Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: [802.3_50G] Back Plane Loss Discussion

Chris 

Yes with minor correction replace 100G with 200G.  There is strong use case for 1 lane and 4 lane, but 2 lane KR2 and CR2 in the past only had limited deployment, but I do see greater use case for 100G-FR2/LR2 than 2 lane Cu.

·         1 lane 50G PAM4 for 50GBASE-KR and 50GBASE-CR
·         2 lane 50G PAM4 for 100GBASE-KR2 and 100GBASE-CR2
·         4 lane 50G PAM4 for 200GBASE-KR4 and 200GBASE-CR4
Thanks,
Ali Ghiasi
Ghiasi Quantum LLC

On Jan 15, 2016, at 4:53 PM, Chris Cole <chris.cole@xxxxxxxxxxx> wrote:

When we discuss 50G PAM4 for backplane and copper, I am assuming we mean the following three application sets:
 
·         1 lane 50G PAM4 for 50GBASE-KR and 50GBASE-CR
·         2 lane 50G PAM4 for 100GBASE-KR2 and 100GBASE-CR2
·         4 lane 50G PAM4 for 100GBASE-KR4 and 100GBASE-CR4
 
Is this the common understanding?
 
Chris
 
From: Mellitz, Richard [mailto:richard.mellitz@xxxxxxxxx] 
Sent: Friday, January 15, 2016 2:20 PM
To: STDS-802-3-50G@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_50G] Back Plane Loss Discussion
 
30 dB… no problem.
Oh…
No crosstalk, no reflections, no noise, no packagesJ
 
From: Vineet Salunke (vineets) [mailto:vineets@xxxxxxxxx] 
Sent: Friday, January 15, 2016 4:32 PM
To: STDS-802-3-50G@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_50G] Back Plane Loss Discussion
 
I agree with Ali that 30 dB should be minimum.
Let’s not forget the same spec needs to work for 50G CR (copper cables),
where total channel total with 3m cables will be 29 dB (as shown by the work in .by).
 
--vineet
 
From: Ali Ghiasi [mailto:aghiasi@xxxxxxxxx] 
Sent: Friday, January 15, 2016 9:41 AM
To: STDS-802-3-50G@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_50G] Back Plane Loss Discussion
 
Joel 
 
I definitely like if we could do 32 dB, but absolutely we can’t go under 30 dB!
 
Thanks,
Ali Ghiasi
Ghiasi Quantum LLC
Office (408)352-5346

<image001.png>
 
On Jan 15, 2016, at 9:13 AM, Joel Goergen (jgoergen) <jgoergen@xxxxxxxxx> wrote:

 

Great summary Ali.
 
I targeted 32dB because at this level, I felt somewhere around 20% of my links would require retimers.  More then 50% links requiring retimers makes it a non starter for me.
 
Take care
Joel
 
From: Ali Ghiasi <aghiasi@xxxxxxxxx>
Date: Friday, January 15, 2016 at 9:10 AM
To: Joel Goergen <jgoergen@xxxxxxxxx>, Richard Mellitz <richard.mellitz@xxxxxxxxx>
Cc: "STDS-802-3-50G@xxxxxxxxxxxxxxxxx" <STDS-802-3-50G@xxxxxxxxxxxxxxxxx>
Subject: Re: [802.3_50G] Back Plane Loss Discussion
 
Joel/Rich
 
I looked at the TE Whisper 1 m backplane with Megtron 6 HVLP with loss of 29.7 at PAM4 Nyquist.  As you know 6 mils trace width with 14 layers as was the TE construction is not a high density line card implementation, for full 1 m backplane to be under 30 dB we may even need to go with Megtron 7 for a more practical implementation!
 
For all thoes promoting 25 or 27.5 dB channel with retimer is non starter because of the added power with racks hitting 20+ KW!  When you add a retimer the link power end up to be (2 KR + 1 VSR) ~ 2.5x  than not using retimer. 
 
Combination of cost, power, and density require more due diligence than just go with OIF-56G-LR loss limit of 27.5 dB.  If we look historically has been leading IEEE on defining the interface specification, but IEEE specification always have been improvement over OIF with higher loss at least in case of 10G/25G backplanes.  OIF specification also need to operate at 56+ Gb/s to support OTN which is added constrain compare to the IEEE specifications which operate at ~53.1 Gb/s.
 
Thanks,
Ali Ghiasi
Ghiasi Quantum LLC
Office (408)352-5346

<Logo3b.png>
 
On Jan 15, 2016, at 8:48 AM, Joel Goergen (jgoergen) <jgoergen@xxxxxxxxx> wrote:

 

I agree with you … but what margin are you looking for?
 
In the designs I target, 25dB has no value.  I would need retimers.  Retimers add space and power.  So if I add them on more then 50% of the links, then I should just as well set the power level to what I need it for to set the retimer number back into the 20ish% links requiring retimers.
 
Take care
Joel
 
From: <Mellitz>, Richard <richard.mellitz@xxxxxxxxx>
Date: Friday, January 15, 2016 at 8:17 AM
To: Joel Goergen <jgoergen@xxxxxxxxx>, "STDS-802-3-50G@xxxxxxxxxxxxxxxxx" <STDS-802-3-50G@xxxxxxxxxxxxxxxxx>
Subject: RE: Back Plane Loss Discussion
 
For this market, it is not only power but lower cost volume manufacturability.
 
I really need 60dB if I use the “fit in the shoe box” argument. J
 
… Rich
 
From: Joel Goergen (jgoergen) [mailto:jgoergen@xxxxxxxxx] 
Sent: Friday, January 15, 2016 11:10 AM
To: STDS-802-3-50G@xxxxxxxxxxxxxxxxx
Subject: [802.3_50G] Back Plane Loss Discussion
 
All
 
My apologies for not starting this discussion sooner.
 
I understand the there is a lot of push-back for a 50G interface 32dB@14Ghz.
I would guess the reason is power related.
 
My argument is that if the limit is 25dB or 28dB, a significant amount of retimers will be required. And therefore the power per link will be almost double anyway.  So power isn’t an argument.
 
However, I would conceded to a lower loss level if we all agreed to completely define the retimer block and recognize it as a valid block function.
 
Take care
Joel
 
<Logo3b.png>