Post of the Day
August 18, 2000

Board Name:
l'union fait la force

Posts selected for this feature rarely stand alone. They are usually a part of an ongoing thread, and are out of context when presented here. The material should be read in that light. How are these posts selected? Click here to find out and nominate a post yourself!

Subject:  Re: Broadcom
Author:  rchampoux

With BRCM acquiring Newport, it seems that BRCM is now a major player in the OC-192 systems - previously a domain controlled by AMCC. I would be interested in any thoughts on this move by Broadcom and what it would mean for players like AMCC.

I guess I can try taking a stab at this one, although you all should beware when a lawyer tries to take on a discussion of opto-electronics. In other words, we're talking "amateur hour" here. But here goes anyway:

First of all, there are four commonly used processes for manufacturing high-speed communications chips. First there is CMOS, which is the cheapest, easiest to manufacture and most efficient in terms of power usage, but also the most difficult to obtain higher speeds from. Next in order is bipolar, then comes silicon germanium (SiGe), then finally gallium arsenide (GaAs). The art and science in making these high speed chips is in the cost-benefit analysis between the need for speed, versus cost, difficulty of manufacture and greater power usage.

At present, the two main competitors in the long-haul, high-speed, optical / SONET / DWDM chip business are AMCC and VTSS. To oversimplify, VTSS was the earlier entrant, initially hung its hat on GaAs for its highest speed products, and is now evidently getting into SiGe on a limited basis. AMCC always preferred SiGe for its highest speed products. However, AMCC (and I would assume, VTSS, although I don't know for sure) always uses bipolar or CMOS whenever possible, as the cheaper, more easily manufactured, and more power-efficient alternative. VTSS initially was much larger than AMCC, but AMCC now appears to be catching up and is growing revenues and earnings at a faster rate. Last quarter, VTSS reported revenues of $114 million and pro forma income of $32.1 million, versus AMCC's revenues of $74 million and pro forma income of $27.9 million.

The primary market for AMCC and VTSS are chips where electronics interface with optical fiber. These chips basically manage a number of tasks related to optical-electronic conversions at very high speeds, including amplifying and/or grooming the signals, coordinating the timing of the incoming optical signals with the outgoing electrical signals, pointing the signals in the right direction (if the incoming signal is optical, there will often be one optical port for the incoming signal, but four or sixteen outgoing ports for the electrical signals), and converting the optical signals from analog (or serial) format, to digital (or parallel) format. To explain the last function further, when an optical signal comes in over fiber, it's like a very high speed Morse Code, where the light is either on or off. However, my understanding (and I may be wrong on this) is that when the signals are converted to electronic signals, the fact that the electronic signals are communicated in bytes (8 separate instances of "1"s or "0"s) means that the signals need to be converted into a parallel form. This involves all kinds of interesting timing and mapping issues.

So when you go to AMCC's product page, you will see a series of different products for different stages of the optical-to-electrical (or vice versa) conversion.

These different stages are often referred to as layers, which I will discuss in order beginning at the closest layer to the optical fiber. The first layer is the physically dependent media (or PMD) layer, which is referred to on the above link as "Opto-electronics". This accounts for a small but growing portion of AMCC's revenues. Next is the physical layer, which is the greatest source of AMCC's revenues at present. This is referred to on the link as "Transceivers SER/DES/CDRs." Transceivers are essentially a combination of receivers and transmitters. Ser-des is serializer-deserializer, which is the conversion from optical/analog/serial to electronic/digital/parallel that I discussed above. "CDRs" is clock and data recovery units, which is recovering data and timing signals from the SONET transmission. AMCC estimates that the overall, existing potential market in the physical layer (a lot of which is still handled in house by companies like Nortel and Lucent) is about $400 million per year.

The next layer is "framer-mapper," referred to on the link as "overhead processors." In AMCC's case, this was the focus of the Cimarron acquisition last year. This is the process of grooming a newly-digitized signal and pointing it in the right direction on its way out of the system. This is a new area for AMCC, but the overall potential market is estimated at $800 million per year, so AMCC is pushing aggressively into it. The next layer above that is the serial backplane, which is essentially the internal switching architecture where various signals are directed toward various ports, on their way out the system. This I believe is the rough area targeted by AMCC's recent acquisition of Yuni Networks. (I'm still unclear on the distinction between serial backplane and switch fabric, although I believe that they are the same thing.)

On the AMCC products page linked above, if you click on any one of the functions in the various layers described above, you will see a series of products that AMCC offers its customers. The beauty of AMCC's business is that a lot of the higher-level functions that used to be handled by software are now incorporated directly onto silicon. This, plus the time-to-market advantages offered by dedicated chipmakers such as AMCC, means that customers are flurrying to buy these chips. Since the demand for bandwidth has been estimated as more than tripling each year (essentially, a 10x increase every two years), and AMCC's business is actually increasing at a rate faster than its equipment manufacturing customers (e.g., Lucent and Nortel) because the trend of these customers is to outsource more and more of the chip design and engineering, you can see why AMCC is in such a sweet spot and why BRCM seems to be exhibiting an interest in it.

Unfortunately, there is no easy way to display all of AMCC's product line, but essentially it would appear to be at least 100 different products, addressing various layers, various functions within each respective layers, and various speeds. VTSS's product list shows a similar broad variety of product line.

And since we're at it, why not take a look at PMCS's product line, where you can see that the main orientation is toward ATM equipment and slower-speed SONET equipment.

The bulk of the long-haul optical communications occur at speed of OC-48 (2.5 gigabits per second) although a faster standard of OC-192 (10 Gb/sec) is becoming more common. VTSS and AMCC offer products at both speeds. PMCS, the bulk of whose revenues are derived from ATM switches, also manufactures some products for SONET optical equipment. Mostly, these are at the OC-3 and OC-12 level, although PMCS is starting to break into long-haul OC-48 on a limited basis. Bob Bailey, PMCS's CEO, has stated (during the 4/00 conference call) that PMCS has no plans to manufacture SiGe chips, and if necessary would "partner" with someone to do so.

Incidentally, "slower" in no way implies "inferior" as to PMCS, because the reality is that for every piece of OC-48 or OC-192 optical equipment, there will probably be several ATM switches or other pieces of metro equipment at OC-3 or OC-12 speeds. So if JDSU is the optical trunk of the tree and AMCC is the main limbs that attach to the trunk, then PMCS is the branches coming off of those limbs. (And BRCM is the twigs off the branches, where the broadband network enters into the home, office, or LAN). So they are really complimentary businesses, although PMCS (like BRCM) seems to be exhibiting an interest in the higher speed products where AMCC is especially strong.

OK, now keeping in the mind the product line of AMCC, VTSS and PMCS (you really need to visit those links to make sense of this) now let's look at the product line of NewPort Communications, Inc.

It looks like we have one chip in production and two on the way. The chip that is in production - an OC-48 transceiver - will be available in 8-10 weeks. The two chips that are on the way are an OC-192 transmitter and an OC-192 receiver (essentially, you're performing the transmitter and receiver functions on one chip - a transceiver - with the OC-48 speed product.) The OC-192 pair are sampling now, and will be in production in the last quarter of this year. NewPort has not announced any products for the PMD or framer/mapper layers.

The big theoretical advantage of NewPort's OC-192 transmitter and receiver pair, as well as its OC-48 transceiver, is that it uses CMOS technology (again, remember that this is cheaper, easier to manufacture and more power efficient) rather than the SiGe used by AMCC. Unfortunately, the pricing on the OC-192 products hasn't yet been announced. And, I'm not enough of an expert to rate the NewPort OC-192 transmitters and receivers on a technical basis against those of AMCC. However, the specs of both are available on each company's respective web sites, so anyone who wants to should easily be able to take a stab at it.

The cost of the chip that is in production - NewPort's OC-48 transceiver - is $245 in quantities of 100 or more.

AMCC offers several SiGe OC-48 transceivers. The most recent one - which to my untrained eye appears to be more functional, because it supports multiple rates of speed in addition to OC-48 - costs $145 in quantities of 1000 or more.

I guess the best conclusion that I can draw is that NewPort's product line at present is limited, that any apparent cost savings does not yet appear evident, and I am unqualified to discuss whether NewPort's product is technically better (although I would assume that it would use lower power.)

Does the NewPort acquisition mean that BRCM is now the "major player" at OC-192? Not in the least. Again, NewPort's product portfolio is extremely limited, and it has not announced design wins or revenues from its existing customers. AMCC is generating substantial revenues from OC-48 and OC-192 at present, and more importantly it generated design wins in the last quarter whose present economic value was quadruple that of its recognized revenues. AMCC's existing design wins will produce significant revenues over the next five to seven years (the typical duration of a chip product placement).

The real concern is AMCC potentially losing the "next generation" of design wins to NewPort, which assumes (1) a significant advantage of NewPort's product, (2) AMCC's inability to match this advantage on its own, and (3) a significant expansion of NewPort's product portfolio to match the breadth offered by AMCC. None of these are proven (or for that matter, disproven) at present, so it would appear that pending further developments, the high-speed chip market is still AMCC's to lose.

Overall, the NewPort acquisition doesn't change my fundamental outlook, which is to hold a basket of the strongest communication chip companies - AMCC, BRCM and PMCS - and to sell any one of them that looks as if it is starting to slip (as VTSS arguably is). However, right now AMCC, BRCM and PMCS are all growing at 100+% rates. So to me there is no harm in holding all three - albeit lightly - which is what I am continuing to do.


Read More Posts by This Author
Go To This Post
More Recommended Posts
Get past Posts of the Day in the Archives