Optical Vendors Face PIC Challenges

COMMENTS

Newest First | Oldest First | Threaded View

<< < Page 2 / 3 > >>

rationator 12/5/2012 | 3:45:05 PM

re: Optical Vendors Face PIC Challenges Hi Mostly Harmless,

"sorry to say but i think you look for excuse to critic pic approach."

Great points. I think that my post was not as well constructed as I'd hoped. I was trying to articulate that INFN's DWDM is a great PIC application. However, PIC is somewhat challenged to be adopted into other applications such as tightly spaced DWDM channels, and client side interfaces.

I don't think that you dispute the challenge of using PIC in tightly spaced DWDM channels as you did not address it in your reply. As this type of channel spacing is being used by the rest of the industry it would mean a dramatic shift in direction from the other system OEMs. This would be a great story for LR to unocver if true.

Regarding client side optics, I'm afraid I couldn't fully understand your comments.
You stated, "nobody else make pic except infinera. so anyone who use pic in system must buy pic from infinera". This was my point. For the past decade+ client side interfaces have been predominantly multi-vendor to drive down costs and create a competitive market. Shifting to a single vendor environment as you suggest, would again be a dramatic shift in direction for the other system OEMs. I think that client side PICs are >5+ years away due to many of the research funding issues highlighted in the article.

In conclusion, I think PIC is great for integrating multiple functions such as the JDSU tunable transmitter. (I hope that this can be manufactured and be available on the market.) I am just very skeptical of PIC being overly hyped and trying to be used as a solution when it is not appropriate in the near term.

rationator

Reply | Post Message | MESSAGES LIST | START A BOARD

Stevery 12/5/2012 | 3:45:05 PM

re: Optical Vendors Face PIC Challenges The 10G market took off at 2.5x the price of 2.5G

That's one point of view.

Another point of view is that the 10G market took off during the internet bubble, a time of unrealistic pricing and market prediction. The CLECs were spending crazy amounts of money on even crazier biz plans. Is that really the reference point that providers and vendors want to use?

I say again: It is just nuts to ask for cutting-edge technology at commodity prices. So let's stop with the bubble-era thinking and get real.

Reply | Post Message | MESSAGES LIST | START A BOARD

Mostly Harmless 12/5/2012 | 3:45:00 PM

re: Optical Vendors Face PIC Challenges hello rationator,

i did not answer the point on tight channel spacing because my english is terrible and i wrote enough already :-)

but i think the infinera dtn system is using 50GHz spacing, just like everyboidy else.

according to web site 800Gbps per fiber in c band. this is 80 channels at 10g so must be 50GHz pacing.

on point of client side optics i am sorry to be unclear. my point is this.

option 1. the dwdm world today is well accustomed to no interop in line side. so if we use multi-vendor pic or other optical chips then the situation does not change.

option 2. infinera make pic to be available on compinetr market. everyone uses infinera pic and so everyone interoperates.

i think option 1 will continue to be the situation tomorrow. i dont think infinera will sell there pics to competitor.

but client side interface does not use pic even on infinera dtn. so there is no problem i think for mid span meet between vedor system.

i agree that jdsu pic is very positive step forward - but is one wavelength only. how can we scale roadm system with only one wavelength? the power of infinera pic with multi wavelenbth is that many channel are available for new service or for rerouting with ospf. not possible with single wave pic like jdsu.

Reply | Post Message | MESSAGES LIST | START A BOARD

hyperunner 12/5/2012 | 3:45:00 PM

re: Optical Vendors Face PIC Challenges Hi stanleymallow,

I don't know where you're getting the info on PIC yield, but this is the kind of FUD we were told by Infinera's competition two years ago. We didn't really believe it then, and today it's clearly BS.

I will acknowledge that Infinera (like other component companies) does not talk about its yields.

I will also acknowledge that the yields are probably (make that certainly) not as good as in the Silicon world, where there's a huge wealth of experience and volume manufacturing economies of scale. But remember they're talking about 50-60 components on a chip. A Pentium has a billion!

However, I cannot seriously believe that yield is an issue for Infinera now. They're a public company with a huge market share , shipping a lot of boxes and a lot of PICs, massive revenue growth, and they're cashflow positive.

If they actually had a yield problem do you really think they could hide it?

I do think Infinera has some real challenges ahead, but I seriously doubt that yield is one of them.

By the way, what are you referring to when you say "partnerships"?

"Nobody wants to pay PhDs to run a production line that still tosses out PICS left and right."

Sorry, but what does that mean?

"Economy of scale doesn't solve this problem of production yield for the next 10+ years (not 5).

Again, what does this mean? There is no yield problem. And if there were, how do you know they can't solve it? You know as little as I do of the actual details of the process.

"By then, other analog devices will challenge the "low cost" OEO PIC on InP substrate."

Gosh, I'm afraid you're just plain wrong on that. I haven't read this Heavy Reading report but I know from our own internal studies that increasing data rates on a wavelength actually put analog devices under more strain, not less. We rejected Infinera's approach a couple of years ago, but now we're pestering our incumbent supplier to explain how analog networks will scale in the future. So far they've got bupkis. But two other systems vendors are already telling us about PICs on their roadmaps. Are they "wrong" too?

hR.

Reply | Post Message | MESSAGES LIST | START A BOARD

rahat.hussain 12/5/2012 | 3:44:58 PM

re: Optical Vendors Face PIC Challenges Does anybody have the numbers on total number of WDM nodes shipped in 2007 and the total number shipped by Infinera? This will give us an understanding of INFN's market penetration.

[Note I did not ask for number of 10G service interfaces shipped within a system (because this creates an opportunity for some fuzzy math), just the number of chassis shipped]

Also, if this is such a great solution, why are other system vendors not forcing their component vendors to provide them with the 10x10Gbs chips for a new architecture? After all, in this highly competitive market, the system vendors are pushing their suppliers on all fronts, why not this one? Or, can no other InP vendors build these devices with reasonable yield?

Is it because most of the world has accepted that ROADMs are the way to go?

Too many questions, too little time ... which is good for the INFN marketing machine.

odo

Reply | Post Message | MESSAGES LIST | START A BOARD

cw.774 12/5/2012 | 3:44:58 PM

re: Optical Vendors Face PIC Challenges My PIC frustration angers you. I think they are just neat!
1.) Yields and component count: Yield is a problem for III-V devices in combination, particulalry >3 devices. Also note a typical CMOS process transistor cost is too low to even calculate ( ok, a billionth of $90?).
A III-V discrete device cost packaged in any manner that actually functions above 6GHz costs well over $10. (note 10chx$10= $100, you'll need this figure later)
2.) Partners: the Laser Mfr. guy is not necesarily the mod mfr. guy is often not the locker guy outside of Infinera-land. From what I have been learning from potential "partners", systems guys had better stick with Infinera for a long, long time... or else.
3.) Yields: 1 of ten channels goes, the whole pic goes in the trash (10 laser, 10 mods., 10 locker, WDM) . Now see point 1.
4.) Economy of scale: see point 1 and 3 and factor a healthy dose of more 40G with a hint of point 6 below
5.) Production lines: I mean a lot, but at the heart is maintenance of a high volume re-growth/selective growth and multi-chamber reactor facility is very high skilled stuff. Didn't they tell you that? My guys do so endlessly.
I never said Infinera has a yield problem but they likely will have one if they ever ship the same # of systems that the CSCO's, ALU's and NT's of the world ship now. Oh, and everyone else sure does consider PIC yield a problem.
6.) Not wrong on OEO and increasing data rates per wavelength at all: It's understood the physics of the serial digital interconnects is reaching it's limits for electronics. Optical/Non-linear implementations will take up slack in the analog domain- completely in concert with FEC etc.. So I mean not scaling analog networks, but creating analog optical devices from some pretty old principals really.
And market share that is currently served by the oft. proprietray Si circuits now will face this competition.

What does Infinera's system cost anyway? And how does that stack up to non-PIC using competition? How much smaller do line-side systems need to be?

On the plus side, you make me think I actually should make PIC's. I hope you have a lot of lke minded colleagues that truly drive a competitive PIC market.

What does FUD mean? Like Elmer?

Thanks for the encouragement, I was really getting down on PIC's and Infinera stock pumpers.

-Stanley

Reply | Post Message | MESSAGES LIST | START A BOARD

hyperunner 12/5/2012 | 3:44:56 PM

re: Optical Vendors Face PIC Challenges Hi stanletmallow,

I'm not angry, but like you I am a little frustrated :-) I think the only reason that we don't have Infinera in our network today is that 2 years ago when we were looking for optical gear, Infinera was cleverly positioned by their competition using FUD. And FUD means "fear, uncertainty and doubt". (I think Elmer has two "d"s).

It's basically where a sales guy will introduce doubt in the prospect's mind about the other guy's product - in this case Infinera.

As I recall the remarks were along the lines of "gee, even the guys who invented PICs 20 years ago can't produce them with commercial yields - and that's just a single wavelength. Imagine the yields Infinera will get on a 10 wavelength PIC."

Don't get me wrong - this wasn't the only reason we chose another vendor. But whenever it came down to a decision, the FUD would come back and work against Infinera. Some service providers are very conservative, and FUD tends to work pretty well.

So today we have three other optical vendors in the network and we're getting outsold by companies who are using the Infinera boxes. Go figure.

PIC yield? I read your comments and I honestly don't know what the detailed answers are to these points.

But one thing I will mention...

"I never said Infinera has a yield problem but they likely will have one if they ever ship the same # of systems that the CSCO's, ALU's and NT's of the world ship now. Oh, and everyone else sure does consider PIC yield a problem."

Oops. Silly me for misinterpreting your earlier comment :-)

"Yield is low. Partnerships will be needed for some time (and all the disorganization associated with it)."

I think you'll find that they are shipping about the same number of systems as those companies. And yet Infinera seem to be able to meet demand with a shorter lead time than any of our current suppliers, and they are cashflow positive.

Honestly, I can only interpret this in two ways:

1. They don't have a yield problem.

2. Yield is "low" for a component company, but because Infinera is a system company their business model means that this is not a problem. So I guess this equates to "they don't have a yield problem".

Stanley, from your other comments I get the impression you're a component guy. So maybe it would be interesting for you to think about the potential yield issues from my point of view - a potential Infinera customer.

All I care about is:

- Does it work?
- Is it "cheap" (and this includes both purchase price and ongoing ops costs)?
- Does it help me to beat my competitors by offering services faster?
- Does Infinera have a sustainable business model?

I guess yields are a component of this last issue, but I'm going to look at the macro level (ie. the numbers, the delivery times, the roadmap etc.) because I'm simply not qualified to understand the chemistry.

And to your other question - should you be thinking about making PICs? Hey didn't you know? Even the guys who invented PICs 20 years ago can't get them to work! :-)

hR.

Reply | Post Message | MESSAGES LIST | START A BOARD

Pete Baldwin 12/5/2012 | 3:44:54 PM

re: Optical Vendors Face PIC Challenges >>>>> I think you'll find that they are shipping about the same number of systems as those companies. And yet Infinera seem to be able to meet demand with a shorter lead time than any of our current suppliers, and they are cashflow positive.

Honestly, I can only interpret this in two ways:<<<<<

Aside from the two you mention, I wonder if there's a third possibility: That Infinera gets some advantage out of owning its own fab. Even if yields are low, they don't have to wait behind someone else to get another group of wafers processed.

That's just a guess. If someone's got a better understanding of the foundry business, please do chime in.

Reply | Post Message | MESSAGES LIST | START A BOARD

Stevery 12/5/2012 | 3:44:53 PM

re: Optical Vendors Face PIC Challenges Aside from the two you mention, I wonder if there's a third possibility: That Infinera gets some advantage out of owning its own fab.

Alternatively, since fab economics are driven by volume, owning a fab is a financial liability. They are often money pits: Note that infn is still not generating any actual cash (except for stock sales) and needs to use software accounting rules.

Reply | Post Message | MESSAGES LIST | START A BOARD

paolo.franzoi 12/5/2012 | 3:44:52 PM

re: Optical Vendors Face PIC Challenges
Well, Infinera claimed around $55M/quarter in revenue last year. Not sure where the optical groups of the other companies were.

Yield is all about cost.

Cost of part = Cost of making part/ Yield

So, if yield is 33% then the part actually costs 3x what the actual cost of materials are. There are mitigations to this based around being able to abort earlier and forgoing some of the cost.

In terms of time - it just means their fab at their current yield is producing enough parts. Their absolute volumes are not huge, so that would seem to be a reasonable assumption. Think in terms of how many good parts per day their fab has to yield for them to ship on time. 10? 100? 1000? We are not talking about the millions popped off semiconductor lines daily.

From a end product cost question, the fundamental is does the part at its yield end up being cheaper when designed in by reducing the cost of the entire product by a given amount. Is this cost reduction more than the R&D to build it and continue it. If so, then it is a good deal. If not, then it is a bad deal. One could add in all the capital overhead costs of the fab, but that would be traded off against the profit given to an external component supplier.

seven

Reply | Post Message | MESSAGES LIST | START A BOARD

<< < Page 2 / 3 > >>