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

Re: [HSSG] Regarding LW & SW VCSELs



Eric.   

I assume by SW you mean the 850nm window.   I'd be interested to know whether you are proposing 4 channels at 25G per channel, or 10 channels at 10G.  If the latter is the existing experience relevant.  I assume you would need the window to be much wider or the channel spacing much narrower than is included in 10G FC.  

Regards,
Mike

-----Original Message-----
From: Eric B. Grann [mailto:ebgrann@xxxxxxxxxxxxx] 
Sent: Thursday, September 28, 2006 5:50 PM
To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
Subject: Re: [HSSG] Regarding LW & SW VCSELs

My 2 cents as well,

1) For datacenters below 100m (typically horizontal runs), pre-terminated parallel fiber is one good solution. 

2) For applications up to 300m (typically vertical runs), parallel fiber may not be the best solution.  A SW VCSEL CWDM solution may suit this distance space must better.  Trying to run pre-terminated fiber ribbon through conduit could be extremely difficult.  Also, the total cost (transceiver + fiber) cross-over point between SW parallel and SW CWDM is below 20 meters.  Both solutions (CWDM & Parallel) have tens of thousands of links in the market so both technologies are well proven.

Eric

-----Original Message-----
From: Jack Jewell [mailto:Jack.Jewell@xxxxxxxxxxxxx]
Sent: Thursday, September 28, 2006 4:29 PM
To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
Subject: Re: [HSSG] Regarding LW & SW VCSELs

My 2 cents:

1) For "data center" applications with maximum reach in the 100-300m range, the general approach of 850nm VCSEL arrays launching down OM3 fiber ribbon (or bundle) is compelling in terms of cost, power, density, technical feasibility, and overlap with similar standards/applications.  Lane rate and number of lanes are still to be determined (as is the aggregate bit rate in HSSG).

2) For 300m-ish fiber ribbons (or bundles), I see no need for 1310nm lasers of any sort for reasons of cost and fiber bandwidth (as Roger and Wenbin pointed out).

Jack


-----Original Message-----
From: Roger Merel [mailto:roger@xxxxxxxxxxx]
Sent: Thursday, September 28, 2006 4:09 PM
To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
Subject: Re: [HSSG] Regarding LW & SW VCSELs

My thinking was that ...
Since we are talking about two new ribbon cables between boxes in a datacenter using these VCSEL arrays, it doesn't really matter if the legacy cable plant is pairs of OM2 or OM1; however, OM3 is indeed the dominant MM media deployed in datacenters as of now.

Now if there is a requirement / SG objective for HS-Ethernet over a single pair of legacy OM2/OM1, that changes things entirely... but I haven't heard anyone considering / proposing this. 


-----Original Message-----
From: Mike Bennett [mailto:mjbennett@xxxxxxx]
Sent: Thursday, September 28, 2006 4:58 PM
To: Roger Merel
Cc: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
Subject: Re: [HSSG] Regarding LW & SW VCSELs

Roger,

What you say regarding OM3 is true, however, I don't think OM3 is dominant in the installed base.  Perhaps that doesn't matter if it is approaching dominance by the time we finish, but I haven't seen any data to suggest it will be. Does anyone have such data that they're willing to share with the study group?

Mike

Roger Merel wrote:
> Wenbin raises an important point.  Going with LW VCSELs on MMF would likely be a disadvantage as the OM3 is only optimized for 850nm and thus modal dispersion limited reach would be shorter for a LW VCSEL (to ~75m) vs a SW VCSEL (up to 300m or so).
>
>
> -----Original Message-----
> From: Wenbin Jiang [mailto:Wenbin.Jiang@xxxxxxxx]
> Sent: Thursday, September 28, 2006 3:26 PM
> To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> Subject: Re: [HSSG] Regarding presentation "Implementation 
> Considerations" by Roger Merel
>
> Frank,
>
> With the right volume, the cost difference between a LW VCSEL chip and a directly modulated DFB laser chip is up for debate. However, an optical isolator is normally needed for the DFB laser, but not for the LW VCSEL.
>
> In the 2nd case, 850nm VCSEL parallel link should match well with the data center reaches of <100m. The advantage of going to LW VCSEL is not obvious for this application if on MMF.
>
> Wenbin
>
>
> -----Original Message-----
> From: Frank Chang [mailto:ychang@xxxxxxxxxxx]
> Sent: Thursday, September 28, 2006 12:35 PM
> To: Wenbin Jiang; STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> Subject: RE: [HSSG] Regarding presentation "Implementation 
> Considerations" by Roger Merel
>
> Wenbin;
>
> I'm acually interested in both, but more for the 2nd case. What is your opinion regarding LW VCSEL parallel link against a DFB based CWDM link for datcenter reaches of <100m? 
>
> Frank
>
> -----Original Message-----
> From: Wenbin Jiang [mailto:Wenbin.Jiang@xxxxxxxx]
> Sent: Wednesday, September 27, 2006 6:40 PM
> To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> Subject: Re: [HSSG] Regarding presentation "Implementation 
> Considerations" by Roger Merel
>
> Frank,
>
> Do you mean to compare a LW VCSEL with a DFB for WDM or to compare a LW VCSEL parallel link with a DFB based WDM link? In the 2nd case, it is a matter of what the break-even distance is between the two links.
>
> Wenbin
>
> -----Original Message-----
> From: Frank Chang [mailto:ychang@xxxxxxxxxxx]
> Sent: Wednesday, September 27, 2006 5:41 PM
> To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> Subject: Re: [HSSG] Regarding presentation "Implementation 
> Considerations" by Roger Merel
>
> Jack et al.;
>
> The fibers.org article made a comment that "A VCSEL solution is approximately five to 10 times cheaper than a WDM solution", I guess it points to multiple 850nm VCSELs or array (if with monitoring??). I amnot sure this statement still hold true for long wavelength (1310nm or even 1550nm) VCSELs. Because of the disruptive nature in recipe or process, the LW VCSEL may have cost structure totally different from its 850nm counterpart. While LW VCSEL singlet could also be different from its array, just like the article indicate its monitoring in an array still chalenging. My question is can someone comment if LW VCSEL array can be potentially built cheaper than a CWDM or WDM solution which is well available today?  
>
> Thanks
> Frank     
>
> -----Original Message-----
> From: Jack Jewell [mailto:Jack.Jewell@xxxxxxxxxxxxx]
> Sent: Wednesday, September 27, 2006 7:16 AM
> To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> Subject: Re: [HSSG] Regarding presentation "Implementation 
> Considerations" by Roger Merel
>
> Marek,
> At OFC 2006, Picolight demonstrated a 1310nm VCSEL operating at 10.3Gbps in an SFP+ module and transmitting over 20km of SMF (10GBASE-LR with an extra 10km of fiber). This week at the European Conference on Optical Communications, AMCC is demonstrating a 10.3Gbps link over 300m of 62.5um fiber (exceeding the 10GBASE-LRM spec). It uses a similar Picolight SFP+ module with a 1310nm VCSEL and a linear receiver. AMCC's EDC chip compensates for the limited fiber modal bandwidth.
> http://www.lightreading.com/document.asp?doc_id=104440&WT.svl=wire1_6
> There have been other 10G demonstrations involving 1310nm and 1550nm VCSELs by various companies.
> Regards,
> Jack
>
> -----Original Message-----
> From: Hajduczenia, Marek [mailto:marek.hajduczenia@xxxxxxxxxxx]
> Sent: Wednesday, September 27, 2006 12:51 AM
> To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> Subject: Re: [HSSG] Regarding presentation "Implementation 
> Considerations" by Roger Merel
>
> Dear Jack,
> I think that clarifies the situation to the point beyond any doubt.
> My questions were still though not answered - is there any valid demonstration of long-wave VCSELs (1310nm, 1550nm) operating at data rate >= 3G that You might be aware of ? 
> Thanks for the answer
>
> Marek Hajduczenia (141238)
> (PhD Student - COM RD1)
> SIEMENS SA - IC
> Rua Irmãos Siemens, 1
> Ed. 1, Piso 1
> Alfragide
> 2720-093 Amadora
> Portugal
> * Marek.Hajduczenia@xxxxxxxxxxx
> *+351.21.416.7472  *+351.21.424.2082
>
> -----Original Message-----
> From: Jack Jewell [mailto:Jack.Jewell@xxxxxxxxxxxxx]
> Sent: terça-feira, 26 de Setembro de 2006 23:55
> To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> Subject: Re: [HSSG] Regarding presentation "Implementation 
> Considerations" by Roger Merel
>
> Roger,
>
> My comments regarding the fibers.org article were not in any way meant to slight the researchers whose work was reported, or their work, or to disagree with any of Jack's (Cunningham's) quotes.  I know Jack Cunningham from way back and respect him greatly.  The work they report only adds to the technological feasibility of 12x10G VCSEL arrays.  I apologize for any possible misunderstanding.
>
> My discussion regarding the status of 12x3G and the technical feasibility of 12x10G VCSEL arrays mostly comprises defensable facts, followed by a reasonably-stated opinion.  "Well on the way" indicates that key capabilities have been demonstrated, in this case: 10G VCSEL speed, significant commercial volume at 1x10G and 12x3G, reliability at 1x10G  and 12x3G, and operation at 12x10G.  Of course challenges remain, but not so fundamental as those already demonstrated.  Yield and reliability are sure to be concerns for any multi-laser approach, especially if very high performance is required.   
>
> You'll be hard-pressed to get any VCSEL vendors (or any other vendors!) to divulge the technical details of their recipes.  Instead I'll address concerns about array reliability and yield in a general and nonproprietary way.  Unless VCSELs are explicitly mentioned, these properties apply to ANY array of lasers, e.g. DFBs, FPs or VCSELs.  WEAROUT RELIABILITY: The wearout lifetime of a 12x array is usually NOT 12x shorter than that of a singlet.  It is only about 2x shorter in VCSEL experiments reported by both Infineon and AOC.  The fact that array wearout lifetime is shorter at all (for any array) is only due to variation among the elements.  If all elements aged identically, the wearout lifetime for a 12x array would be identical to that of a singlet.  Better uniformity results in better array lifetimes.  NON-WEAROUT RELIABILITY: Worst-possible-case early-failure rate is 12x higher for a 12x array compared to a singlet.  But again, the array penalty is often lessened by t!
 he!
>   nature of the root cause.  For example, ESD is one culprit for early, non-wearout, failure.  If an ESD event kills a laser array for a multi-laser product (parallel or WDM), the effect is the same as if it kills a singlet in a serial product: it kills one module.  In this case there is no "array penalty," assuming the array is equally susceptible to ESD as the singlet.  VCSEL early-failure rates are already extremely low. [Any numbers here would lead down a rathole of definitions, conditions, exclusions, qualifications, etc.]  YIELD: Array yield behaves similarly to wearout reliability.  The better the uniformity, the closer the array yield approaches the singlet yield.
>
> Since nearly all concepts discussed at HSSG involve multiple-laser sources of one kind or another, I hope the above is generally useful.
>
> Regards,
> Jack
>
> -----Original Message-----
> From: Roger Merel [mailto:roger@xxxxxxxxxxx]
> Sent: Tuesday, September 26, 2006 9:05 AM
> To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> Subject: Re: [HSSG] Regarding presentation "Implementation 
> Considerations" by Roger Merel
>
> John (goes by "Jack") Cunningham is well versed in VCSELs. The article is rather short and doesn't say very much.  They are specifically working on approaches for higher speed VCSELs.  So I'm sure the "experimental" is in the higher speeds.
>
> Jack Jewell below would suggest that single VCSEL products at 10G + arrays products at 2.5/3.125G, and a demonstration with IBM in 2003 are together proof that 12x arrays at 10G not challenging... but by doing so in such a way specifically raises the exact red flags we are all concerned about... both yield and reliability for VCSELs scale inversely with direct modulation speed (tied in part to current density)... such that unless there is some new design and new material incorporated (rather than just improved manufacturing process control), one would have to be suspect that (10x or) 12x10G VCSEL arrays are and will continue to be "challenging".
>
> With that all said, I am NOT suggesting that a 10x or 12x VCSELs should be precluded from a possible HS-Ethernet PMD.  HS-Ethernet is going to be challenging all-around.  The VCSEL array solutions are a viable avenue for some of the shorter distance applications.
>
> It would be beneficial if Jack or other VCSEL vendors would explain what has changed in the technical basis of VCSELs to ameliorate the yield and reliability concerns especially tied to modulation rate and current density.  This could provide the SG with comfort for moving from Demos to Standards for (10x or) 12x 10G VCSELs.
>
> -Roger
>
> -----Original Message-----
> From: Jack Jewell [mailto:Jack.Jewell@xxxxxxxxxxxxx]
> Sent: Tuesday, September 26, 2006 7:19 AM
> To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> Subject: Re: [HSSG] Regarding presentation "Implementation 
> Considerations" by Roger Merel
>
> The tone of the fibers.org article misleadingly suggests that the implementation of VCSEL arrays is "still at rather experimental stage."  Suffice it to say that there are well over 100,000 12-channel parallel VCSEL-array link products operating in the field at present.  Significant volumes began several years ago and keep rising.  Manufacturers and users of these parallel optical interconnections are represented at 802.3 and the present HSSG.  There is no need to discuss the technological details of monitoring (the subject of the fibers.org article) here; it is handled by various means.  The parallel optical interconnect products have undergone rigorous qualification by customers with the highest standards of quality and reliability.  VCSEL-array based parallel links are real, and this email is surely traveling through some of them.
>
> The present links typically operate at a lane rate (channel rate) of 2.7-3.3Gbps. Customer-driven initiatives are already underway to increase the lane rate to 10Gbps. IBM (with Picolight) demonstrated a 12x10Gbps link operating over 300m of MMF at OFC 2003. It used "SNAP-12" modules, i.e. the same modules used today. Since then, 10Gbps VCSEL technology has matured significantly, driven by 10GBASE-SR sales.
>
> Parallel optical links using VCSEL arrays, detector arrays, and MM parallel fibers are well on the way to showing "technically feasible" at speeds up to 120Gbps.
>
> Jack
>
> -----Original Message-----
> From: Hajduczenia, Marek [mailto:marek.hajduczenia@xxxxxxxxxxx]
> Sent: Tuesday, September 26, 2006 2:36 AM
> To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> Subject: [HSSG] Regarding presentation "Implementation Considerations" 
> by Roger Merel
>
> Dear all,
> If I recall it correctly, Roger expressed his concerns for the wavelength selection plan on slide number 8 of his presentation and pointed out that most likely the VCSEL arrays with 10 or more integrated sources can be technologically challenging. I would only like to point out that perhaps we should investigate his proposal to use integrated VCSEL arrays operating at 10+ Gbit/s -> here is why: http://fibers.org/articles/news/8/8/10. The arrays seems to be still at rather experimental stage though this is not the first piece of evidence that such devices can be achieved and may be a much cheaper solution than standard WDM systems ... 
> Looking forward to receiving any feedback ...
> Best wishes
>
> Marek Hajduczenia
> (COM RD1)
> SIEMENS SA - IC
> Rua Irmãos Siemens, 1
> Ed. 1, Piso 1
> Alfragide
> 2720-093 Amadora
> Portugal
> * Marek.Hajduczenia@xxxxxxxxxxx
> *+351.21.416.7472  *+351.21.424.2082
>
> -----Original Message-----
> From: Andy Moorwood [mailto:amoorwood@xxxxxxxxxxxxxxxxxxx]
> Sent: segunda-feira, 25 de Setembro de 2006 18:36
> To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> Subject: [HSSG] Reach Ad Hoc, call for participation and conference 
> call announce ment
>
> Dear HSSG Members,
> thank you to those who have already replied with their intent to join this activity, let me echo John's request for participation.  I want to schedule a conference call for Thursday October 5th at 11 am to 12:30 Pacific with the following objectives:
> 1. Identify what media and reach information is already available to the group for the nominated applications: "HPCC, Data Center, Metro, others"
> 2. Identify what media reach questions are of particular interest to the group.  For example, the 10 gigabit LR specification has a link budget capable of supporting a reach of 10km in many installations. Are end users actually using this budget? considerably less ? or they could use more ?
> 3. Define a schedule for future conference calls To facilitate data sharing/viewing  at the meeting we need to use some kind of collaborative environment in addition to voice conferencing.  Extreme uses "Conference Place" by Intercall and I will host this kick off meeting (If other members care to donate resources for future meetings this would be appreciated).  Live data sharing does however impose some restrictions in that a member needs to be invited to the meeting place.  
> Consequently I need a minimum of 24 hours notice prior to the meeting if a member intends to participate.
> If  a member wants to present at the meeting I need a minimum of 48 hours notice  and the slides (PDF or PPT) format Any requests not meeting these timelines will be supported on a best efforts basis.  If it is possible to support audio only participation on a "drop in basis" I will forward this to the reflector just prior to the meeting.
> Thanks for your support !
> Andy
>   


--
Michael J. Bennett
Sr. Network Engineer
LBLnet Services Group
Lawrence Berkeley Laboratory
Tel. 510.486.7913