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

Re: [802.3_OMEGA] Motion #3 for PMD adoption. Understanding the negative votes.

Dear Ruben,

I'm afraid I cannot attend this afternoon's meeting due to a plenary, but I just wanted to underline for the TF my concerns as I've expressed to the group and to you privately in our correspondence over the last few months and why I must maintain that 980nm + OM3 has not been shown to meet the Technical Feasibility CSD.

This automotive application is not like most applications in IEEE 802.3 in that the technology we choose will directly affect the safety of billions of people. This is not an exaggeration. There are a billion cars in the world today. If even 20% of those adopt internal optical communication systems, billions of people will be dependent on these optical "nerves" between the safety critical sensors around the car, on which automotive is becoming more and more dependent, and the onboard computer where sensor input is processed and reactions executed. These optical links will therefore be critical to the safety of cars and their billions of passengers in the coming years.

The most serious cause of failure in VCSELs operating beyond 100 degrees, while being directly modulated is random failure, in particular due to so called "dark line defects" or "etch point defects".
As per Joseph's contribution in June (https://www.ieee802.org/3/cz/public/22_jun_2021/pankert_3cz_01_220621_random_failures.pdf), while the density of defects in 850nm, 940nm and 980 nm has decreased to 100 / cm2, any one of those defects can propagate given the right conditions and cause a sudden failure. The risk of defect propagation and the resulting random failure increases exponentially as we approach and exceed 100 degrees and the risk is further strongly compounded by direct modulation i.e. turning the laser on and off rapidly. This is analogous to standing on a plane of glass compared to jumping up and down on a plane of glass. Therefore the risk of random failure is much more severe under automotive conditions and needs to be addressed.

However, in spite of random failure being the most serious potential cause of 980nm VCSEL failure in the target environment, there has been no experimental, statistical or otherwise empirical evaluation in this TF of the actual risk of random failure on 980nm VCSELs under these extreme automotive conditions. We still have no idea whether it would be 0.1%, 1%, 10% or 25% (none of which would be acceptable of course) or indeed 0%.

My concern is that most of the contributions on 980 nm VCSELs have focussed on other "safer" areas such as the lower risk of wear-out failure, which is a different mechanism and, for post-burn-in VCSELs, not expected to be high. As stated in Joseph's contributions, while wear-out failure is much lower in 980 nm VCSELs compared to 850nm VCSELs, the occurrence of these etch point defects is exactly the same for 850nm, 940nm and 980nm VCSELs.

In addition to this we mustn't forget that 980nm VCSELs are largely unproven for transceiver applications. Any TF member can carry out their own quick litmus on the general availability of 980nm transceivers by simply googling "980 nm transceiver". I still found nothing apart from pump lasers, but someone mentioned that they exist so I will assume that this is the case.

As I've assured you before, I am not seeking to exclude VCSELs, I have worked with VCSELs for over 20 years, it's a technology I trust for data centres, but you cannot claim to meet Technical Feasibility when you have not addressed the greatest cause of failure in the target environment i.e. random failure. If our roles were reversed I think you would not accept the same from me (and of course you'd be right not to).

I hope you'll agree that this is a serious matter and I hope that it can be discussed without contention going forward.

Kind regards
Richard Pitwon


On Thu, Oct 14, 2021 at 7:59 PM Rubén Pérez-Aranda <rubenpda@xxxxxxxxx> wrote:
Dear Colleagues,

Related with the technical opinion expressed below, I would like to emphasize a comment that Kjersti Martino already did
in our last meetings. She commented that the -40ºC performance is a problem for 850nm VCSELs because once the 850nm 
VCSEL is adjusted to try to meet performance requirements at 125ºC, its performance at -40ºC is compromised. This comment
is consistent with my observation in the lab experiments, and it should not be ignored.

Fortunately, I already reported data in several contributions that support this comment.

https://www.ieee802.org/3/cz/public/11_may_2021/perezaranda_3cz_01a_110521_50Gbps_850nm_demo.pdf : 
  • This contribution reports experiments of 50 Gb/s transmission using 850nm VCSEL across temperature.
  • Specifically slide 5 reports the threshold current as a function of back-side temperature. Ith is a key indicator of how VCSEL performance change with temperature.
    • For your information, lower Ith is better (bandwidth, distortion, noise), and higher Ith is worse.
  • This is the typical curve that I observed in all the 850nm VCSEL that I tested. See also the following contributions of July 2020:
  • The 850nm VCSELs are optimized for room temperature (25C), and the slope of the Ith curve vs temperature is abrupt in the right and to the left sides of the optimum temperature.
  • On top of that, trying to move the optimization point to higher temperature, it would further penalize low temperature and viceversa. 

https://www.ieee802.org/3/cz/public/may_2021/perezaranda_3cz_01_0521_VCSEL_980nm.pdf :
  • This contribution reports experiments of 50 Gb/s transmission using 980nm VCSEL across temperature.
  • See slide 6 for Ith.
  • With similar Ith at 125ºC (~1 mA), the 980 device performs much better than the 850nm device of https://www.ieee802.org/3/cz/public/11_may_2021/perezaranda_3cz_01a_110521_50Gbps_850nm_demo.pdf.
  • And for low temperature, the Ith is even lower than at room temperature. You do not have the bathtub curve.
  • The dependence of Ith with temperature is reduced, so the design for operation in wide range of temperature is easier for 980nm devices.

Thank you and best regards,

Rubén Pérez-Aranda
KDPOF


El 13 oct 2021, a las 23:55, Rubén Pérez-Aranda <rubenpda@xxxxxxxxx> escribió:

Dear Colleagues,

I have received a reason behind one negative vote: "I would like to see 850nm wavelengths included in the PMD, per Ramana’s proposal.”
I appreciate the feedback. Please, send the responses to the reflector, so we can use it for consensus building in a common discussion.

Next I am going to elaborate my technical opinion about including 850nm in the PMD:
  1. Reliability topic:
  2. Assembly topic: 
    • Pluggable and AOC devices are not valid for automotive: cost, lack of volume scalability, no standard processes, not automated assembly processes (it would not pass automotive qualification auditory), EMC constraints, power consumption, space requirements, etc.
      • Automotive and data-center are very very very different applications.
    • OEMs and Tier1s expect from PHY vendors single component PHY chips that are assembled in the ECU PCB using standard processes of pick and place and reflow soldering.
    • For the optical PHYs, everything needs to be integrated in a single chip: electronics, photonics, optics, fiber alignment, etc. and this component needs to be assembled in a fully automated line. In this point, flip-chip assembly option is crucial because it allows to drastically reduce the assembly time for achieving a given chip positioning accuracy. Flip-chip is not a demand for signal integrity, but for assembly cost reduction. Nevertheless, flip-chip makes easier design for signal integrity as everyone knows (e.g. return loss in high speed lines).
    • Flip-chip assembly requires of light emitters with back emission and photodetectors with back illumination. For doing that, substrate transparency is necessary. The industry has done several attempts for solving flip-chip with 850nm devices (e.g. substrate back etching), however it has not resulted in a suitable process in terms of cost and quality.
    • Someone may think on 940nm as a valid option (see contribution), because reliability levels would be similar to 980nm. However, flip-chip assembly is a problem with 940nm due to lack of transparency in the photodetector substrate.
  3. Broadband receiver topic:
    • Broadband receiver proposal has been presented twice:
    • I will summarize reactions and requests (not satisfied by proponent) to the broadband receiver proposal already presented:
    • There is no doubt about the existence of SWDM PD. However, are they suitable for 802.3cz (automotive)?
    • SWDM PD broad spectral response is not just a matter of AR coating. It is also a matter of photons absorption, InGaAs doping, band-gap engineering, vertical structures, etc. 
      • What is the availability of such kind of photodetectors developed for SWDM and which kind of technology is used to overcome the limitations of conventional GaAs and InGaAs photodiodes
      • How many providers are in the market with this kind of PD technology? — Name of the broad band PD providers was requested to be given to the TF
      • Maturity level: this technology is very recent compared with InGaAs PDs. What is the experience of the industry about aging and PVT variations? 
      • Is this PD technology suitable for automotive application? — Internal structure, compressive strains, lattice mismatches, failure modes, operation temperature range, reliability testing, qualification
      • What is the relative cost of SWDM PD with respect to widely available and mature InGaAs PDs?
    • By specifying the receiver has to be sensible between 840 and 990 nm, it precludes flip-chip assembly option, which has been argued as advantageous for automotive application.
      • Standard die and wire bond assembly is more expensive than flip-chip when positioning accuracy is required
    • What is the value proposition of broad band PMD for automotive application?
      • Validation, PVT characterization, qualification, PD wafer-sort, and PHY final-test are going to be more expensive, because reception along the full wavelength range have to be tested. Why more expensive devices, qualification and production tests would be good for automotive application?
      • A solution based on OM3 + 980nm VCSEL have been demonstrated to be optimal in terms of technical and economical feasibility and complete. 980nm VCSELs are easier to manufacture, with lower relative cost and bigger number of suppliers (https://www.ieee802.org/3/cz/public/may_2021/king_3cz_01a_0521.pdf ). Why does P802.3cz need a broad band PMD?
      • There is no 850nm backward compatibility requirements in automotive application. We are free to choose the right wavelength, specially because fiber link is limited to 40 meters. Why do we have to choose more than one wavelength?
      • Do we think market segmentation originated by multi lambda transmitters is positive for automotive?
      • Do we think OEM is going to be happy with the extra cost of supporting broad band PMD?
      • Most important: Are we solving a real OEM necessity with broad band PMD proposal?


I am sorry for the long response, but wavelength is not an easy topic in this project. There are a lot of arguments that we, the proponents of PMD based on OM3 + 980nm VCSEL, have
considered. I hope you appreciate the effort to put all this information together; I did it to make easier to follow arguments and relevant contributions.

Thank you and best regards,

Rubén Pérez-Aranda
KDPOF


El 12 oct 2021, a las 23:13, Rubén Pérez-Aranda <rubenpda@xxxxxxxxx> escribió:


Dear Colleagues,

Today, in the 802.3cz Interim Meeting we have succeeded in the adoption of 4 baseline proposals (i.e. 50G PCS/PMA,
EEE/LPI, Loopback modes, and BER test mode). However, the most important motion, the one to adopt a PMD failed and 
most of the negative voters did not provide any reason for explaining their vote, and what for me is more important, they did 
not indicate what should be changed in that PMD baseline proposal to be accepted by them.

Today we voted the only one PMD proposal presented until now that is technically complete, fulfills the 100% of the objectives 
and is 100% consistent with the CSD responses. The other PMD proposals (i.e. GIPOF and Si-Photonics) are not technically 
complete (at least today), and either do not fulfill none of the objectives or only some of them. I understand that someone can vote
against option “A”, when he/she offers an option “B”, being both A and B complete and consistent with the objectives and CSD. 
However, I can't understand a negative vote against option A with nothing to offer (at least today) and with no arguments. 

Which part of the motion text and referenced contributions should be changed, with proposed remedy, in order the motion being accepted? 
Should the PMD based on OM3 + 980nm VCSEL not be included in 802.3cz draft and the task force should wait until other PMD options are complete and meet (new) objectives? Why (technical reason)?
Should we have also included the other PMD options in the motion, even if they are incomplete and inconsistent with objectives, going against CSD?
Did the motion text or the referenced contribution prevent the adoption of future PMD options like GIPOF and Si-Photonics (provided the objectives and CSD are revised first)?

Therefore, I kindly ask the following individuals to provide arguments for supporting their negative vote against the motion #3
for PMD adoption of today. These arguments may be used to identify gaps in understanding the procedures and building 
consensus in the future.

  • Tadashi Takahashi, Nitto Denko Corporation
  • Kazuya Takayama, Nitto Denko Corporation
  • Kenji Yonezawa, AGC
  • Hidenari Hirase, AGC
  • Yuji Watanabe, AGC
  • Satoshi Takahashi, POF Promotion
  • Ichiro Ogura, Petra
  • Taiji Kondo, MegaChips
  • Hideki Isono, Fujitsu Optical Components
  • Kenneth Jackson, Sumitomo

Thank you and best regards,

Rubén Pérez-Aranda
KDPOF

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


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


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

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