INTEGRATED PHOTONICS: Integrated F-P laser array solves 100G problems

A primary challenge in the 100 Gbit/s or 100G optical interconnect space is the design and manufacture of high-speed, low-cost lasers that can support the several-kilometer distances of a large data center.

An array of twelve lasers on a single bar are flip-chip bonded onto a silicon photonics chip and lined up against waveguides
An array of twelve lasers on a single bar are flip-chip bonded onto a silicon photonics chip and lined up against waveguides

A primary challenge in the 100 Gbit/s or 100G optical interconnect space is the design and manufacture of high-speed, low-cost lasers that can support the several-kilometer distances of a large data center. Although widely used at 1 and 10 Gbit/s, vertical-cavity surface-emitting lasers (VCSELs) are not an option for 100G systems since their distance is limited to 100 m on OM3 fiber (multimode, 50 μm core/125 μm cladding)—much too short to span a large data center.

“The IEEE 802.3 specification for 10 km reach, 100GBase LR (100 Gbit/s, baseband modulation, long reach) would span almost any data center and more; but this specification requires four directly modulated distributed feedback (DFB) or electro-absorption (EA) lasers with wave-division multiplexing (WDM) to combine all four channels onto a single fiber,” says Arlon Martin, VP of marketing at Kotura (Monterey Park, CA). “The problems with this approach are cost, power, and size. Due to complexity, these transceivers cost one hundred times 10G transceivers of similar reach.”

Martin says that customers expecting next-generation transceivers to transmit more bits at a lower cost are disappointed, and cost is not the only concern: Power consumption is also way too high—20–24 W compared to 1–2 W for 10G. Finally, the 100G C form-factor pluggable (CFP) package is so large (larger than an iPhone) that only four of them fit across the face plate of a 19 in. switch, making density worse than using 10G transceivers.

Martin says 100G transceivers are expensive because of the stringent requirement for the lasers, which are specified to an 800 GHz grid (corresponding to roughly 4.5 nm spacings) in the 1310 nm region, meaning the laser wavelengths must center exactly at 1295.56, 1300.05, 1304.58, and 1309.14 nm. In addition, to allow for multiplexing and system interoperability, only 2.1 nm of the 4.5 nm window can be used by the laser. Not only does the laser manufacturer have to develop high-performance 25 GHz lasers to match this unusual grid, they also have to throw away the distribution of perfectly good lasers that don’t match the grid.

A better way
Fabry-Perot (F-P) lasers—one of the easiest to manufacture—have a broad lasing spectrum without the complicated, DFB-style grating and not requiring the EA stage of a multisection EA laser. These lasers are so small that 10,000 or more fit on a single indium phosphide (InP) wafer at high yield. However, F-P lasers usually cannot be used for high-speed data-center applications because their modulation speed is far too slow and their lasing modes cannot meet the requirements of a WDM system.

But Kotura has developed a new approach that converts a low-cost, F-P style laser to a high-speed WDM laser for data-center applications. A laser array is flip-chip bonded onto a silicon photonics chip using an automated pick-and-place machine and passive alignment (the lasers are not powered on). Using physical features and alignment marks, the entire array is soldered into place, precisely aligning each laser with its corresponding waveguide on the silicon photonics chip (see figure).

An array of twelve lasers on a single bar are flip-chip bonded onto a silicon photonics chip and lined up against waveguidesAn array of twelve lasers on a single bar are flip-chip bonded onto a silicon photonics chip and lined up against waveguides
An array of 12 lasers on a single bar are flip-chip-bonded onto a silicon photonics chip and lined up against waveguides.

Conversion of the broadband laser spectrum to the precise WDM wavelength requires a grating on each of the waveguides of the silicon chip. The gratings are imprinted simultaneously using a photolithographic mask so that each laser has its own exact grating defined by the silicon process, not by the InP process. This means that every laser can be used to generate the required wavelength and almost any wavelength plan can be accommodated. If more channels are required, then a larger array is used and additional waveguides and gratings are added to the silicon chip, allowing expansion from 4 WDM channels to 12 or even 40 channels on a single chip.

Because Kotura uses a 3-μm-diameter waveguide that roughly matches the beam profile of the laser and special waveguide facets and coatings on the array, the need for collimators, lenses, and double-stage isolators is eliminated, lowering costs and simplifying assembly. And to accomplish the increased modulation speed of 25 GHz or more, the laser is operated in CW mode and modulators are integrated into the silicon chip—fast enough for 100G with four encoded data streams at 25 Gbit/s each.

The Kotura WDM 100G engine fits inside a compact quad small-form-factor pluggable (QSFP) package, increasing front-panel density by more than a factor of 10 to more than 4.4 Tbit/s, all while consuming far less power than many CFP solutions at data-center reach lengths of 2 km and more.

More in Lasers & Sources