Wednesday, May 20, 2015 at 8:36AM
The second part of the survey of developments at the OFC 2015 show held recently in Los Angeles.
Part 2: Client-side component and module developments
- CFP4- and QSFP28-based 100GBASE-LR4 announced
- First mid-reach optics in the QSFP28
- SFP extended to 28 Gigabit
- 400 Gig precursors using DMT and PAM-4 modulations
- VCSEL roadmap promises higher speeds and greater reach
First CFP4 100GBASE-LR4s
Several companies including Avago Technologies, JDSU, NeoPhotonics and Oclaro announced the first 100GBASE-LR4 products in the smaller CFP4 optical module form factor. Until now the 100GBASE-LR4 has been available in a CFP2 form factor.
“Going from a CFP2 to a CFP4 results in a little over a 2x increase in density,” says Brandon Collings, CTO for communications and commercial optical products at JDSU. The CFP4 also has a lower maximum power specification of 6W compared to the CFP2’s 12W.
The 100GBASE-LR4 standard spans 10 km over single mode fibre. The -LR4 is used mainly as a telecom interface to connect to WDM or packet-optical transport platforms, even when used in the data centre. Data centre switches already favour the smaller QSFP28 rather than the CFP4.
Other 100 Gigabit standards include the 100GBASE-SR4 with a 100 meters reach over OM3 multi-mode fibre, and up to 150m over OM4 fibre. Avago points out that the -SR4 is typically used between a data centre’s top-of-rack and core switches whereas the -LR4 is used within the core network and for links between buildings. The -LR4 modules also can support Optical Transport Network (OTN).
But in the data centre there is a mid-reach requirement. “People are looking at new standards to accommodate more of the mega data centre distances of 500 m or 2 km,” says Robert Blum, Oclaro’s director of strategic marketing. These mid-reach standards over single mode fibre include the 500 meter PSM4 and the 2 km CWDM4 and modules supporting these standards are starting to appear. “But today, on single mode, there is basically the -LR4 that gets you to 10 km,” says Blum.
JDSU also views the -LR4 as an interim technology in the data centre that will fade once more optimised PSM4 and CWDM4 optics appear.
QSFP28 portfolio grows
The 100GBASE-LR4 was also shown in the smaller QSFP28 form factor, as part of a range of new interface offerings in the form factor. The QSFP28 offers a near 2x increase in face plate density compared to the CFP4.
JDSU announced three 100 Gigabit QSFP28-based interfaces at OFC - the PSM-4 and CWDM4 MSAs and a 100GBASE-LR4, while Finisar announced QSFP28 versions of the CWDM4, the 100GBASE-LR4 and the 100GBASE-SR4. Meanwhile, Avago has samples of a QSFP28 100GBASE-SR4. JDSU’s QSFP28 -LR4 uses the same optics it is using in its CFP4 -LR4 product.
The PSM4 MSA uses a single mode ribbon cable - four lanes in each direction - to deliver the 500 m reach, while the CWDM4 MSA uses a fibre to carry the four wavelengths in each direction. The -LR4 standard uses tightly spaced wavelengths such that the lasers need to be cooled and temperature controlled. The CWDM4, in contrast, uses a wider wavelength spacing and can use uncooled lasers, saving on power.
"100 Gig-per-laser, that is very economically advantageous" - Brian Welch, Luxtera
Luxtera announced the immediate availability of its PSM4 QSFP28 transceiver while the company is also offering its PSM4 silicon chipset for packaging partners that want to make their own modules or interfaces. Luxtera is a member of the newly formed Consortium for On-Board Optics (COBO).
Luxtera’s original active optical cable products were effectively 40 Gigabit PSM4 products although no such MSA was defined. The company’s original design also operated at 1490nm whereas the PSM4 is at 1310nm.
“The PSM4 is a relatively new type of product, focused on hyper-scale data centres - Microsoft, Amazon, Google and the like - with reaches regularly to 500 m and beyond,” says Brian Welch, director of product marketing at Luxtera. The company’s PSM4 offers an extended reach to 2 km, far beyond the PSM4 MSA’s specification. The company says there is also industry interest for PSM4 links over shorter reaches, up to 30 m.
Luxtera’s PSM4 design uses one laser for all four lanes. “In a 100 Gig part, we get 100 Gig-per-laser,” says Welch. “WDM gets 25 Gig-per-laser, multi-mode gets 25 Gig-per-laser; 100 Gig-per-laser, that is very economically advantageous.”
QSFP28 ‘breakout’ mode
Avago, Finisar and Oclaro all demonstrated a 100 Gigabit QSFP28 modules in ‘breakout’ mode whereby the module’s output fibres fan out and interface to separate, lower-speed SFP28 optical modules.
“The SFP+ is the most ubiquitous and standard form factor deployed in the industry,” says Rafik Ward, vice president of marketing at Finisar. “The SFP28 leverages this architecture, bringing it up to 28 Gigabit.”
Applications using the breakout arrangement include the emerging Fibre Channel standards: the QSFP28 can support the 128 Gig Fibre channel standard where 32 Gig Fibre Channel traffic is sent to individual transceivers. Avago demonstrated such an arrangement at OFC and said its QSFP28 product will be available before the year end.
Similarly, the QSFP28-to-SFP28 breakout mode will enable the splitting of 100 Gigabit Ethernet (GbE) into IEEE 25 Gigabit Ethernet lanes once the standard is completed.
Oclaro showed a 100 Gig QSFP28 using a 4x28G LISEL (lens-integrated surface-emitting DFB laser) array with one channel connected to an SFP28 over a 2 km link. Oclaro inherited the LISEL technology when it merged with Opnext in 2012.
Finisar demonstrated its 100GBASE-SR4 QSFP28 connected to four SFP28s over 100 m of OM4 multimode fibre.
Oclaro also showed a SFP28 for long reach that spans 10 km over single-mode fibre. In addition to Fibre Channel and Ethernet, Oclaro also highlights wireless fronthaul to carry CPRI traffic, although such data rates are not expected for several years yet. Oclaro’s SFP28 will be in full production in the first quarter of 2016. Oclaro says it will also use the LISEL technology for its PSM4 design.
Industry prepares for 400GbE with DMT and PAM-4
JDSU demonstrated a 4 x 100 Gig design, described as a precursor for 400 Gigabit technology. The IEEE is still working to define the different versions of the 400 Gigabit Ethernet standard. The JDSU optical hardware design multiplexes four 100 Gig wavelengths onto a fibre.
“There are multiple approaches towards 400 Gig client interfaces being discussed at the IEEE and within the industry,” says JDSU’s Collings. “The modulation formats being evaluated are non-return-to-zero (NRZ), PAM-4 and discrete multi-tone (DMT).”
For the demonstration, JDSU used DMT modulation to encode 100 Gbps on each of the four wavelengths, although Collings stresses that JDSU continues work on all three formats. In contrast, MultiPhy is using PAM-4 to develop a 100 Gig serial link.
At OFC, Avago demonstrated a 25 Gig VCSEL being driven using its PAM-4 chip to achieve a 50 Gig rate. The PAM-4 chip takes two 25 Gbps input streams and encodes each two bits into a symbol that then drives the VCSEL. The demonstration paves the way for emerging standards such as 50 Gigabit Ethernet (GbE) using a 25G VCSEL, and shows how 50 Gigabit lanes could be used to implement 400 GbE using eight lanes instead of 16.
NeoPhotonics demonstrated a 56 Gbps externally modulated laser (EML) along with pHEMT gallium arsenide driver technology, the result of its acquisition of Lapis Semiconductor in 2013.
The main application will be 400 Gigabit Ethernet but there is already industry interest in proprietary solutions, says Nicolas Herriau, director of product engineering at NeoPhotonics. The industry may not have decided whether it will use NRZ or PAM-4 [for 400GbE], “but the goal is to get prepared”, he says.
Herriau points out that the first PAM-4 ICs are not yet optimised to work with lasers. As a result, having a fast, high-quality 56 Gbps laser is an advantage.
Avago has shipped over one million 25 Gig channels in multiple products
The future of VCSELs
VCSELs at 25 Gig is an enabling technology for the data centre, says Avago. Operating at 850nm, the VCSELs deliver the 100m reach over OM3 and 150m reach over OM4 multi-mode fibre. Avago announced at OFC that it had shipped over one million VCSELs in the last two years. Before then, only 10 Gig VCSELs were available, used for 40 Gig and 100 Gig short-reach modules.
Avago says that the move to 100 Gig and beyond has triggered an industry debate as to whether single-mode rather than multi-mode fibre is the way forward in data centres. For VCSELs, the open questions are whether the technology can support 25 Gig lanes, whether such VCSELs are cost-effective, and whether they can meet extended link distances beyond 100 m and 150 m.
“Silicon photonics is spoken of as a great technology for the future, for 100 Gig and greater speeds, but this [announcement] is not academic or hype,” says I-Hsing Tan, Avago’s segment marketing manager for Ethernet and storage optical transceivers. “Avago has been using 25 Gig VCSELs for short-reach distance applications and has shipped over one million 25 Gig channels in multiple products.”
The products that account for the over one million shipments include Ethernet transceivers; single- and 4-lane 32 Gigabit Fibre Channel, each channel operates at 28 Gbps; Infiniband applications, with 4-channels being the most popular; and proprietary optical interfaces with the channel count varying from two to 12 channels, 50 to 250 Gbps.
In other OFC data centre demonstrations, Avago showed an extended short reach interface at 100 Gig - the 100GBASE-eSR4 - with a 300 m span. Because it is a demonstration and not a product, Avago is not detailing how it is extending the reach beyond saying that it is a combination of the laser output power and the receiver design. The extended reach product will be available from 2016.
Avago completed the acquisition of PLX Technologies in the third quarter of 2014 and its PCI Express (PCIe) over optics demonstration is one result. The demonstration is designed to remove the need for a network interface card between an Ethernet switch and a server. “The aim is to absorb the NIC as part of the ASIC design to achieve a cost effective solution,” says Tan. Avago says it is engaged with several data centre operators with this concept.
Avago also demonstrated 40 Gig bi-directional module, an alternative to the 40GBASE-SR4. The 40G -SR4 uses eight multi-mode fibres, four in each direction, each carrying a 10 Gig signal. “Going to 40 Gig [from 10 Gig] consumes fibre,” says Tan. Accordingly, the 40 Gig bidi design uses WDM to avoid using a ribbon fibre. Instead, the bidi uses two multi-mode fibres, each carrying two 20 Gig wavelengths travelling in opposite directions. Avago hopes to make this product generally available later this year.
At OFC, Finisar demonstrated designs for 40 Gig and 100 Gig speeds using duplex multi-mode fibre rather than ribbon fibre. The 40 Gig demo achieved 300 m over OM3 fibre while the 100 Gig demo achieved 70 m over OM3 and 100 m over OM4 fibre. Finisar’s designs use four wavelengths for each multi-mode fibre, what it calls shortwave WDM.
Finisar’s VCSEL demonstrations at OFC were to highlight that the technology can continue to play an important role in the data centre. Citing a study by market research firm, Gartner, 94 percent of data centres built in 2014 were smaller than 250,000 square feet, and this percentage is not expected to change through to 2018. A 300-meter optical link is sufficient for the longest reaches in such sized data centres.
Finisar is also part of a work initiative to define and standardise new wideband multi-mode fibre that will enable WDM transmission over links even beyond 300 m to address larger data centres.
“There are a lot of legs to VCSEL-based multi-mode technology for several generations into the future,” says Ward. “We will come out with new innovative products capable of links up to 300 m on multi-mode fibre.””
For Part 1, click here