Editor: Linley
Gwennap In
This Issue
Networking Silicon Market Share 2006 is now available, bringing you all-new market share data. Find out who's gaining share and who is losing. For more information, visit our web site. EZchip Discloses 30Gbps and 100Gbps NPUs During May, EZchip made several significant disclosures regarding new network processors. In its 1Q07 earnings release, EZchip revealed it is sampling one version of its NP-3, specifically the version it codeveloped with Marvell for an unnamed Tier One customer. Other customers will have to wait for the standard NP-3, which EZchip says will sample later this year. It is unclear how the two NP-3 versions differ, but the company has released details on the standard product. Whereas the NP-2 is built at TSMC in a 130nm process, the NP-3 will use a 90nm process at IBM. And while the NP-2 had two 10Gbps interfaces, the NP-3 adds a third simultaneously operating 10Gbps port. As a result, EZchip calls the NP-3 a 30Gbps NPU. The other significant addition to the NP-3 is OAM offload features. This week, EZchip disclosed limited details of its NP-4, which is due to sample some time in 2008. Using 65nm technology, the NP-4 will be a single-chip 100Gbps NPU with integrated traffic management. Compared with the NP-3, the NP-4 will add functions normally handled by a separate fabric-interface chip (FIC). EZchip says the FIC function will enable system-wide QoS in designs using standard 10GbE switch fabrics. The NP 4 will include eight 10Gbps ports, 24 GbE ports, and one 100Gbps port. A single NP-4 should support a variety of designs, including 24xGbE or 4x10GbE line cards as well as a 24xGbE+4x10GbE standalone configuration. With NP-2 revenue ramping, the NP-3 in the hands of a lead customer, and a 100Gbps NPU under development, EZchip is demonstrating both business and technology momentum. Meanwhile, Bay is sampling its Chesapeake 40Gbps NPU/TM, and Xelerated has fresh funding in hand; both have hinted at future 100Gbps products. Although the 100Gbps Ethernet standard won't be completed before 2009, the 100Gbps NPU race is clearly on. —Bob Additional coverage of EZchip appears in our report A Guide to Metro Network Processors. Leveraging the network-processor technology it acquired as part of Agere, LSI last week announced a family of communications processors, the APP2200. These highly integrated products combine dual 280MHz ARM11 CPUs, a pair of packet engines from the APP NPU products, a flexible encryption engine, Ethernet MACs, and other network interfaces. The high-end APP2250 is rated at 3Gbps of throughput, whereas the low-end APP2210 scales down to 300Mbps. Pricing ranges from over $100 for the 2250 to about $45 (in high volume) for the 2210. LSI also disclosed the previously unannounced APP200, which uses an ARM9 CPU and sells for just $20. As in other APP products, the packet engines offload the entire data plane, implementing a flexible classification engine and a robust traffic manager. As networks transmit a mixed of voice, video, and data, traffic management becomes more important. The security engine, apparently licensed from SafeNet, handles a variety of bulk and public-key algorithms and offloads security protocols from the CPU. IPSec performance ranges from 500Mbps to 1,500Mbps depending on packet size. This architecture frees the two CPUs to handle voice encoding, networking management, and high-level applications. The new chip is scheduled to sample to customers in August. The APP2200 targets the emerging SMB gateway, which combines several functions such as router, VPN/firewall, VoIP, and storage. Some even manage video from surveillance cameras. Merging multiple functions into one box reduces overall cost but drives up the CPU requirements. The dual ARM11 CPUs, combined with extensive hardware offload, provide horsepower and flexibility for these various functions. With this new architecture, LSI will compete directly Freescale's PowerQuicc. Although the APP2200 can't match Freescale's fastest processors, it provides plenty of horsepower for SMB along with proven security and packet engines. Now, LSI must build the software support and the credibility to steal design wins from the market leader. —Linley Additional coverage of the APP family appears in our report A Guide to Access Processors.
Last week at Interop, AMCC and Broadcom separately announced availability of their 10GBase-KR transceivers. The IEEE recently approved this new standard (802.3ap) for transferring 10Gbps serial data over the backplane. The initial target applications include blade servers and ATCA systems. Each company had a live demonstration showing the transceiver passing data at 10Gbps using connectors from Tyco or Molex. Both companies are currently sampling their KR chips and are in the process of qualifying them for production. Broadcom's BCM8071 and AMCC's QT2055 each include equalization and transmit pre-emphasis, to ensure error free communications over the backplane. Each device bridges between a XAUI interface and a 10Gbps serial interface. Each can auto-negotiate operation at either 1Gbps or 10Gbps and is packaged in a 13mm BGA. The BCM8071 is manufactured on a 90nm process and has a typical power dissipation of 1.5W. In contrast, the QT2055 is manufactured on a 130nm process and has a slightly lower power dissipation of 1.3W. AMCC and Broadcom have sampled the first KR transceivers at about the same time and with similar features. Each company has reused in-house technologies to reduce execution risk and deliver early samples. We expect volume for these products to ramp in 2008 or later. On the basis of power dissipation, AMCC's QT2055 has a slight edge—especially in designs that will use multiple 10Gbps links. Potentially, AMCC could further reduce power dissipation by migrating its design to 90nm. On the other hand, Broadcom can promise a roadmap that integrates the KR transceiver with its popular switch products—thus simplifying system upgrades. —Jag See what the vendors had to say about 10G Ethernet at our recent seminar on High-Speed Interconnects. At Microprocessor Forum last week, AMCC disclosed the first details of its next-generation Power CPU, code-named Titan. The CPU, codeveloped with Intrinsity, is slated to achieve 2GHz in a standard 90nm process. This amazing speed comes in part from Intrinsity's Fast14 technology. Titan uses a dual-issue, out-of-order architecture but keeps the pipeline to a modest 11 stages, reducing penalties. AMCC expects Titan to burn just 2.5W (typical) per CPU, allowing it to be used in SoC designs with moderate power dissipation. The CPU is designed to work in dual- and quad-core designs. The first Titan-based products are scheduled to sample in early 2008. —Linley Additional coverage of AMCC's Power processors appears in our report A Guide to High-Speed Embedded Processors. Earlier this month, Iamba Networks announced its GPON eco-system (iGES) product line. For the OLT, the startup offers the iSL2402 dual-GPON chip, which is currently sampling. By the end of the year, the company plans to sample a quad-port GPON controller. For the ONT, Iamba announced the iSN1000 product line along with a production-ready system design. The iSN1000 design has been validated for interoperability with equipment at several carriers. The stealthy startup is only the second company to sample a complete GPON solution. The level of completeness has allowed Iamba to get design wins at TEMs and ODMs. By the end of the year, it should have a product line to rival BroadLight's. —Jag Additional coverage of GPON products appears in our report A Guide to Broadband Interface Chips.
On July 18th, The Linley Group will host the next seminar in its Linley Tech 2007 series. This one-day event in San Jose will focus on the technologies driving the future of enterprise and data-center networks. This seminar is intended for network-equipment vendors, server OEMs, system designers, service providers, enterprise-network managers, software developers, press, and the financial community. This event will feature in-depth presentations on the newest chips and technologies for switching and server connectivity from leading suppliers of OEM products for this market. You won't want to miss this outstanding lineup of technical presenters, including:
The seminar will open with a presentation from The Linley Group highlighting market and technology trends and providing context for later presentations. The remainder of the day will include talks and panel discussions covering a broad range of topics such as 10GbE switching, 10GbE backplanes, optical interconnects, and application-aware networking. Full details of the program will be announced shortly. Register now! Attendance is free to qualified attendees that pre-register. This event is sponsored by Freescale, AMCC, Solarflare, Luxtera, Mellanox, Dune, Cavium, and Fujitsu Microelectronics. To receive The Linley Wire via e-mail, click here
|
