[SI-LIST] Re: How to estimating signal waveforms at inaccessible points in DDR3 or 4?
- From: Hermann Ruckerbauer <Hermann.Ruckerbauer@xxxxxxxxxxxxx>
- To: luliu@xxxxxxxxxx, "si-list@xxxxxxxxxxxxx" <si-list@xxxxxxxxxxxxx>
- Date: Tue, 27 Jan 2015 08:59:23 +0100
Hello,
for a Point to Point connection (Single Slot, single Rank) with only
writes you will get quite reasonable results even when just embedding an
ideal transmissionline delay from your measurement point to the die
(assuming you are at least close to the memory component).
This will remove the most critical point in your measurement which is
the reflection that is causing a plateau.
So far in most of my measurements I did this and did not utilize
interposers .. but I have to admit, using such interposers would have
made life simpler several times ...
As e. g. DDR3 memory subsystems are designed to handle dual slot, dual
channel at 1600 there is often enough margin in such a P2P system that
this simplification already gives reasonable results and allow a
judgement between good and bad design.
In case you use the interposer you have at least the interposer model
from your vendor available and you can easily de-embedd the stub and the
probe (at least for the keysight solution, but I guess for other vendors
as well). What is still missing is the package model, that you can take
from the IBIS. But the closer you are at the die the simpler and more
accurate will be the de-embedding setup. So in such a configuartion this
should not be the big issue any more.
But I have to admit for my own interposer design I have still to create
the model for de-embedding .. let's see when there is real need (and
time ;-) ) to do so.
But as discussed in the past: you can spend endless time in optimizing
your setup and getting better models for embedding/de-embedding.
Overall I think the specification should adjust to reality ;-)
For DIMM based solutions maybe one can adjust to some of the serial link
standards techniques:
- e.g. define the signal quality at the connector.
- This leaves still an issue for solder down solutions .. but this is in
High speed serial links there as well and maybe interposer would solve
this issue ...
- define a low level test support in the Controller/DRAM that drive out
defined data (even the system is not really powered up, a bit more than
current read calibration data Testmode).
- Define Testboards that allow simple measurments (e. g. CLB as DIMM to
test a motherboard or CBB as Motherboard to test a DIMM)
- Splitting RX/TX and Controller/DRAM measurments
In the meantime I think I have an idea how the PCIe Gen3 RX compliance
is working (still trying to verify this one) and maybe this concept
would be also reasonable ..
Last but not least (maybe even most important): I think the on-Die
measurements will overtake this part before it is critical enough that
it is absolutely required to change anything.
On the controller side there are already many controllers out that can
handle this, and on the DRAM side we had already some prototypes that
did this 6 years ago when I was working at Qimonda. Especially with DDR4
Vref Adjustment capability there is nearly everything available that is
required even without additional DRAM features ..
I think the equipment vendors will loose their grounds here ... But this
might be topic for another discussion ...
Sorry for again writing to long ..
Hermann
P.S. is anybody interested in developping such a Memory "Compliance
Test" setup ? Not sure how far we will come without controller or DRAM
vendor support, but there might be some options .. e. g. for Altera
contrtollers I think it should be possible to generate a testsetup to
measure CA, DQ Write and maybe even DQ Read performance.
It would be really a nice and interesting project ..
Our next Events:
Seminar "Open the Black Box of Memory" (13-14.04.2015) in Stockholm Sweden
Embedded World 2015 (24-26.02.2015) in Nuremberg
EKH - EyeKnowHow
Hermann Ruckerbauer
www.EyeKnowHow.de
Hermann.Ruckerbauer@xxxxxxxxxxxxx
Itzlinger Strasse 21a
94469 Deggendorf
Tel.: +49 (0)991 / 29 69 29 05
Mobile: +49 (0)176 / 787 787 77
Fax: +49 (0)3212 / 121 9008
Am 27.01.2015 um 07:32 schrieb Robbie Liu:
> Hi Hermann:
> Thank you for your reply, yes , both the question you mentioned, the ODT
> monitor and separate the read/write , is still challenging, especially for a
> dual slot dual rank system. But we may make the thing simple first , for
> example , the 1 slot and 1 rank system, with an always write control.
>
>
> Thanks and Regards,
> LIU Luping
>
>
>
>
> -----Original Message-----
> From: Hermann Ruckerbauer [mailto:Hermann.Ruckerbauer@xxxxxxxxxxxxx]
> Sent: Saturday, January 24, 2015 1:24 AM
> To: Robbie Liu; si-list@xxxxxxxxxxxxx
> Cc: tom@xxxxxxxxxxxxxxxxx; Charles.Grasso@xxxxxxxxxxxx;
> heidi_barnes@xxxxxxxxxxxx; sherman.chen@xxxxxxx
> Subject: Re: [SI-LIST] How to estimating signal waveforms at inaccessible
> points in DDR3 or 4?
>
> Hello,
>
> In addition to getting the passive models for embedding the multipoint and
> variable termination of the DDR memory bus is causing big problems.
> So even if you have the models ot the channel your run into the issue, that
> you need to know which termination settings are applied to use tools like
> infiniisim.
> So you need exactly to know the termination scheme for e. g. writes to the
> given rank. To know this you might need to monitor ODT signals to other ranks
> and also need to know dynamic termination.
> One can build quite complex embedding structures with Infiniisim, but a
> double slot memory system will bring also such a tools to the edge what is
> possible.
> You need also to know the Memory controller TX Ron and have a model of the
> complete channel to do this correct.
>
> And last but not least the separation between Write/Read and also Write to
> the specific rank need to be decoded exactly.
>
> And your scope application might take for overshoots the complete trace and
> not only the write bursts, but for read bursts the De-Embedding setup is not
> correct.
> So even if you do have the models available it is a really tricky task to
> implement the setup for the embedding correctly.
>
> Question is how accurate you want/need to be .. often i work with the very
> simplified setup that I measure after the serial termination on the DIMM and
> just embedd the rest of the routing on the DIMM (and assume some specific
> termination scheme).
> But I know that this is only a very rough estimation of the problem ...
>
> Hermann
>
> Our next Events:
> Seminar "Open the Black Box of Memory" (13-14.04.2015) in Stockholm Sweden
> Embedded World 2015 (24-25.02.2015) in Nuremberg
>
> EKH - EyeKnowHow
> Hermann Ruckerbauer
> www.EyeKnowHow.de
> Hermann.Ruckerbauer@xxxxxxxxxxxxx
> Itzlinger Strasse 21a
> 94469 Deggendorf
> Tel.: +49 (0)991 / 29 69 29 05
> Mobile: +49 (0)176 / 787 787 77
> Fax: +49 (0)3212 / 121 9008
>
> Am 23.01.2015 um 10:06 schrieb Robbie Liu:
>> Dear All:
>>
>> You may treat this question as a follow up discussion of recently
>> hot discussion "tips of using TDR probe " and "Pin vs. Die", so I may
>> thanks to those experts first ,for their wonderful knowledge sharing.
>>
>> The question will be met by many hardware engineers such that our
>> DDR3/4 chip may be imaging mounted on board , or the board use HDI so you
>> can't have test point close enough to the end.
>>
>> Take DDR write as example, If we just have a test point (TP1)10mm
>> away from the DDR3 chip, and you want to estimate the signal waveform at
>> the DDR3 die (TP2 ), we need to know the S21 from TP1 to TP2, but for most
>> of us didn't have such capability to access the TP2, so the problem comes.
>>
>> Certainly you can use 2D or 3D EM tool to get the channel S
>> parameter, but the high accuracy need experience , and the package file is
>> difficult to get.
>>
>> So can we just use one port test from TP1 ,to get the S parameter of
>> this two port network?
>>
>>
>> We notice two papers recently mentioned this topic , one is by
>> Tomohiro Kinoshita, Shoichi Hara, Eiji Takahashi ,Panasonic, A technique for
>> estimating signal waveforms at inaccessible points in high speed digital
>> circuits, 2013 9th International Workshop on Electromagnetic Compatibility
>> of Integrated Circuits (EMC Compo), December 15-18, Nara. Japan, it use
>> their own Novel Electromagnetic Tool 'MomCACE' to get the ABCD matrix from
>> TP1 to TP2, then estimate the waveform at the channel end.
>> Another is by Evelyn Mintarno#1, Steven Ji*2 , A Practical Method to
>> Characterize Interconnect in a Fully Loaded System, and its Application to
>> DDR3 Channel, Proceedings of the 38th European Microwave Conference, , based
>> on my understand (maybe wrong ), the method is use for characterize the
>> channel between the CPU and the SO DIMM connector, right? I was wondering
>> if we can push it further , like the PCB +Package, so that we can use the s
>> parameter to estimate signal waveforms at inaccessible points in high
>> speed digital circuits, for example , the imaging mounted DDR3 chip on
>> board. The condition at port 2 may change between ODT 40, ODT 75 and
>> ODT120, or ODT OFF, not sure if it is possible.
>> What is your opinions ? do you have better solutions?
>>
>>
>>
>>
>> Thanks and Regards,
>> LIU Luping
>>
>>
>> ________________________________
>>
>> Confidentiality Notice.
>> This message may contain information that is confidential or otherwise
>> protected from disclosure. If you are not the intended recipient, you are
>> hereby notified that any use, disclosure, dissemination, distribution, or
>> copying of this message, or any attachments, is strictly prohibited. If you
>> have received this message in error, please advise the sender by reply
>> e-mail, and delete the message and any attachments. Thank you.
>>
>> ------------------------------------------------------------------
>> 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
>>
>>
>
> ________________________________
>
> Confidentiality Notice.
> This message may contain information that is confidential or otherwise
> protected from disclosure. If you are not the intended recipient, you are
> hereby notified that any use, disclosure, dissemination, distribution, or
> copying of this message, or any attachments, is strictly prohibited. If you
> have received this message in error, please advise the sender by reply
> e-mail, and delete the message and any attachments. Thank you.
> ------------------------------------------------------------------
> 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:
