Comms chips

Mindspeed Claims SerDes Superiority

Competition -- and bickering -- among manufacturers of SerDes (serializer/deserializer) transceiver chips for 10-gigabit Ethernet equipment is hotting up following a further product announcement in this field from Mindspeed Technologies on Monday (see Mindspeed Unveils Transceiver).

Like all SerDes transceivers, MindSpeed’s new chip, called the SkyRail M27205, is designed for use in high-speed connections between bays of optical networking equipment or in backplanes that link together cards in a single chassis. The chip prepares signals by converting them from a parallel, low-speed data interface to a high-speed, serial interface, and vice versa.

In the case of this particular chip, the conversion is between an XGMII interface on the low-speed side and a four-lane XAUI interface on the high-speed side. The SkyRail M27205 implements the latest version of the evolving 10-gig Ethernet standard, according to Eli Massabki, director of product marketing.

It's the latest addition to the industry's widest portfolio of transceivers in this field, he claims. Mindspeed now has single, quad, and octal chips (see Conexant SkyRail Transceiver, Mindspeed Intros Transceiver, and Mindspeed Unveils Octal SkyRail).

Offering such a wide choice of chips is a big deal, according to Massabki, who says a customer might need an eight-channel device on a switch card and a single or four-channel device on a line card. "Customers will want to get all these [chips] from a single vendor," he contends.

Another important feature of all of Mindspeed’s SerDes chips is that they support variable speeds, thanks to their "Flexiclock architecture." This makes it possible for Mindspeed to claim that its chips will support a range of protocols, based on the fact that it can tinker with the bit rate if it becomes necessary.

As a result, the company says the M27205 can be used in 10-gig Fibre Channel equipment, an emerging standard that's very closely based on 10-gigabit Ethernet. "At this point it's identical; the only change is the bit rate," says Massabki, although he acknowledges that the standard is immature and could change significantly in the future.

As well as supporting different protocols, this feature also allows Mindspeed's part to be backwards compatible with old backplanes and backplanes that operate at non-standard data rates.

Most other SerDes chips operate at fixed line rates. The equivalent transceiver from Broadcom Corp. (Nasdaq: BRCM) offers a 3.125 Gbit/s data rate only. The one from Marvell Technology Group Ltd. (Nasdaq: MRVL) operates at 1.0, 1.25, 2.0, 2.5, or 3.125 Gbit/s.

Another point in Mindspeed's favor is that its four-channel SerDes chip takes advantage of clock-forwarding, a technique that allows the full bandwidth of the signal to carry data. It does this by eliminating the need for data encoding, thus reclaiming a 20 percent overhead. "This is equivalent to 4.25 Gbit/s with 8b/10b encoding," Massabki notes.

But while Broadcom and Marvell's chips might not be as flexible or cover as many applications, both companies insist that their chips perform better in the functions they can do.

Broadcom, for instance, claims to have the only SerDes chips that employ receive-end equalization -- a technique that automatically corrects for tranmission distortion after the signal has travelled across the backplane. Other SerDes vendors, including Mindspeed, use pre-emphasis, which involves manually tuning each channel at the transmit end. (Gennum Corp. is planning to offer chips that use receive-end equalization, but its chips will not have full SerDes functionality -- see Gennum Gets Dense.)

Marvell, for its part, throws cold water over Mindspeed's claim to have industry-leading jitter performance. In Monday's press release Mindspeed claims the M27205 is the industry's "most robust" SerDes transceiver, with a jitter performance exceeding that required by the 10-gig Ethernet standard.

Marvell begs to differ. Both companies exhibited their wares in their booths at the Communications Design Conference tradeshow in early October. "The eye diagram in the Mindspeed demo indicated poor jitter performance compared to the Marvell demo," Gary Smerdon, VP of marketing for Marvell's communications business group, writes in an email to Light Reading. "Also Marvell's demo showed performance over 40 inches of backplane with six connectors, whereas Mindspeed's was measured directly off the device." In other words, in Smerdon's view, Marvell's device was demonstrated under real-world conditions, whereas Mindspeed's was not.

Mindspeed notes that the chip in its demo was not the M27205, but the M27211, an octal SerDes transceiver that was unveiled at the beginning of October. However, given that the M27205 is based on the technology developed for the M27211, according to Massabki, this doesn't appear to let Mindspeed off the hook completely.

Mindspeed’s Massabki also manages to raise the hackles of Broadcom and Marvell by claiming that the SkyRail M27205 is the only SerDes chip that's "strictly compliant" with the latest version of the 10-gig Ethernet standard. Although the standard has yet to be finalized, the important parts of it haven’t changed in the last three revisions, he says. "Mindspeed has been working with version 3.3... 3.x is fairly stable. Anything below that is significantly different and is not interoperable."

Marvell unveiled its 10-gig Ethernet product, the Alaska X, in September 2001 and says it's compliant with version 3.1 –- a claim that Massabki disputes, saying it’s compliant with an older, 2.x version of the evolving standard.

Broadcom was first to market with a 10-gig Ethernet transceiver chip, its BCM8011, in May 2001, at which time it was based on version 1.2 of the standard (see Broadcom Offers 10-Gig Ether Bits). "Since then, we have updated/revved the design so that we are consistent with draft version 2.3, and we are unaware of any interoperability issues between draft 2.3 and 3.2," emails Dave Sorenson, Broadcom's product line manager for SerDes.

In any case, Broadcom will most likely upgrade its chip again, after version 3.3 is finalized, which is supposed to take place at next week's meeting of the IEEE 802.3ae working group, Sorenson adds.

Other vendors, such as PMC-Sierra Inc. (Nasdaq: PMCS), Texas Instruments Inc. (NYSE: TXN), and Velio Communications Inc., also make SerDes transceivers, but their devices are compatible with, as opposed to "strictly compliant" with 10 gigabit Ethernet (see Velio Cleans Up).

— Pauline Rigby, Senior Editor, Light Reading

Want to know more? This very topic is the subject of a session at Lightspeed Europe,Light Reading’s annual conference, on December 4-6, 2001, in London. Details: Optical Ethernet

Page 1 / 2   >   >>
metrodude 12/4/2012 | 7:37:10 PM
re: Mindspeed Claims SerDes Superiority Tried and True. Although there are many
choices in SERDES components. One must also
consider the service, support and competence of FAE's.
As an Engineer, dealing with FAE's for issues and I demand prompt curtious and stellar service/assistance if needed with instrumental execution and delivery.

They have a customer service/sales organization that is first rate, and actually a pleasure to deal with.

Hat's off, a job well done.
sdarch 12/4/2012 | 7:37:05 PM
re: Mindspeed Claims SerDes Superiority There're more players of CMOS 10GbE transceiver (4x3.125G link). Take a not-so-wild guess: PMC-Serria has their sample chips plugged in Cisco's prototype boxes, 'cos you know their close relationship, don't you? TI successfully sent their samples to various customers long times ago, yet it was lacking for some critical features required for ethernet. Don't know whether their latest one works. Startup Velio made quite some noise. they do 8-channel chip from the very beginning. If any of their customers complete a box with Velio chip, maybe you'll hear it firstly from Velio's marketing guys (of 'cos their chip might actually work). another startup, much quiet, much smaller BitBlitz might indeed have very good design as they claimed, such as low power consumption, can they survive the economic slowdown?
Silicon lab announced single channel 2.5G for sonet which was picked by Ciena. technically they should be capable of 4x3.125G chip, too, given enough R&D budget. AMCC offers 2.5G for sonet only. I wonder why I can't find any 3G stuff on their website?

Did I miss any big name network chip company?

hey, do you still remember: PMC-Sierra snapped AAnetcom for $300M, Connexent bought HotRail for $400M and put it into mindspeed later, and Broadcom, wow, hand out $1.3B for NewPort. How's their 10G CMOS serdes doing?!
toll booth willy 12/4/2012 | 7:37:03 PM
re: Mindspeed Claims SerDes Superiority The key to faking out other vendors is the clammy hands. It's a good non-specific symptom. I'm a big believer in it. A lot of people will tell you a good phoney fever is a deadlock, but you get a nervous vendor, you could wind up in a doctor's office--that's worse than 10G. You fake a stomach cramp, and when you're bent over, moaning and wailing, you lick your palms. It's a little childish and stupid, but then so are the public markets.
sailor 12/4/2012 | 7:36:51 PM
re: Mindspeed Claims SerDes Superiority You can kiss PMC goodbye. They canned almost all of the former Aanetcom engineers. They should totally exit this market very soon.
toll booth willy 12/4/2012 | 7:36:43 PM
re: Mindspeed Claims SerDes Superiority I love the smell of SerDes Transceivers in the morning.
ajo2 12/4/2012 | 7:36:38 PM
re: Mindspeed Claims SerDes Superiority I just wish Broadcom would stop acting (pretending) like they were the only game in town. It is as if they dismiss the idea that anyone can have a better solution than theirs. Then when you tell them it not only fits the application better, but is cheaper, they are at a loss for words.
toll booth willy 12/4/2012 | 7:36:37 PM
re: Mindspeed Claims SerDes Superiority And even if we beat Broadcom, if we win, HAH! Even if we play so far above our heads that our noses bleed for a week to ten days; even if God in Heaven above points his hand at our products; even if every man woman and child joined hands together and prayed for us to beat Broadcom, it just wouldn't matter because all the really good looking girls would still go out with the guys from Broadcom because they've got all the money! It just doesn't matter if we win or if we lose. IT JUST DOESN'T MATTER
mofstedahl 12/4/2012 | 7:36:36 PM
re: Mindspeed Claims SerDes Superiority The article and the Companies cited have missed two very important points. The first one revolves around signal integrity. There is so much more to this than preemphasis or receiver EQ. Running multi-Gigabit channels across existing FR4 backplanes is not trival and requires an overall system knowledge and on-chip techniques in order to be successful. The other key area is power consumption. This is a critical area to OEM's who have severe constraints with re to system level power consumption at the central office which also directly affects form factor. Velio Communications is the only Company out there that has the system and signal integrity experience which is delivering, in real system environments, high speed I/O with both a high level of signal integrity as well as the World's lowest power and smallest form factor SerDes solutions.
Passing Shot 12/4/2012 | 7:35:08 PM
re: Mindspeed Claims SerDes Superiority I've been very impressed with the SerDes products these guys at Mindspeed have introduced. I guess all these products come from the HotRail acquisition in San Jose. I saw their octal at the Network Processor Forum and it was driving 47" of FR-4 @ 3.2Gbps and the device wasn't even lukewarm. You should check them out!
smartjohn 12/4/2012 | 7:34:33 PM
re: Mindspeed Claims SerDes Superiority So everyone wishes they were like mindspeed!

broadcon and marval claim that they are as 10GE compliant as mindspeed. (Odd since they admit only complying with previous versions.)
veleo says its signal integrity is as good as mindspeed.
pmc and ti claimed to have a serdes as good as mindspeed, but for some reason arenGăÍt able to show it to anyone. (Is troubled TI going to follow PMCs serdes cancelling lead?)

That velio egghead says blah-blah about signal integrity. The problem is that I WAS THERE at CDC and Mindspeed was demonstrating 3.2Gbps over 47Găą with all channels running and a perfect data eye. velio was nowhere to be seen. marvel was embarrassing to even look at, showing only a single channel hooked directly up to the scope under a Găúfirst 10GEGăą sign. It made you sick.

It looks an awful lot like Mindspeed has the highest speed, lowest power, highest integration serdes Gă˘PERIOD Gă˘and the confidence to show it in broad daylight.

We WISH you could all be more like mindspeed! Can the rest of you please spend more time on products and less on blah-blah so we can get 10GE off and rolling???
Page 1 / 2   >   >>
Sign In