networkZONE Products for the week of April 8, 2002
Ed. Note - Alas, Transpectrum is no more as a company! Read the ZONE editorial
Requiem for Magnificent Failure.
Transpectrum made a grand entrance into the analog semiconductor
market with the launch of the world's first single-chip, 40 Gigabit per
second (Gbps) CMOS transceiver. Transpectrum, founded one year ago by prominent
engineers in the semiconductor industry, is a provider of leading-edge,
integrated CMOS semiconductor solutions that enable broadband communications
and networking of data, video and voice services. The company launched today
and unveiled the XPT140, an ultra-high-speed integrated circuit (IC), manufactured
entirely in CMOS process technology.
"Transpectrum is pioneering a new class of communications semiconductor solutions," said Dr. Fred Mohamadi, CEO, Transpectrum. "Others have talked about bringing this level of technology to market, but through our unique design techniques we have been able to bring the benefits of lowest power, lowest cost, and smallest footprint chips years ahead of the competition."
"CMOS technology has demonstrated that it is capable of performing the OC-192 PHY function. This provides important power and integration advantages for optical networking equipment," said Allan Armstrong, director of optical transport semiconductors at RHK, Inc. "Transpectrum plans to further lower the power of OC-192 transceivers, deploy quad transceivers in DWDM applications, and pioneer the use of CMOS for OC-768."
PASIFX Product Family
Transpectrum's PASIFX portfolio brings unprecedented benefits to enterprise
LANs, SANs, WANs and MANs by reducing the amount of physical space and power
required for deploying more bandwidth; decreasing operating equipment costs
while maximizing bandwidth capacity; and simplifying packaging and cooling,
reducing overall system complexity and size.
Transpectrum's multi-channel 10Gbps PASIFX product portfolio includes the following:
Transpectrum XPT410, the world's first single-chip, 40 Gbps CMOS transceiver, is a fully integrated quad 10Gbps transceiver, that features the following:
"Our multi-channel PASIFX product portfolio is the only family of
transceiver solutions manufactured entirely in CMOS process technology,"
said Dr. Behzad Razavi, founder of Transpectrum. "The PASIFX product
portfolio, which we will roll out during the next six months, positions
us as a dominant player in the communications semiconductor industry."
networkZone's Lee Goldberg Says . . .
If the folks at Transpectrum can deliver on their
claims made for their 10-Gbit/s and 40-Gbit/s all-CMOS transceivers, they
do stand to become one of the key players in making cost-effective OC-192
equipment possible, and bringing OC-768 closer to mass deployment. Truth
be told, their claims are a bit tough to swallow, and if it were not for
some fairly convincing discussions I had with them about how do what they
claim to do with CMOS, I'd have a hard time reviewing these two chips with
a straight face.
To get a better sense of what I talked with Transpectrum CEO, Dr. Fred Mohamadi, who I know from his days at Newport, before it was acquired by Broadcom. He seems to be building on the expertise he acquired at Newport, developing all-CMOS OC-48 parts two years ago. One indicator of this is that the majority of his designers are analog and RF engineers. While he was vague on enough details to add a half-saltshaker to my rating, he did talk frankly about some of the challenges his team faced in building OC-192 and OC-768 components using CMOS transistors that have an ft barely twice the operating range of the part (most digital designs require an ft about 4X the maximum clock rate of the chip.)
In addition to the difficulties of the actual 40 Gbit/s interface, there is also a bunch of work keeping the data synchronized once it is converted to a parallel path within the bowels of the chip. The multi-layer stripline design metal runs used for the wide data paths within the chip help keep the bits clocking through in lock step. Both the 10-Gbit and 40-Gbit parts employ several cross-linked mechanisms which monitor the relative position of each bit and adjust timing to maintain lane coherency on a per-bit basis. (See the figures for the XPT140 by viewing the transmitter & receiver block diagrams.)
To ride herd on the fast clocks required for a chip like this, Transpectrum claims to have developed a highly accurate digital PLL driven by a VCO of proprietary design. The novel CDR/PLL and synchronization scheme probably also accounts in good part for the tremendous power savings (50%-75%) they claim to have over competing 10 Gbit/s solutions from Broadcom, Silicon Labs, and Vitesse. This, and a tight layout ensured by their suite of custom-written design tools allows for a very aggressive, cutting-edge design.
Although the block diagrams for both the XPT410 four-channel 10-Gbit/s part, and the XPT140 single-channel 40 Gbit/s chip are nearly identical (see the block diagrams again), and depict a 16-channel internal data path, I suspect that this is not quite true. Although I have no hard data to back me up, my guess is that the 40-Gbit part actually uses four, 16-bit data paths to move the data around the chip. The elaborate, and most likely effective data path synchronization architecture used in both chips would make such an arrangement practical. Besides, only having to worry about super-fast signals in a few areas of the design probably makes it easier to work within the limits of a bulk CMOS process.
While I feel that Transpectrum has the massive talent required to deliver working product by the end of 3Q '02 as they claim, it is likely that at least their first generation of 40-Gbit/s parts may not perform quite as well as an equivalent part fabricated in SiGe, InP, or other processes. I am not enough of an expert to competently speculate how the problems will manifest, but likely places could be jitter generation and tolerance, or sensitivity to minor design issues at the SFI-5 interface.
This being said, what they learn here should be invaluable to making their Quad 10-Gbit/s parts work at least acceptably well. Since the market for OC-192 equipment is much more viable in the short term, it looks like Transpectrum may be able to have both parts working well enough to start making an impact on the market and help drive down the cost of both OC-192 and OC-768 equipment.
|
|