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

Re: [802.3_B400G] Meeting Material Posted



Hi Cedric,

 

Instead of using projections, let’s use actual data. As an industry, our track record of projections that predict dramatic departures from what happened before, have not been accurate. A great example are volume projections of 400GBASE-FR4, which have been off by 4 to 5 years. This has been obscured by being lumped together with 400GBASE-DR4 volume, which is mostly 4x 100GBASE-DR and -FR, not 400Gb/s Ethernet. (As an aside, the higher volume of break-out vs. duplex is the only recent paradigm shift that I know).

 

For a developing market, we have to be careful in choosing the time frame for comparisons. 400GBASE-LR8 was the first duplex SMF interface in the market, -FR4 came years later, so it was the only option inside or outside the datacenter. If we look too early we will conclude that the ratio of LR to FR is infinite.

 

If we want to use your graphs, let’s use that portion which is based on actual shipments.

 

400G shipments.png


 

For 2022, two orders of magnitude ratio of FR to LR looks reasonable. 

 

Finally, a sanity check on any conclusion we get from statistics is always in order. 10x ratio of inside to outside the datacenter optics volume doesn’t make sense.  


Thank you

 

Chris


On Mon, Jul 25, 2022 at 9:47 AM Cedric Lam <000011675c2a7243-dmarc-request@xxxxxxxxxxxxxxxxx> wrote:
Chris and Frank:

I had checked with Vlad from Lightcounting of the volume of LR interfaces vs.  DR and FR.   He graciously let me share the following charts with the group here.   We can eyeball from his statistics that the volume of 10km interfaces is not 3 orders of magnitude smaller than that for DR and FR volume combined.  It is about 1 order of magnitude smaller. 

Vlad is willing to come to IEEE to present such market data if it helps us. 

Cedric

image.png
image.png
--
Cedric F. Lam
Cell: +1 (949) 351-2766


On Sat, Jul 23, 2022 at 6:55 PM Frank Chang <fymchang@xxxxxxxxx> wrote:
Hi Xiang, 

Thanks for your cost comment, I'd like to chime in for comparing 3 key building blocks: TOSA, ROSA and DSP ASIC. I think this argument is more analogous with the current 400G LR4-10 vs 400G ZR by doubling the baud rate.   

1) TOSA part: 
It's true 800G IMDD LR4 requires 4 lasers, but take EML as example, that leverage 4 DFB + EAM monolithically integrated chips under volume production, so no separate modulator portion is necessary. Coherent TX did require one highly powerful CW DFB laser (if not costly TEC controlled tunable transmitter with DWDM grade), but the additional MZ portion also separated using much more complex dual polarization IQ modulators and polarization properly controlled PBS (polarization beam splitter). 

Even though coherent uses one cw laser literally, its separate complex DP-IQ MZ still makes it much more costly than 4 EML CoC chips. 

Considering power consumption, coherent TX will require 2*Vpi higher swing to drive DP-IQ MZ by 4 ADDITIONAL linear MZM drivers which add extra power and cost.    

2) ROSA part

Coherent RX requires a much more complicated Integrated Coherent Receiver (ICR) and 4 dual balanced TIA with polarization properly controlled PBS (polarization beam splitter). It will be very hard to believe its cost is going to be close to the simpler IMDD RX using 4 PIN PD + 4 TIA with standard DMUX splitter.         

3) DSP part 

I assume coherent DSP will be 5-10x more complex than PAM4 DSP. Do you agree with that considering the difference in EQ taps, FW control, DSP algorithm, and FEC scheme? 

Keep in mind IMDD LR4 10km reach will piggyback a huge base of DSP (and volume) developed and deployed for 500 m and 2 km reach solutions, so no separate R&D cost associated for this specific reach. 

Best regards, 
Frank

On Sat, Jul 23, 2022 at 11:59 AM Xiang Zhou <000011dbeaa0229f-dmarc-request@xxxxxxxxxxxxxxxxx> wrote:
Hi Chris,

In your Laser power comparison, it seems you missed a factor of 4 for the IM-DD, since the coherent only needs a single laser while the IM-DD requires 4 lasers.
Additionally, to support 10km reach, tighter channel spacing like LAN-WDM4 is needed, and thus TEC is also needed for IM-DD. 

Regarding the ASIC cost. My understanding is that the ASIC cost consists of two portions: 1) the R & D cost and  2) the  BOM (Si area) cost.
To the first order, only the R & D cost  (per chip) reduces as the volume increases. As you mentioned, if we simply add a low-power LR mode to a 800G ZR chip, the R & D cost is almost 'zero' 
since the R & D is needed for ZR/ZR+ and will be amortized to ZR and ZR+ volume anyway. It is true the BOM of a coherent ZR ASIC will be larger than an IM-DD ASIC, but that portion of 
cost is a much smaller piece of the total coherent ASIC cost.

thank you
Xiang

On Fri, Jul 22, 2022 at 9:31 AM Chris Cole <chris@xxxxxxxxxxxxxxx> wrote:
 

Dear 802.3df TF Participants,

 

The July Plenary meeting had detailed presentations on 800G 10km PMD approaches, enabling further discussion of Cost and Power. 

 

Cost

 

The key PMD cost driver is the ASIC. A ZR Coherent ASIC is significantly higher cost and power than the same rate IMDD PHY ASIC (see email below). During the meeting, it was shown how to reduce ZR complexity by restricting reach to 10km, which requires a new ASIC. Unfortunately the resulting PHY volume is now three orders of magnitude lower than IMDD’s, instead of ZR’s two orders of magnitude lower, negating any comparative per die cost savings. Even more unfortunately, CMOS development cost, including masks, given projected PMD lifetime volumes raises serious questions about Economic Feasibility of a dedicated 10km ASIC. 

 

Power

 

Eric’s presentation includes TX Loss, which enables comparison of Coherent and IMDD laser DC power.

 

TX Technology

TX Loss

Link Loss

TX + Link Loss

Effective Loss
after RX LO

Unit

SiPh MZM Coherent

23

6.3

29.3

14.6

dB

InP EML IMDD

5

6.3

11.3

11.3

dB

Loss Difference

 

 

 

3.3

dB

 

A 40% efficiency of the TEC stabilizing the Coherent laser, results in the Coherent source requiring 5 times more DC power than the IMDD lasers.


Similar cost and power insights are likely behind the proposal from Chris Doerr, who understands Coherent as well as anyone, or better. For his PHY and TX, he proposes standard IMDD ASIC and EMLs, respectively.

 

Thank you

 

Chris

 

From: Chris Cole
Sent: Tuesday, April 05, 2022 5:18 PM
To: STDS-802-3-B400G@xxxxxxxxxxxxxxxxx
Cc: jdambrosia@xxxxxxxxx
Subject: RE: [802.3_B400G] Meeting Material Posted

 

Dear Task Force Participants,

 

During the 3/29/22 meeting, we started debating IMDD and Coherent alternatives for campus reach PMD.

 

In Transport, the debate was settled long ago as soon as the feasibility of Coherent was established. It’s obvious that eliminating DCFs and reducing the number of EDFAs is a good idea.

 

In Datacom, the CD, PMD and Sensitivity advantages of Coherent are not clear benefits, and Coherent is not an automatic choice. We have to do the hard work of analyzing technical and market trade-offs. Unfortunately, the 3/29/22 discussion was not a good start, so a reset is in order.

 

We know that when there is an FR interface, LR is not going to ship in the millions; it will ship in the tens of thousands. Similarly, we know that ZR interfaces don’t ship in the millions. Using LightCounting Vendor Survey Results through Q4 2021, 100G/200G Coherent volume was ~1⁄2 million over the last five years, which is less than 2% of the 100G IMDD Datacenter volume over the same time period.

 

This means neither an IMDD nor Coherent LR solution will have a dedicated DSP ASIC. IMDD 800G LR4 will use the same DSP ASIC as FR4 and DR4, because these will ship in the many millions. Coherent 800G LR1 will use the same DSP ASIC as ZR or ZR+ because those will ship in the tens and possibly hundreds of thousands. The IMDD DSP ASIC will be much lower cost than the ZR DSP ASIC because its volume will approach two orders of magnitude higher, like at 100G. We don’t have to speculate about the cost ratio of DSP ASICs at 800G; we have the answer at lower rates.

 

The specs on IMDD LR4 optics are more stringent than FR4, which raises cost. The Coherent LR TX source doesn’t have to be tunable, which lowers cost. However this fixed laser is nothing like an IMDD source, it has to be cooled and held to a precise wavelength otherwise it won’t match the RX LO.

 

The most interesting argument in favor of Coherent is that if nothing is done, the standard ZR solution works for LR, without the need for another code. This has both R&D and OpEx advantages.

 

Since the debate between IMDD and Coherent is likely to be replayed many times, let’s set a good example for the future by how we conduct it in .3df.


Thank you

 

Chris

 

From: John D'Ambrosia <jdambrosia@xxxxxxxxx>
Sent: Sunday, March 27, 2022 4:48 PM
To: STDS-802-3-B400G@xxxxxxxxxxxxxxxxx
Subject: [802.3_B400G] Meeting Material Posted

 

All,

Presentation material for our Task Force meeting on 29 March has been posted - https://www.ieee802.org/3/df/public/22_03/index.html [ieee802.org].

Please note that Mr. Williams presentation has undergone legal review by IEEE legal.  Mr. Law and I will provide guidance regarding this presentation prior to Tom presenting, and ask that all comments are held to that time.

As a reminder, All interim teleconference meeting participants should review the following documents prior to participation in an interim meeting teleconference: 

·       IEEE SA patent policy

·       IEEE SA Copyright Policy

·       IEEE SA Participation Policy

All of these policies may be found at http://ieee802.org/3/policies.html [ieee802.org].

Regards,

John D’Ambrosia

Chair, IEEE P802.3df Task Force


To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1


To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1


To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1


To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1


To unsubscribe from the STDS-802-3-B400G list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G&A=1