy go-around:
>
> 1. divide lumped RLC into sections of Pi/T circuit and interconnect them
> together.
>
> 2. instead of RLC, use a transmission line with impedance of Z
> =sqrt(L/C) and delay calcuated from length of packaging trace and L/C.
>
> I have doubt about both approaches. Did any one tried and verified them
> ?
>
> Thanks in advance
>
> Thanks in advance
>
> Perry
>
> --
> Perry Qu
>
> Product Integrity | 600 March Road
> Alcatel Canada | Ottawa, ON K2K 2E6, Canada
>
> DID: (613) 7846720 | FAX: (613) 5993642
>
> ------------------------------------------------------------------
> 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
--
Perry Qu
Product Integrity | 600 March Road
Alcatel Canada | Ottawa, ON K2K 2E6, Canada
DID: (613) 7846720 | FAX: (613) 5993642
------------------------------------------------------------------
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:
