Sckipio, Broadcom in G.fast Race
Sckipio Partnership With Lantiq Brings G.fast to GPON, VDSL2 Suite
Sckipio is sampling a chipset for distribution-point implementations of G.fast (G.9700/G.9701), the new International Telecommunication Union (ITU) standard for 1Gbps over 250 meters of copper. The company will tape out and sample a single-channel device for customer premises equipment (CPE) in 1Q15. Both the CP1000 for CPE and the DP3000 for multiport distribution-point units (DPUs) consist of a DSP processor and an analog front end (AFE), which integrates a line driver.
Broadcom, long considered Sckipio’s toughest competitor, demonstrated a DSP/AFE combination chipset for G.fast at the recent Broadband World Forum (BWF). The DPU-oriented BCM65200 and BCM65900 are sampling now, building on Broadcom’s central-office DSL designs. The company’s original ADSL2+/VDSL2 residential-gateway SoC (BCM63138) is in full production, but a new version is sampling to select customers with small silicon changes and upgraded firmware that add G.fast support. Broadcom insists that for both DPUs and CPE, designs that combine aspects of xDSL and G.fast will enable the easiest transition to faster services.
By contrast, Sckipio says it avoided spinning its processor architecture and vectoring methods from original VDSL designs, instead purpose-building them to achieve optimum power for TDD-based bundles dominated by G.fast. Signaling its belief that G.fast complements VDSL2 and GPON, the company agreed to a design and marketing pact with Lantiq; this pact covers the latter company’s latest GPON and VDSL designs.
Sckipio touts G.fast as incurring inherently lower operational and capital costs because the technology uses time-division duplexing instead of VDSL’s frequency-division duplexing. The four-port DP3000 chipset uses vectoring, a far-end-crosstalk cancellation method, to support as many as 64 subscribers per DPU. Although Broadcom touted the six-port capability of its BCM65200, its reference design at the BWF used two chips to support eight ports because, to date, on-chip vectoring extends to only four ports.
Sckipio’s founders, who came from home-networking specialist CopperGate, raised an initial $10 million in Series A funding with the explicit intent of bypassing xDSL technologies to pursue gigabit twisted-pair modulation. Company principals helped steer the G.fast standard to frequency bands and noise floors that could support 4K HD video over unshielded twisted pair in the final 250 meters to the residence. Absent such a goal, according to Sckipio, the local exchange carrier would lose broadband business to any cable-TV multisystem operator (MSO) with hybrid fiber-coax infrastructure.
Vectoring can serve in VDSL designs but is optional, whereas its use is mandatory in G.fast. VDSL requires a DSL access multiplexer (DSLAM), which traditionally resides up to 1,000 meters from the customer premises. G.fast uses a DPU that relies on the same optical-line-termination systems in DSL networks but sits less than 250 meters from the customer premises. The small DPUs employ reverse powering: they receive power from the customer premises, compliant with Power over Ethernet, helping to simplify installation (see NWR 1/20/14, “G.fast Boosts Copper to 1Gbps”).
In October, Lantiq announced a board-level reference design for a G.fast residential gateway that integrates Sckipio’s CP1000 subscriber modem and the GRX330 gateway processor. Lantiq, Infineon’s former wired-communications business unit, offers a family of GRX3xx processors based on dual MIPS 34Kc cores that integrate packet-handling accelerators. A year ago, the company offered its first VDSL2 chipset to support vectoring: the VRX300 (see NWR 6/24/13, “Vectoring Doubles DSL Rates”). Sckipio and Lantiq will collaborate on marketing opportunities to employ their chips in fiber-to-the-distribution-point (FTTDP) topologies.
Far From Rehashed DSL
Sckipio used a common DSP in both the CP1000-D and DP3000-D digital-front-end (DFE) devices, as well as dedicated blocks to handle DMT modulation. The DP3000-D also integrates a vector control engine that can handle vectoring group sizes of up to 64 lines for multiple DP3000-D devices. A single dedicated analog front end (1000-A) with integrated line driver serves in both CPE and DPU designs—one per CP1000 and four per DP3000. The end-to-end transceiver TDP is 1.5W for the CP1000 and 1.5W per line for the DP3000. Both products feature many power-saving modes as well as a discontinuous mode: an ITU-defined TDD mode that can trade off data throughput and power consumption.
The CP1000-D’s receive path incurs less than –150dBm/Hz of far-end reflected noise in two/four-wire hybrids. The device integrates an Ethernet MAC and PHY with RGMII, SGMII, and 2.5G SGMII as well as MDIO (SMI) management master and slave interfaces. The DP3000-D’s interface options include 1Gbps, 2.5Gbps, and 10Gbps with G.999.1 support for channelization over Ethernet. Also supported in DP3000-D are RGMII, GMII, SGMII, 2.5Gbps SGMII, XAUI, MDIO management, Ethernet backplane support via 1000Base-KX and 10GBase-KX4, and support for 1000Base-X. Both the CPE and DPU devices feature fast online reconfiguration and short train and retrain times. They also offer high immunity to “disturbers”: noise sources in xDSL or G.fast bundles.
Because the DP3000 integrates an L2 switch, it handles 1:4 Ethernet-to-G.fast port switching, which the ITU G.999.1 standard uses when a G.fast transport stream carries Ethernet traffic. Switching can be based on VLAN addresses, Ethernet MAC addresses, or G.999.1 stream IDs. The product supports up to eight QoS priority queues per G.fast transceiver port. Sckipio designed its single-port CP1000-D with a full fiber-termination-unit transceiver; the DP3000-D implements four such transceivers in a single die.
In both the customer-premises and distribution-point versions, the DSP devices integrate IEEE 1588 time-synchronization capability, which is particularly advantageous for mobile backhaul. They also support an 8kHz network timing reference, as well as time-of-day synchronization with an external clock. Both the CP1000-D and DP3000-D come in industrial-temperature-range versions for outdoor use.
The company simultaneously released two development platforms for CPE and distribution points. The CP1000-EVM is a Sckipio reference design that integrates the subscriber modem with DDR memory, a SLIC for connection to circuit-switched phone service, a Wi-Fi MAC, and flash memory. As Figure 1 shows, the DP3016-EVM DPU reference design combines four Sckipio DP3000-D four-port modems (for a total of 16 channels) with a PMC-Sierra WinPath3 SuperLite NPU and G.999.1 software. The four modems use 2.5Gbps SGMII to connect with the NPU, which links to two SFP optical modules for PON uplinks.
Figure 1. Block diagram of Sckipio’s DP3016-EVM DPU reference design. Fiber signals from the service-provider PON enter the PMC chip, and the Sckipio DP3000-Ds bridge to DMT-modulated signals carried over twisted-pair cabling at up to 1Gbps. The 16 Sckipio AFEs integrate a line driver.
Unlike VDSL’s pre-determined frequency window of 17–30MHz, G.fast uses programmable bandwidths of up to 106MHz (though most implementations use 17–88MHz). This capability should increase to 212MHz in the second generation, following a planned G.fast extension developed under ITU auspices. Support for both VDSL2 and G.fast requires reserving 17–30MHz for VDSL, so no real-world G.fast implementation will achieve 1Gbps.
In addition to playing a critical role in developing G.fast standards, Sckipio worked with Celtic-Plus, part of the E.U.-funded Eureka network for joint industry telecom research. Celtic-Plus is field-testing G.fast components in 4Q14.
Two years ago, Lantiq began shifting its VDSL business to vectored VDSL2 by releasing its reverse-powered FTTDP development system, which is based on the Falc-ON GPON controller and Vinax V3 VDSL2 processor. Company principals met Sckipio founders 18 months ago and decided to collaborate on common FTTDP architectures. Lantiq stresses to telcos that the case for vectored VDSL2 or G.fast depends highly on a carrier’s existing topology and its anticipated fanout of multi- or single-family dwellings, but it found that many local carriers are clamoring for G.fast simply because of hype. That urgency led it to work with Sckipio to accelerate its product development.
Both companies bring unique software adjuncts to their chipsets. Sckipio developed a persistent management agent (PMA) for each distribution-point unit, as well as a PMA aggregator to send information on multiple PMAs to a service provider’s network management system. Lantiq offers its Universal Gateway software, which includes AnyWAN port configuration, to set up generic ADSL2, VDSL2, LTE, Gigabit Ethernet, and GPON ports, and it has extended the software to cover G.fast.
If You Build It, Who Will Come?
Sckipio takes the bullish view that G.fast adoption by copper-based telcos will become a matter of sheer survival. Even many facilities-based telcos are rejecting DSL deployments in favor of millimeter-wave radio (see NWR 5/19/14, “Millimeter Wave Seeks Best Home”). Cable MSOs, at least in North America, are positioned to eliminate traditional telcos. Fiber to the home depends highly on full-deployment underwriters like Google Fiber, leaving FTTDP-based G.fast as the only economically viable option for carriers that lack hybrid fiber-coax infrastructure.
Industry watchers such as David Burstein of DSL Prime say that G.fast can play a special role as the only gigabit-per-second option for copper, but that carriers who are reticent to deploy vectored VDSL2 DSLAMs won’t automatically leap to G.fast—even despite its small reverse-powered DPUs. Lantiq says G.fast deployment costs per subscriber are higher than VDSL2’s but much lower than FTTH’s. Transition to the new technology could follow the pattern that the move from ADSL to VDSL established: even though VDSL deployment costs remain more than double those of ADSL, most telcos have already shifted to VDSL for new deployments because it supports new services such as HD streams. The move to G.fast from VDSL2 could follow a similar tack.
Lantiq expects the easiest sell will be carriers that have already deployed fiber to the multidwelling unit. In this case, a G.fast line card can replace a DSL line card. Because DPUs must replace DSLAMs, single-family dwellings may receive upgrades depending on the density of surrounding neighborhoods.
Broadcom supports deployment first to MDU sites with more fiber infrastructure, but it believes a faster move to G.fast is possible if the fiber-to-copper transition point (DPU or DSLAM) can support multiple vectored VDSL2 and G.fast services. By mid-2015, the BCM65200 will enable three vectoring modes: intrachip, board-level interchip, and system-level (board to board). The current reference platform can vector among four ports in one chip and soon will vector among all six ports. Broadcom said that in 4Q14, it will offer vectored elimination of FEXT across multiple chips on a single board. The BCM65200, which should reach production in 2015, will support system-level vectoring when multiple boards employ this chipset.
Few semiconductor companies have tried to retain a portfolio that includes GPON, xDSL DSP devices, and G.fast. For the foreseeable future, Sckipio and Broadcom will be the only DPU competitors, as Table 1 shows. Huawei could still license its own G.fast solution—perhaps through a spinoff, but probably not through its wholly owned HiSilicon, because a chip-level sale requires system-level expertise that exceeds HiSilicon’s capabilities. Ikanos was expected to introduce a Neos G.fast architecture to augment its existing Velocity-Uni VDSL2 chip and Fusiv processor, but it faced significant cash-flow problems in 3Q14 that could postpone or even kill these plans.
Table 1. Comparison of Sckipio and Broadcom chipsets for G.fast distribution-point units (DPUs). Broadcom claims six ports, but it only demonstrated vectoring across four ports at the 2014 Broadband World Forum. (Source: vendors)
When Fiber Is Not an Option
When it announced its chipsets in October, Sckipio unveiled nine design wins. Suttle, XaVi, and Zinwell designed CPE bridges and DPUs; VTech will offer a CPE bridge and DPU as well as a residential gateway. Although designs at midrange companies help validate the market, G.fast will require that large OEMs work with large carriers to overcome the initial hurdles. Huawei, owing to its system-level presence based on its own chip design, is an example of the multinational giants the industry is expecting. Alcatel-Lucent, another large OEM, announced on October 15 that a customer, the A1 domestic-services subsidiary of Telekom Austria, made its first residential G.fast connection (believed to be the world’s first) earlier in the month. A secondary ODM market for G.fast may one day emerge, but we anticipate such an occurrence only after the standard matures and has larger OEMs behind it.
When marketing to telcos, proponents of broadband twisted-pair technologies assume that unsubsidized urban and suburban FTTH networks, such as Verizon’s FiOS, are running out of steam owing to trenching and right-of-way costs. Sckipio avoids positioning G.fast directly against vectored VDSL2, and it allowed that many customers can benefit from 200Mbps performance. Assuming, however, that consumers demand UltraHD or 8K HD—which requires 500Mbps of bandwidth for a single stream—G.fast will be the only suitable twisted-pair technology.
Shifting from traditional DSLAMs (line powered unless configured with a remote power supply) to reverse-powered DPs also entails a learning curve for telcos. Lantiq said that true reverse-powered units in 2014 have yet to see extensive deployment, though it expects trial networks soon.
We see reverse-powered FTTDP topologies as the end game for all current xDSL deployments, though Burstein cautions that a slowly advancing standards effort and concern over power surges in early field trials could hinder the deployment of reverse-powered DPUs. Unless Google or other companies expand FTTH underwriting, the extension of fiber-to-the-network interface devices appears unlikely in the next decade, at least for most North American residences. This situation gives G.fast a clear opportunity to challenge cable’s dominance in HFC topologies. But many struggling regional telcos may be unable to seize this opportunity.
Capital equipment costs remain problematic for players such as CenturyLink, which expects delays even in moving from ADSL2 to vectored VDSL2. Broadcom’s support for combined VDSL2/G.fast bundles could give telcos a less painful way to implement services beyond 300Mbps. But if a shift to vectored DSLAMs takes a decade or more, G.fast and its rearchitected DPUs may fail to achieve a significant presence in North America until the 2020s. Sckipio and Lantiq anticipate faster movement in geographical regions with green-field opportunities, including China and Latin America. If the global economy slows in 2015, however, G.fast could face delays owing to the inherent spending conservatism of many telcos.
Price and Availability
Sckipio’s DP3000 chipsets are in early sampling, as is the Broadcom BCM65200/65900. Sckipio expects the CP1000 to tape out and sample in 1Q15. Broadcom’s existing CPE SoC, the BCM63138, is in full production; a revised version that supports G.fast under the same part number is sampling. Both companies withheld pricing. Lantiq’s G.fast gateway reference design will ship by the end of 2014.
Sckipio products: http://www.sckipio.com/products-2/
Lantiq residential gateway: http://www.lantiq.com/Gfast
Broadcom BCM65200: No product briefs available, general information at http://www.broadcom.com/press/release.php?id=s877202