Dear all,
It is very clear based on the discussions in TF that reliability topic is a complex topic. Affirmation such as any VCSEL made for datacom during the last two decade already in the field is valid for automotive applications is misleading and equivalent to say that we have been loosing our time in the task force discussing a topic that was already not a problem. I fully disagree. The VCSEL reliability topic was already in the discussions and contributions of the study group.
The last contributions presented about reliability in the TF are inconsistent between them, showing reliability results that are different in many orders of magnitude (Horse vs Murty). And the contributions from Murty try to demonstrate that the methodology used by industry is incorrect and that using a new math is possible to demonstrate that VCSEL reliability was not and is not a problem. This is against the general methods the automotive and other industries have been used for many decades for analyzing all the failure modes that can happen in an IC. For your information, same reliability model and unreliability lognormal distribution function is behind electro-migration, however no one in the industry is trying to go against the accepted methodology that has worked to produce reliable ICs, because it would be really dangerous.
I am not affiliated with VCSEL vendor. I am affiliated with a PHY vendor company, the only one PHY vendor with optical Ethernet PHYs in series production in cars. I have tested a lot of VCSEL devices since 2019 from many manufacturers. I can affirm the reality in the qualification process, and more important, in the field will be consistent with the physical laws, regardless the math and lobby behind the decisions. This is the most beautiful thing of engineering, i.e. the reality that makes us to be humble and open to learn.
But, reliability is not the only topic behind wavelength discussion, i.e. single wavelength (with its tolerances) vs wide-range. The other topic is testing. I already sent an email to this reflector the 13th October with questions that I believe we should answer to make better decisions.
Let’s focus on SRS test (166.6.4.13, figure 166-45). With the current specification (970 ~ 990) I can argue that just using ONE reference transmitter is OK for testing the device (in P (process) V (voltage) T (temperature) characterization phase to get the design indexes for each parameter, in qualification tests, in final test, etc). Just one, like with 850nm receivers for 10BASE-SR, 25GBASE-SR, etc.
But when I have to support a wide-range, e.g. 840 ~ 990 nm: - How many different reference TXs do you think do we need in the lab? Just 850 and 980? 940nm too? Why not 880, 910, 950, 970 as well?
- As far as I know there are no tunable multimode reference TX in the market in this range. The cost of each multimode reference TX is important.
- How can I argue to the Tier1 and the OEM that just one wavelength is enough in the SRS test to validate my PHY implementation?
- Should we write in our 802.3cz standard that just one wavelength is enough in the SRS test?
- Which one and why?
- Is it up to the implementor? How can we guarantee the interoperability then?
- Is the Tier 1 and OEM going to accept single wavelength test report? Consider that 20 nm width justification is not the same of 150 nm width justification.
- Why are we adding these complications to the standard?
- Backward compatibility? — No, because automotive is a green field.
- Limitations of the fiber (CD, EMB, Att)? — No, because we are addressing 40 meters max length.
- More suppliers? Really? The yearly produced VCSEL in >= 940 nm is 2x orders of magnitude 850nm devices.
Best regards,
Rubén Pérez-Aranda, KDPOF
Dear 802.3cz Task Force participants, I am expressing an opinion regarding the subject matter of comments I-107 and I-108 against 802.3cz D3.0. I am very concerned about the REJECT response. The automotive industry will benefit from the participation of multiple VCSEL suppliers. The short optical links can be served by VCSELs of any wavelength where fiber has sufficient bandwidth. I support expanding the wavelength range to include 850 nm, 910 nm, and 940 nm VCSELs for broad market participation. Virtually all VCSELs made for datacom are at 850 - 940 nm with more than two decades of operation in the field. A lot of discussion has focused on reliability. Both Mirko Hoser (affiliated with Coherent Corp) and Ramana Murty (affiliated with Broadcom) have presented the excellent reliability of 850 nm VCSELs. Both show lifetime with margin for the automotive application. The reliability calculations presented by Ramana make the point that VCSELs are screened for lifetime and that is accurate. Several statements in the response to the comment I-108 are demonstrably inaccurate. 850 nm VCSELs do not need to be biased below 5 mA to achieve a long lifetime and there is no difficulty in achieving performance over -40 deg C to 125 deg C. Also, our experience with commercial implementation of SWDM transceivers has proven that there is no adverse impact on manufacturing or testing cost for receivers designed to accept a wide range of wavelengths. It is difficult enough to predict the future. Restricting our implementation options unnecessarily is not a wise decision. Best Regards, Vipul Bhatt Affiliation: Coherent Corp
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
|