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

Re: [BP] Opinion about the Channel Model Ad Hoc



I tend to agree with Jia. We heard from many system vendors that they are interested to design 20G and 40G line cards. In these cases, cutting the number of backplane traces by 2X is essential and, as Jia point out, not having to use esoteric materials and costly manufacturing techniques as required by 10G SERDES results in a substantial economic benefit.
 
Regards,
 
Reuven
 
Reuven Segev
Director of Marketing
TeraChip, Inc.
Tel: (650)320-8148 ext 204
Cell: (415)307-7683


From: Jia Gongxian [mailto:jgxian@HUAWEI.COM]
Sent: Wednesday, March 23, 2005 1:56 AM
To: STDS-802-3-BLADE@LISTSERV.IEEE.ORG
Subject: Re: [BP] Opinion about the Channel Model Ad Hoc

Hello everyone,
 
As a new designed backplane for 10GBASE-KR, maybe its economical, however for 1000BASE-KX and 1000BASE-KX4 application, it must be expensive. The only advantage is that it is compatible with 10GBASE-KR.
 
As for 6.25G interface, I personally think it is necessary for IEEE 802.3 ap to cover it. According to our analsysis through simulation and meausrement, systems for 2.5/3.125 Gbps application can be easily upgraded to 5/6.25 Gbps,given the backplane design be slightly changed. We can even use common FR4 instead of improved FR4. However, there lies in great difficulties for 2.5/3.125 systems to be upgraded to serial 10Gbps.
 
 
Jia Gongxian
Huawei Technologies Co.,Ltd.
Tel:     86 755 89651154
FAX:     86 755 89650731
Email:   jgxian@huawei.com
website: http://www.huawei.com
----- Original Message -----
Sent: Friday, March 18, 2005 12:52 PM
Subject: Re: [BP] Opinion about the Channel Model Ad Hoc

Interesting post.
 
Brings forward the question on whether there is economic justification for the 10G per lane, given the incremental cost incurred by the backplane and the line cards.
Any comments/opinions?
 
An additional question would be the reason for not including a 6.25Gx 2 lanes solution. After all, the OIF does have a specification for a 6.25G interface. What would be the PCB (backplane and line card) trade off for 3.125, 6.25, 10?
 
I would be curious to learn more about that.
 
Reuven
 
Reuven Segev
Director of Marketing
TeraChip, Inc.
Tel: (650)320-8148 ext 204
Cell: (415)307-7683


From: Jia Gongxian [mailto:jgxian@HUAWEI.COM]
Sent: Thursday, March 17, 2005 5:34 PM
To: STDS-802-3-BLADE@LISTSERV.IEEE.ORG
Subject: [BP] Opinion about the Channel Model Ad Hoc

Hi all IEEE P802.3ap Task Force Members,
 
I have been concerned for IEEEE 802.3 ap Task Force for a long time,  and have read through all the relevant materials available from the website up to now.  I do hope I have a chance to discuss with all members about the project  as a system backplane designer, to share what I think about it. Since it is inconvenient for us to take part in the face to face meeting and teleconference, I hope I can use the "e-mail reflector" as a platform to share with you all.
 
The following is what I want to know and what I care about most:
 
1)  From  the object   objective,  the project will cover 1 G per lane, 3.125 GbpsX4 lanes and  10Gbps per lane. Does this mean the channel will  meet all the scope at  the same time ? 
For the legacy backplane used for around 1 Gpbs , most of them is made of common FR4 and general connector,such as 2mm connector without shield pin.  I think it is very hard to upgrade to serial 10 Gbps, at most up to 3.125 Gbps, economically.
For the legacy backplane used for around  3.125Gbps per channel,  we still use common FR4, not improved FR4, and apply common manufacturing technology.
For the NRZ coding 10Gbps application, if we use this kind for backplane, I think this is very hard to realize. Thus I think for serial 10G backplane, we may using Greenfield backplane, not legacy backplane.
 
According to our simulation and measurement results,  the insertion loss of channel up to 3 GHz or so, is mainly effected by attenuation of copper loss and dielectric, impedance discontinuity caused by trace and connector, beyond 3 GHz,  the via stub play a more and more important role, firstly,  backplane via stub, secondly, line card stub.
 
Given  we want each layer in PCB can be routed high speed signals,for NRZ coding 5/6 G application, backplane via maybe need be backdirlled, for NRZ coding 10 G application, line card via maybe backdrilled also.
 
Does this project plan to using common PCB technology to make the backplane and linecard ? or using backdirlling and other technologies to reduce the via stub ?
 
2) As for the channel model, I think the channel specifications from TP1 to TP5 is important, especially for 10G serial application. for 1G per lane and 3.125G X 4 lanes application,  maybe we can ignore the influence caused by the component from TP4 to TP5 (two line card vias and a compactor), since before 3G Hz, the via's influence is limited, however, for 10G serial application, especially for NRZ coding, line card vias also play a great role for the channel performance. According to our simulation result, for the via routing in layer 2 (line card is about 2.5mm thick), beyond 5 GHz, the insertion loss of the via drop's sharply. We need define specifications from TP1 to TP5, instead of the specification form TP1 to TP4.
 
3) As for the test sample, I think we should define some specifications  to guide how to design. The test samples must actually reflect the real product design, or else, the measurement result and conclusion according to the result is meaningless.  For example, different types of test launches we used  will greatly influence our test result. I think the rules is, the design will reflect the practical design, not just for test.  For instance, we can use optimized SMT sma and with  a PTH (plated through hole) with 8/10 mil diameter to reflect the BGA via in practice.
 
4)  As for the channel crosstalk,  some member have mentioned using power sum to define the spec, however, the project is aim at high speed digital application, all the standards are defined as voltage in practice, such as eye height, imput sensitivity, ect. We usually using voltage sum to calculate the worst case of crosstalk, not power sum.
 
5)As for the Serdes, It's better to integrated ac compactor into Serdes,  which will significantly improve the channels performance, for we can reduce 2 line card vias and a compactor in linecard. Serdes for 10G serial applications must have  adaptive equalization and BIST function.
 
Any comments will be welcomed.
   
Thanks and Best regards.
 
Jia Gongxian
Huawei Technologies Co.,Ltd.
Shenzhen China
 
Tel:     86 755 89651154
FAX:     86 755 89650731
Email:   jgxian@huawei.com
website: http://www.huawei.com