i/oZONE Products for the week of June 3, 2002
Transpectrum says . . .
Pushing The Envelope - Transpectrum's All-CMOS All-In-One
10-Gbit Chip Includes TIA, Retimer, Laser Drivers, and More
Ed. Note - Alas, Transpectrum is no more as a company! Read the ZONE editorial Requiem for Magnificent Failure.
Transpectrum has launched the XPT110, the world's first CMOS 10 Gigabit per second (Gbps) all-in-one chip for enabling broadband communications and networking. By reducing the amount of physical space, power and cost for deploying more bandwidth, the XPT110 brings unprecedented benefits to customers in networking, storage management and mobile communications.
The XPT110 is the latest addition to Transpectrum's PASIFX product line, the only family of transceivers manufactured entirely using CMOS process technology. The XPT110 integrates the functionality of several chips - a transimpedence amplifier, limiting amplifier, bi-directional re-timers, signal boosters, and programmable driver - into a very small footprint single chip.
As an example, this functional consolidation enables significant size and cost reductions of the optical modules that link switches and fiber cables. The XPT110 can be used in SAN switches to improve data transfer performance, as well as to bring down the cost of high-speed network links across the board.
"Transpectrum is committed to shaping the future of the industry by supplying the highest bandwidth, lowest cost, and smallest footprint chips years ahead of the competition," said Dr. Fred Mohamadi, CEO, Transpectrum."The XPT110 truly signals what lies ahead - this is an entirely new class of semiconductor solutions."
"Our fully integrated chip is another breakthrough in substantially reducing pin count and power, hence, reducing copper traces in high-speed systems," said Dr. Behzad Razavi, founder and acting CTO of Transpectrum. "The XPT110 will roll out during the fourth quarter, positioning us as a dominant player in the SAN market."
Transpectrum XPT110 features the following:
analogZONE Says . . .
Transpectrum's latest part falls somewhere between a network and an I/O component, but because it has such a large potential to cut the cost of FibreChannel and other SAN applications, I've posted it in the i/oZONE area. The claims for this highly integrated 10-Gbit optical driver chip also border between the miraculous and impossible. If it was not for the potential impact of this component, and some frank discussions I had when I reviewed their OC-192 and OC-768 transceivers back in April , I'd have trouble reporting on this chip with a straight face.
Transpectrum has set itself a formidable task, to reduce the current target cost for a 10-Gig short-haul optical module by half or more by the end of this year. With the cost of a module hovering at a tad below $1K now, people talk of $400 modules in '03, which translates to more like $5-$600 in the real world. Such a cost reduction would help fuel a much bigger market for 10-Gig links in everything from switch stacking busses to FibreChannel-based SANs, but to hit anywhere close to the $400 mark, there are going to have to be some significant improvements is all areas of module manufacturing.
Besides the electronics costs that Transpectrum addresses, module designers will have to find ways to reduce costs in the packaging, assembly, optics, and electro optics that go into these products, something the SFP 20-pin package being proposed to supercede the Xenpak module should help with. It's much smaller (38% of Xenpak) and designed to let manufacturers take advantage of high-volume packaging and assembly technologies. The XPT110 is designed to fit both the form factor, and the lower power (1 W for the combined chip vs. 7-9 W for Xenpak solutions) allowed by the new SFP standard.
Like its companion transceivers, the XPT110 is an all-CMOS design which packs a previously unheard-of amount of functionality onto a single chip. The IC includes a programmable laser driver and laser boosters (transmit pre-emphasis or possibly adaptive EQ on the receive side), plus all the CDR and re-timing circuitry.
According to Fred Mohamdi, Transpectrum's chief exec, this aggressive level of integration has its challenges, but also offers many advantages. In return for the sweat involved with managing noise from trace and substrate coupling, you get lowered rise-times because you do away with intra-chip parasitics. This alone sharpens rise time to 40 ps, allowing the chip's internal retimers to clean up both the transmit and receive eyes. This retiming is essential to clean up incoming signals coming from a transceiver across 8"-10" of PC board. Since this is a single channel device, there are no VCO pull issues that worried me in their earlier multi-channel transceivers.
If Mohamdi and crew can really produce this part, they will significantly reduce the space, power and cost of 10-Gig connections. Since the part can handle a wide range of speeds, and its gain, pre-emphasis, and drive parameters can be programmed through an I2C interface, the same chip can drive OC-192 or 10-Gig SONET connections.
Perhaps most intriguing is Transpectrum's targeting of this product for SANs. They say that once the FibreChannel standard is upgraded to 10X (using technology borrowed from 10Gig Ethernet), this chip will support it. Mohamdi expects that high-capacity SAN applications will lead LAN markets by a year or two, with a sales surge in 10X FibreChannel-based products predicted by late '03.
One of the reasons I'm reviewing this part with a clean conscience is that the company is doing due-diligence in its design by running elements separately on multiple chips. Surprisingly (at least to me) the total power of test chips around 0.8 W. Fully integrated samples are expected Q4 '02, Transpectrum expects to deliver the chip some time in Q4 of this year. Pricing will be under $200 for samples, under $100 in production quantities.
|