DC Blocking Capacitor for 10Gbps Ethernet

2017-06-30 by:CAE仿真在线来源:互联网

Hello colleagues,

I would like to understand the perspective of the group with regards to external DC block capacitors in 10Gbps channels such as the one being designed by this group. My past experience indicates it will be difficult to find DC block capacitors that are specified for operation at 5GHz and beyond - at least capacitors that are priced reasonably such that their cost is a small proportion of the link cost. In the farily recent past I have had difficulty finding capacitors with low enough ESRs even at 4Gbps operation. Even if such a capacitor is found the transmission line discontinuities (vias, pads etc) associated with such apacitors present a significant degradation that would be nice to eliminate.

I will ask a secondary question.

Has it been considered to require the DC block capacitor to be internal? Many SERDES vendors have DC blocks integrated within their receivers and located downstream from the internal 50 ohm termination but upstream from their receiver bias network (which has much larger impedance than 50 ohms). Such placement allows the capacitor to be much smaller in value and thus integrateable and still allows the receiver to be biased independently of the transmitter which I believe is the main purpose of the DC block. Another benefit of the internal capacitor is that it will not have the associated discontinuties that an external capacitor will have.

Thanks in anticipation of your reply.

Vicente Cavanna

HP ProCurve Networking

Hi Vicente,

803.3ap really doesn’t say the caps are on the board. The cap is after TP4 and is the domain of the Rx. Yes it's a challenge for chip folks as they will need to tell there customers how to deal with this issue.

Regards,

Rich Mellitz, Intel

Hi Vicente, all:

I share your concern. I don't think that the draft standard correctly deals with the topology that you describe. I commented on this in an earlier draft of the standard.

I think we need to say what is meant by AC coupling by providing one or more specifications that place some bounds on it.

Regards,

Steve A.

Hi Vicente,Steve,All,

I have same concern about the AC coupling, according to the draft, AC compactors does belong to RX, but doesn't clearly say, whether its in linecard or in IC chip, I do think these will cause problems while we design a system. if in linecard, the pad and vias will surely cause the impedance discontinuity, different processing in AC coupling area of PCB will have different effect on the channel performance. How well we will process the AC compactors area ? As a system designer, I don't know, because the spec of backplane channel doesn't include this part.

In order to reduce the risk, I do agree Vicente that it's a better choice that place the AC coupling into the IC package.

Regards,

Jia Gongxian

Huawei Technologies Co.,Ltd

While I recognize that putting the AC coupling capacitor on chip would be desirable it is not really that easy.

The 100nF off chip coupling capacitor leads to a coupling time constant of 10 us. Such a long time constant is needed to limit baseline wander with scrambled data, as independent analysis by Rick Walker, Steve Anderson, and me indicate.

Generating such a long time constant is going to be hard in today's deep sub-micron processes. One could use, for instance, a 100pF capacitor and a 100 kOhm resistor, but size of both components would be a problem, parasitics in the capacitor will make meeting return loss specs hard and leakage current through the 100 kOhm resistor may casue offset problem.

Charles Moore

Hi Charles, All,

I recognize the problem with 64B/66B coding and AC coupling, but 8B/10B, with its guaranteed DC balance over the span of a codeword, does not appear to share that problem, and could tolerate a much much smaller time constant.

It appears that the goal to make the SERDES fairly independent of the coding may be preventing the integration of the AC coupling capacitor and thus forfeiting the many benefits of such integration. That is unfortunate for the many applications that are likely to use 8B/10B coding and that will be forced to share the same difficulties with implementation of DC blocks with channels that use 64B/66B or similar codes.

Vicente

Vince,

Our goal is not "to make the SERDES fairly independent of the coding", it is to make 1000BASE_KX, 10GBASE_KX4, and/or 10GBASE_KR work. 10GBASE_KR uses 64/66 encoding and will need coupling down to quite low frequency, 1000BASE_KX and 10GBASE_KX4 use 8B/10B encoding and can get away with quite a bit higher AC coupling cutoff frequency.

Some IEEE802.3ap compliant transceivers may not support 10GBASE_KR and would not need as large coupling capacitors and for those cases we recommend 4.7 nF.

Charles Moore

If one is considering designing the serdes to support multiple applications, is it reasonable to expect that this same interface could be used to connect to XFP optical modules?

The XFP module specification requires the DC blocking capacitors to be in the XFP module, not on the line card.

I suspect that would complicate designing a common interface for KR and XFI if KR required on-chip capacitors.

Dave

Hi Charles,

While putting the AC coupling capacitor on chip is 'not really that easy', we want to make sure we do not preclude it. Especially since we are not limited to just passive components to realize these functions.

In that vein, the best choice for putting the AC coupling capacitor on chip may involve putting the cap AFTER the termination (the configuration Vincente describes).

So the standard needs to be clear that putting the AC coupling capacitor after the termination is allowed and that for system designers, there may be DC current flowing through the termination resistors. Up until know, the description of AC coupling appears to be incomplete.

Brian

Hi Charles,

I agree with Brian that the spec should not preclude internal DC blocks (if indeed it does).

It does not seem unreasonable (in regards to the concerns you raised about size and offset voltage) to integrate a 10pF capacitor and a 5K bias resistor to obtain a 50ns time constant. Such a time constant would be adequate for an 8B/10B code (even at 1 Gbps) or for other codes whose disparity is better bounded than for the 64B/66B code.

Such a SERDES with integrated DC block capacitor, unless it had some means to bypass the DC block or to comensate for baseline wander, would not be suitable for 64B/66B codes, but would be very attractive for many other applications.

Vicente

All,

Be aware that the draft standard for p802.3ap not only does not mandate the value of an A.C. coupling capacitor but it also does not mandate the type of A.C. coupling to be implemented. How this "D.C. block" is implemented is solely up to the implementor of the link. Yes there are some test setup where A.C. coupling caps of certain values are specified, this does not mean that is how one has to implement construct a link.

Howard Baumer

Vincente and Brian,

I am new to this reflector so please excuse me if this has been brought up before.

I support allowing the implementer to determine where the DC block capacitor is but isn't it important to ensure that the implementation of the termination does not limit the signal swing at the receiver? If the capacitor comes after the termination then the common mode point of the termination effects the allowed swing. If the termination is on-die and common mode point is ground, like in PCI Express, then the single ended swing is limited by the clamping diodes of the technology. The implementer will have to ensure that the clamp can not turn on if the maximum swing arrives at the receiver. Will this be a problem in this configuration?

Also, this kind of configuration, with the capacitors after the termination, typically increase the receiver capacitance since you have the parasitic capacitance of the termination and then the parasitic capacitance of the network setting the common mode point of the receiver after the DC blocking capacitor. This creates another type of it challenge on meeting any reasonable return loss requirements.

Steve Waldstein

开放分享：优质有限元技术文章,助你自学成才