Hello Isono-san,
Thank you for clarifying which set of cost
numbers we should use for discussion.
Your email highlights a confusing point in
the traverso_02_0708 presentation. The conclusion for 2010 Serial (at 120K
volume) is 1 x CWDM 2010 cost for GPPO IF, and 0.78 x CWDM 2010 cost for PCB
(GPPO-less) IF. The conclusion for 2010 CWDM is 6 x 10GE LR 2010 cost.
Lightcounting data (see cole_04_708) gives
the 2008 40G VSR module cost as 48 x 10GE LR 2010 cost (= 40 x 10GE 2008 LR
cost.)
This means that there is an 8x reduction
in cost from 2008 to 2010 for GPPO IF based module, and an additional 1.3x (10x
total) cost reduction for GPPO-less IF based module. Page 16 of
traverso_02_0708, identifies main drivers for this drop in Serial 2010 cost:
-
Optics packaging
-
4:1 SerDes
instead of 16:1 SerDes
-
Low cost
SerDes packaging
-
Low cost
RF interconnect
-
Higher
Volume
There is an in-depth discussion of low
cost GPPO-less IF packaging and interconnect technology on pages 6, 7, 8, 9 and
10 of the presentation, which supports three of the above bullets. However, this
addresses the 1.3x (of the 10x total) cost reduction factor since it applies to
GPPO-less IF.
This leaves two other above bullets to
account for majority of the 8x cost drop in two years.
There is no specific discussion in the
presentation of why a 4:1 SerDes is cheaper then a 16:1 SerDes, although
comments were made during Q&A in Denver
that the I/O count is reduced. Since there is general agreement that SerDes die
cost is a small fraction of the overall cost, this presumably is a minor
component of the 8x cost drop.
This leaves the Higher Volume bullet to
account for the majority of the 8x cost drop in two years, with page 13 giving
the volume assumption as 120K in 2010. In his 8/2/08 email, Takai-san estimated
the cumulative 40G shipment as 10K. This gives a volume increase of 10x to 20x,
depending on exact annual assumptions.
A general rule is that a 10x increase in
volume results in 2x drop in cost.
Therefore, it is very hard to see how the
8x cost drop in two years of GPPO IF based 40G module can be justified by the projected
10x to 20x volume increase.
A much more reasonable conclusion is that there will be a 2x to 3x cost drop in
two years, as projected in cole_08_0708, page 9, and traverso_04_0308, page 8.
Chris
From: Hideki Isono
[mailto:isono@xxxxxxxxxxxxxx]
Sent: Tuesday, August 05, 2008
7:58 PM
To:
STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3BA] Discussion
on 40G for ="" 10 km SMF
Alessandro-san
Regarding the cost difference you pointed out, the background reason isas follows.
a) In this proposal timing, the specific 40GE volume is not discussed,and also
the existing technology adaptation is assumed. This is the reason whythis
estimation is too high.
b) From this estimation, new technology such as low cost TOSA/ROSA(XLMD)
and also low cost
SerDes package are assumed.
As the result of these assumption, the cost becomes very close to thelatest
estimation.
C) From this estimation, the specific volume (120K pcs for 2010)
isassumed and GPPO-less
package is optionally introduced. We concluded that the cost is 0.78 xCWDM for
GPPO-less and
1 x CWDM for GPPO IF.
Estimation described in (C) is the latest and the most accurate one.
Please refer to this document hereafter.
Best regards,
Hideki Isono
Fujitsu Ltd
At 22:01 08/08/04 -0700, Alessandro
Barbieri (abarbier)wrote:
Hi
Atsushi,
> I am afraid that to chose 40GbE CWDM will ignore data centerapplications
that is cost sensitive and may require less than 4x10Gcost.
or
>"We have to resolve this to achieve <2x10G cost."
it is not clear how you derive the conclusion that 2X is needed for thedata
center space. Is it a gut feeling or is there a rationalexplanation?
At least below I attempted to articulate briefly why 4X if veryreasonable on
the optics (which BTW is just a part of the total systemcost) to ensure market
success for 40G SMF and I would like to understandif you have any specific
disagreement:
"I think ~4X the cost of 10G with *just* the benefit ofconsolidating 4
metro fibers will work quite well.
On top of it add the operational advantage of simplifying the network
byreducing the dependency on LAG by a factor of 4 and one could argue that4X on
SMF is perfectly fine."
Last I am now getting confused with this latest 2X 10G cost on top of allthe
cost projections presented on 40G serial:
a) http://www.ieee802.org/3/ba/public/mar08/traverso_04_0308.pdf :
2012 Serial still more than 1X CWDM
b) http://www.ieee802.org/3/ba/public/may08/jewell_02_0508.pdf:
2011 Serial is 1X CWDM
c) http://www.ieee802.org/3/ba/public/jul08/traverso_02_0708.pdf
2010 Serial is 0.78X CWDM
d) Now in 07/08 the claim is 2X 10GBASE-L which is anywhere between 0.5Xand
0.3X CWDM (even lower than traverso_02_0708 in 2012)
Why the story keeps changing on serial?
Thanks,
Alessandro
> -----Original Message-----
> From: Atsushi Takai [mailto:atsushi.takai@xxxxxxxxxx]
> Sent: Monday, August 04, 2008 5:48 PM
> To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> Subject: Re: [802.3BA] Discussion on 40G for ="" 10 km SMF
>
> Team
>
> First of all, in case of CWDM, volume will never resolve the issue.
> The cost will be always 4x10G+WDM.
> The cost will go down according to 10G.
>
> I agree with Chris's comment that integration will not
> resolve the cost issue in case of 40GbE CWDM.
> We also had experiences of such technologies and found difficulties.
> Thus in case of 40GbE CWDM, The cost seems 4x10G+(WDM).
>
> ((100GbE WDM is different.
> In case of 100GbE WDM, the cost will be considered almost 4x25G+(WDM).
> Thus to achieve <10x10G cost is to achieve 25G cost to be
> closer to 10G.
> We believe we can achieve it.))
>
> It seems that CWDM has no way to achieve 4x10G and may be 6x
> with reasonable estimation.
> I am afraid that to chose 40GbE CWDM will ignore data center
> applications that is cost sensitive and may require less than
> 4x10G cost.
>
> I agree that even 4x10G is much cheaper than current cost of
> 40G 2km serial.
> But it is not the discussion point.
>
> In case of serial, volume and well-known technology will
> resolve the cost issue.
> (1) Today's market is small, may be < 10K/year.
>
I believe team confirmed market.
>
Also carrier started to install 40Gbit/s transport
> systems recently.
>
We are receiving 40Gbit/s Infinibandthat is another market.
>
Volume pulls the investment for thetechnology and it is
> happening.
> (2) Optical device companies that I discussed have no concern
> on technology.
>
Especially this is 1310-nm transmission.
>
(I hope you understand this is very important)
> (3) IC cost is always related volume to compensate investment.
> (4) Thus only big challenging technology is interconnection
> technology.
>
We have to resolve this to achieve<2x10G cost.
>
This is the interconnection betweendriver to Laser module
>
with the length of about 1 inch orless
>
that 40G signal go from driver chip tolaser chip.
>
We have resolution today semi-ridgedcoaxial cable.
>
However we need low cost alternativeto achieve such low cost.
>
We may overcome using customizeddesign with IC and module.
>
However we need universal design toreduce cost and wide usage.
>
Anyway I am optimistic for this.
>
> I believe we have to choose serial.
>
>
> =~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
> Atsushi Takai
> Marketing Division, Opnext Japan, Inc
>
> =~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
> ----- Original Message -----
> From: "Alessandro Barbieri
(abarbier)" <abarbier@xxxxxxxxx>
> To: <STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx>
> Sent: Tuesday, August 05, 2008 8:00 AM
> Subject: Re: [802.3BA] Discussion on 40G for ="" 10 km SMF
>
>
> Hi Frank,
> I have a couple of brief of comments inline...
>
> > Steve;
> >
> > I share with your viewpoint, but still have similar concerns,
> > so put us in the category of "undecided" regardingthis
after
> > listening to customers from either datacom or telecomside.
> >
> > The problem is if neither 40g serial or 4x10 CWDM canprovide
> > a cost trend more favorable than 4x individual 10ge, then
> > 40gbe will be hard to take off, everyone may stick to install
>
> I think ~4X the cost of 10G with *just* the benefit of
> consolidating 4 metro
> fibers will work quite well.
> On top of it add the operational advantage of simplifying the
> network by
> reducing the dependency on LAG by a factor of 4 and one could
> argue that 4X
> on SMF is perfectly fine.
>
> > 10ge a bit longer, especially for data centers (which is more
> > cost sensitive), so 40gbe SMF may eventually unable to build
> > up significant volume. Think about the 10gbe volume
> > difference regarding LX4 vs. 10ge serial.
>
> Comparing the volumes of optics destined to different
> applications is not an
> apple-to-apple comparison. Even though LX4 works on SMF, I
> believe less than
> 5% use it for that purpose.
>
> Thanks,
> Alessandro
>
>
> > 4x10g CWDM option may provide a competitive cost point from
> > day one, are we underestimating the LX4 mfg issues in terms
> > of photonics integrated circuits to drive further cost down?
> >
> > Are we too optimistic on 40g innovation for cost reduction
> > (obviously maybe lengthy and expensive development) keeping
> > in mind close to limits of current electronics?
> >
> > Feel like both 4x10g CWDM and 40ge serial face "breakthrough"
> > ahead. Without any reasonable/realistic consensus with these
> > "hard" data, seems there would be difficult for thegroup to
> > reach the decision unanimously.
> >
> > The likely scenarios is if both 40g serial or 4x10 CWDM
> > cannot be built more cost-effectively than 4x individual
> > 10ge, 40ge SMF will have very limited time window, then
> > people will escape it and jump straight to 100GE.
> >
> > My 1 cent.
> > Frank
> >
> > -----Original Message-----
> > From: Trowbridge, Stephen J
(Steve)
> > [mailto:sjtrowbridge@xxxxxxxxxxxxxxxxxx]
> > Sent: Monday, August 04, 2008 8:50 AM
> > To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> > Subject: Re: [802.3BA] Discussion on 40G for ="" 10 km SMF
> >
> > Hi Mori-san and others,
> > It goes without saying that the cost of a 40G 4x10G CWDM
> > transceiver will not over the long term fall below 4x the
> > cost of a 10G transceiver, but for a very trivial reason: Any
> > cost reduction that results from development or volumes of 40
> > GbE will also reduce the cost of a quad 10G transceiver. So
> > this is a meaningless comparison and not helpful for the
> > decision. The decision needs to be made based on how the
> > costs of 4x10G CWDM and serial 40G compare to each other, not
> > how they compare to the cost of 10G. 40G serial technology
> > has been in the market for ~6 years, and is still stubbornly
> > expensive. Costs are finally decreasing somewhat, but the gap
> > is not being closed vs. 10G because the cost of 10G is
> > decreasing even faster than the cost of 40G.
> > Regards,
> > Steve
> >
> > -----Original Message-----
> > From: Kazuyuki Mori [mailto:mori.kazuyuki-1@xxxxxxxxxxxxxx]
> > Sent: Sunday, August 03, 2008 10:21 PM
> > To: STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx
> > Subject: Re: [802.3BA] Discussion on 40G for ="" 10 km SMF
> >
> > TF members,
> >
> > I'm Kazuyuki Mori, Fujitsu Labs. I support 40G Serial and I
> > basically agree with Takai-san regarding below points.
> >
> > (1) Cost
> >
My understanding is that, generally speaking, the final
> > cost target for 40G optical transceiver should be 2 -3 times
> > of 10G transceivers. In order to achieve this target , I
> > again and again discussed the cost reduction approaches with
> > TOSA/ROSA suppliers, IC suppliers, optical module suppliers
> > and our laboratory experts. In case of CWDM, I haven't found
> > any technical solution to achieve this target (<x4 cost of
> > 10G transceiver), and there has been no presentations in IEEE
> > to show this cost reduction approaches until now. On the
> > other hand, 40G serial is feasible to achieve this target as
> > shown in traverso_02_0708.
> >
I wonder that 40G transceiver cost will remain more than 4
> > times of 10G in future if once CWDM solution is authorized.
> >
> > (2) Size
> >
I think the size reduction is another big challenging
> > target in CWDM as Chris already agreed in recent dialogues.
> > Some people say that monolithic DFB array enables high
> > density package solution, but CWDM option is almost
> > impossible to be realized because the same active layer ofLD
> > cannot be applied. Also the hybrid integration using PLCwith
> > an integrated AWG MUX is sometimes picked up, but it isquite
> > challenging due to high insertion loss of AWG caused by
> > intrinsic Gaussian profile, and also due to AWG temperature
> > dependence. In my perspective as a researcher, optical
> > integration approach in 40G CWDM has some intrinsic problems
> > and leads the cost increase.
> >
Please remember that this isn't the case of Vcsel array,
> > but the case of DFBs and also with optical mux.
> >
> > (3) Power
> >
Steve pointed out that '40G SerDes are very power hungry',
> > but this is not correct. Current SerDes is for 16:1 and 1:16,
> > however 4:1 and 1:4 SerDes
> > should be asuumed in We need to
compare using 4:1 and 1:4
> > SerDes. In our
> > estimation, 2W is possible by deleting unnecessary circuits
> > from today's SerDes even when SiGe was used.
> >
> > Kazuyuki Mori
> >
> > ----- Original Message -----
> > From: "Atsushi Takai" <atsushi.takai@xxxxxxxxxx>
> > To: <STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx>
> > Sent: Sunday, August 03, 2008 12:34 PM
> > Subject: Re: [802.3BA] Discussion on 40G for ="" 10 km SMF
> >
> >
> > > Jeff
> > >
> > > I do not know your background.
> > > However I found, in your comment below, you misunderstood
> > the optical
> > > transmission technologies.
> > > I do not want to argue line by line.
> > >
> > > Just I would like to point one sentence:
> > > "The biggest downside of 40G serial, seems to be the
> > physics problem of
> > > PMD.".
> > > This is not true for 10km SMF.
> > > The 40Gbit/s PMD was a technical challenge in several years
> > ago but now it
> > > is not downside.
> > > Even, we are discussing 1310-nm devices while current
> > module includes
> > > 1550-nm devices.
> > > (We can neglect dispersion issue in case of 1310nm transmission)
> > >
> > > The biggest cost in current serial module is silicon chip
> > that is much
> > > more
> > > volume sensitive.
> > > I hope you know the accumulative shipment of 40Gbt/s client
> > module is
> > > around
> > > 10K peaces or such range.
> > > However IEEE confirmed market of 40GbE 10km serial enough for
> > > standardization, you can expect much lower cost with
> higher volume.
> > >
> > > Also investment for 40Gbit/s transmission networks started
> > these years,
> > > thus
> > > industry started invest for 40Gbit/s technologies.
> > > You will find much activity in the industry.
> > >
> > > All
> > >
> > > I am very concerning during the meeting and e-mail discussion,
> > > many of members may misunderstand the 40Gbit serial
> > technology status and
> > > activity in the industry,
> > > and understand only the surface.
> > > We, who has technology especially optical device
> > technology, should be
> > > responsible to let people understand the technology.
> > > I will think about it.
> > >
> > > =~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
> > > Atsushi Takai
> > > =~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=
> > > ----- Original Message -----
> > > From: "Jeff Meyer" <jmeyer@xxxxxxxxx>
> > > To: <STDS-802-3-HSSG@xxxxxxxxxxxxxxxxx>
> > > Sent: Sunday, August 03, 2008 2:33 AM
> > > Subject: Re: [802.3BA] Discussion on 40G for ="" 10 kmSMF
> > >
> > >
> > > Chris has a excellent point about
> > >
> > >
_Other Aspects_
> > >
> > >
It is no longer possible to simplyincrease Baud to
> > match data rate,
> > >
because of fundamental electrical andoptical
> propagation limits.
> > >
This was recognized during the 100G SMF PMD discussion,
> > with Serial
> > >
never a viable alternative for the 10km or 40km reach. In the
> > >
future, all data rates beyond 100G will use some form of
> > multi-lane
> > >
technology. 40G is the inflection point where cost and
> > difficulty of
> > >
Serial rises dramatically compared to multi-lane alternatives.
> > >
Optical communication has reached the point that all
> > other forms of
> > >
communication (wired or wireless) reached many years ago, where
> > >
simple modulation format serial solutions are not practical.
> > >
> > >
> > > The biggest downside of 40G serial, seems to be the physics
> > problem of
> > > PMD. However there are an increasing number of long haul
equipment
> > > providers who have solved this problem. There have been
> > thousands of 40G
> > > serial long haul installations deployed to date.
> > >
> > > As far as the Cost, Power, Size & Reliability I think this
favors
> > > serial. The cost saving of CWDM seems largely driven by the
> > large number
> > > of vendors providing 10G IC's and components. But let us
> > ponder, if the
> > > 10GE fathers chose 4x 2.5G WDM to reduce risk in the late
> > 1990's would
> > > we be benefiting from the low costs and the large number of
> > vendors? All
> > > we need is multiple vendors of 40G serial components and
> > the prices will
> > > plummet. Lets face it the cost of SiGe is not that much
> > higher than CMOS
> > > unless you get to volumes greater than 100,000 parts. By
> then, CMOS
> > > processes will catch up to SiGe in FT. I am a microwave guy
> > and the 40G
> > > packaging is not difficult these days ( there are many
> > vendors that can
> > > do LTCC fine line packages and they are "Open Tooled"
so
> > you can get a
> > > reference design for the 40G electrical packages for no
> NRE ). If we
> > > compare microwave packaging to flip chip mounting of lasers
> > and optics,
> > > I would imagine optics costs more, but I have no "hard
> > data" to support
> > > this.
> > >
> > > The biggest reason why I favor serial over CWDM is the
> > leadership for
> > > the future. Lets take the risk like the 10G serial
> > innovators did in the
> > > late 90's. Once we get several manufacturers of 40G parts
> > this prices
> > > will plummet.
> > >
> > > Schedule Risk. Albeit the risk for serial is higher but how
much?
> > >
> > > Let's keep technology moving forward for the future generations.
> > >
> > >
> > >
> > > Jeff Meyer
> > >
> > >
> > >
> > > Chris Cole wrote:
> > >
> > >> Takai-san痴 7/31/08 email discusses a
number of points. Our
> > arguments
> > >> concerning his first two points (Cost and Time to Market)
are
> > >> unchanged from cole_04_0708, so are not repeated here. The
> > remaining
> > >> points are addressed below.
> > >>
> > >> _Power_
> > >>
> > >> The long term power consumption of 40GE CWDM and 40GE Serial
is
> > >> similar. Four 10G un-cooled DFBs and associated Laser
Drivers use
> > >> about the same power as one cooled 40G EML and
> associated Modulator
> > >> Driver. The remaining ICs are also about the same if
> > advanced process
> > >> nodes and new designs are assumed. As was pointed out by
> > Joel Goergen
> > >> during the Q&A session in Denver, a 40GE Serial block diagram has
> > >> comparable circuitry to 40GE CWDM block diagram when drawn
> > fairly to
> > >> permit apples to apples comparison.
> > >>
> > >> There is no basis for a claim at this late stage in the
> debate that
> > >> Serial has a power advantage over CWDM, and that CWDM
> > 菟ower reduction
> > >> plans are invisible.・In jewell_03_0508,
p.9 and again in
> > >> traverso_02_0708 p. 12, ratios of power between an
> > aggressive Serial
> > >> implementation and CWDM implementation are 0.96 and 0.97,
> > i.e. clear
> > >> statements in pro-serial presentations that there is no
> advantage.
> > >>
> > >> _Size_
> > >>
> > >> For future generation products, CWDM has an advantage over
> > Serial for
> > >> fitting into a smaller form factor like QSFP because similar
to a
> > >> 10GE-LR SFP+, the re-timing CDRs can be moved outside of
> > the module.
> > >> Serial always has to have the 4:1 SerDes function in the
> > module. Even
> > >> with aggressive projections about future component size
> and power,
> > >> Serial has a packaging and thermal management design
> > challenge to fit
> > >> into QSFP.
> > >>
> > >> What is required to fit 40GE CWDM into QSFP is optics
> integration.
> > >> This type of technology has been described in numerous
> > presentations
> > >> to the HSSG and involves flip-chipping lasers onto a PLC
with an
> > >> integrated AWG Mux. The CWDM grid prevents use of a
> monolithic DFB
> > >> array and requires flip-chipping discrete DFBs, but that
> is a yield
> > >> and cost issue not a feasibility or size issue. The time
> > line for such
> > >> an advanced development program is lengthy, but is similar
to
> > >> realistic PCB RF-interconnect 40GE Serial development
> > schedules. The
> > >> investment required to bring this advanced technology to
> market is
> > >> high, again similar to one required for low cost 40GE
Serial.
> > >>
> > >> In contrast, no advanced technology development is
> > required to quickly
> > >> bring to market first generation low cost CWDM products
based on
> > >> discrete optics packaged in a larger form factor.
> > >>
> > >> _Reliability_
> > >>
> > >> There is no current 1310nm 10G DFB failure data that
justifies
> > >> bringing up concerns about the reliability of a 4x10G
> CWDM PMD. 10G
> > >> 1310nm PMDs ship in volume today with very high
> > reliability. If there
> > >> is actual field failure data behind this concern, it would
add
> > >> credibility to have it presented.
> > >>
> > >> _Other Aspects_
> > >>
> > >> It is no longer possible to simply increase Baud to match
> > data rate,
> > >> because of fundamental electrical and optical propagation
> > limits. This
> > >> was recognized during the 100G SMF PMD discussion, with
> > Serial never a
> > >> viable alternative for the 10km or 40km reach. In the
> > future, all data
> > >> rates beyond 100G will use some form of multi-lane
> > technology. 40G is
> > >> the inflection point where cost and difficulty of Serial
rises
> > >> dramatically compared to multi-lane alternatives. Optical
> > >> communication has reached the point that all other forms of
> > >> communication (wired or wireless) reached many years ago,
> > where simple
> > >> modulation format serial solutions are not practical.
> > >>
> > >> Chris
> > >>
> > >>
> > >
> >
>