OMAP-L138 DQGATE Routing Specification

Milan

CKB0B1 is the sum of the length of the CK net plus the average length of the DQS0 and DQS1 nets. Typical DQGATE length is equal to CKB0B1 with max 100mils skew. 

Thanks

David

David,
If I understand you properly, the 100mils spec is a tolerance of DQGATE trace length with respect to definition you provided. 

Let me raise the following question.

Datasheet Table 5-34 defines three NetClasses - ADDR_CTRL, D0 and D1. Table 5-36 and Table 5-37 specify routing reqs for these individual NetClasses, however, I am NOT aware of ANY mutual requirement for the three NetClasses in terms of trace lengths.

With that being said, I am not quite sure whether I fully understand the rationale behind the equation for DQGATE you provided.

Do you think you could help?

Many thanks,
Milan

Milan

DQGATE0 and DQGATE1 are the loopback signal for timing adjustment (DQS gating). You can route the three net classes first following the requirement in table 5-36 and 5-37, then route the DQGATE0 and DQGATE1.

David

David,
I still do not understand the point here.

The documentation does not impose any mutual requirement between D0 and D1 NetClass. Thus, D0 can be 150mils long whereas the length of D1 can be of 500mils with no violation whatsoever.

WHY do you call for an accurate DQGATE loopback within 100mils that covers BOTH data lanes then?

There must be something I have been missing here. Please help. 

Many thanks in advance,
Milan

Milan

A 16 bit DDR memroty system will have two sets of data net classes, one for data byte 0 and one for data byte 1, each with its associated DQS. But there is no need to skew match across data bytes.

DQGATE is used by the OMAPL138 to predict the round trip delay of the PCB layout. The length equals to the length of CK plus the average length of DQS0 and DQS1 net classes.

Thanks

David

David,
Please explain how you can predict a round trip delay if there is no skew match across data bytes.

Thanks,
Milan

Milan

This is why we have the requirement that DQGATE trace length need to be equal to the length of CK plus the average length of DQS0 and DQS1 net classes.

David

David,
I am giving up.

Best,
Milan

Milan

I am not sure what the confusion is. You have the D0 and D1 net class, The nominal trace length for D0 and D1 class is DQLM mils. You also have the CK net class with nominal trace length of CACLM. So if CACLM is 100mils, D0 is 100mil, D1 is 150mils. then DQGATE length will be equal to 100mil + (100+150)/2 = 225mils.

David

David,
What confuses me is the following statement from you previous message:

"So if CACLM is 100mils, D0 is 100mil, D1 is 150mils."

I do not see the reason why D0 should be 100mils when CALCM is 100mils.
I am not aware of any mutual spec among CALCM, DQLM0 and DLQM1. That's the root cause of my problem.

Milan

Milian

There is no mutual spec among CALCM, DQLM0 and DQLM1. Constrained by the overall area size between DDR and OMAPL138, you can't have significant skew difference between CALCM, DQLM0 and DQLM1. Even if there is 50mil difference, it is only around 7ps skew.

Thanks

David  

Hi David,

Why is there the DQS/D nominal trace length as No.3 in the DQGATE Routing Specification?

Is this a typo? Does proper specification indicate that the DQGATE nominal trace length is range of "CKB0B+50 Mils" from "CKB0B-50 Mils"?

Best regards,

Daisuke

David-san,

Thank you for your reply.

Best regards,

Daisuke