Editor: Linley
Gwennap Special Interop Issue
In This Issue
Save the date! Mark your calendars for July 16 for the Linley Tech seminar on Embedded Network Security Design. This seminar is sponsored by Freescale, AMCC, Netronome, LSI, SafeNet, and Cavium. Details of this event will be announced soon. For an idea of what you can expect, check out last year’s detailed program. Broadcom Debuts 10GbE Copper PHY Last week at Interop, Broadcom introduced several new products, including Ethernet switch chips and its first 10GbE-over-copper (10Gbase-T) PHY. All the new products are sampling and are built in 65nm process technology. Broadcom’s Senior VP of the Enterprise Networking Group, Nariman Yousefi, states that in 2007 transistors started costing less on the 65nm technology node than 90nm, despite higher wafer costs. The company’s 10GBase-T PHY, the BCM8481, is a single chip capable of supporting all Ethernet speeds up to 10Gbps. It supports the standard PHY features such as XAUI and SGMII, loopback modes, and cable diagnostics. A XAUI pass-through mode allows the device to connect to an optical module. The chip supports jumbo packets of up to 18KB for improved efficiency. It supports the full 100m distances over Cat6a or Cat7 wiring as well as a 30m low-power mode. One bad sign: Broadcom declined to share power dissipation, which is the most important metric for 10GBase-T PHYs. We believe the power dissipation for the BCM8481 to be significantly greater than the 5.5W-5.9W of competing devices from Aquantia and Solarflare. Although this shortcoming will leave room in the market for the smaller 10GBase-T PHY vendors, we expect Broadcom to move quickly to improve the competitiveness of its product. In any case, the successful release of this product indicates that Broadcom is unlikely to acquire a 10GBase-T PHY startup; these small companies must now find a way to be long-term competitive against the 800-pound Ethernet gorilla. —Jag
Broadcom’s new Ethernet switch chips include the BCM56624 and the BCM56720. The BCM56624 is a 48xGbE switch with 4x10GbE or HiGig2 uplinks. Broadcom has increased the HiGig2 data rate to 16Gbps, enabling a non-blocking switch. Compared to the company’s earlier product, Broadcom has doubled the number of ports and improved several other features. For example, the new device increases the number of tables for L2, L3 and ACL entries. The BCM56624 also has a port to connect with an external TCAM, which can be used to further extend these tables. In addition, the new chip adds carrier Ethernet features such as Q-in-Q and MPLS tunneling. The BCM56624 works with the BCM56720 to create a fabric for a scalable Ethernet switch. The two devices schedule traffic using strict priority, weighted round robin, or deficit fair queuing. The fabric manages congestion using per priority class of service and link pause. —Jag Additional coverage of Broadcom’s Ethernet products appears in our report A Guide to Ethernet Switch and PHY Chips.
Also at Interop, Marvell rolled out its first high-speed embedded processors under the Discovery Innovation brand. These ARM-compatible chips leverage the brand name and system interfaces of Marvell’s popular Discovery system controllers while adding the company’s home-grown Feroceon CPU. The lineup includes three product families: the dual-CPU MV78200, the single-CPU MV78100, and the lower-cost MV76100. Marvell is one of the few companies with a license to develop its own ARM-compatible CPUs. The company first debuted the Feroceon CPU in 2005, but the initial products were all application-specific. The new Discovery processors target a range of applications, including communications, networking, consumer, and other embedded designs. With two superscalar 1.2GHz CPUs, the MV78200 should be competitive in performance with most of Freescale’s PowerQuicc processors while offering lower cost and power. Although the current products do not scale to the large number of CPUs offered by Cavium and others, they hit the sweet spot of the market, and Marvell promises to quickly deliver multicore products. Marvell’s biggest challenge is convincing its traditional communications and networking customers to switch to ARM. The Discovery system controllers support only MIPS and PowerPC processors; these customers will have to recompile their software for the Discovery processors, although they can at least leverage the I/O drivers. ARM has been perceived as too low in performance for most networking applications, but the new Marvell processors may change that perception. —Linley Complete coverage of the new Marvell processors appears in our upcoming edition of A Guide to High-Speed Embedded Processors.
Marvell also made several announcements targeting Carrier Ethernet applications. The company announced two switch products: the 98DX2100 and 98DX4100. To minimize power dissipation, each is manufactured in 65nm. To reduce cost, the devices integrate Flash memory and an ARM CPU operating at up to 1GHz as well as a DDR2 DRAM controller. The CPU and memory controller are similar to those in Marvell’s new Discovery processors. The DX2100 is a Fast Ethernet switch with Gigabit Ethernet uplinks. It targets MTU applications that connect subscribers over FE and use the GbE uplinks for active Ethernet or GPON connections. The uplinks can operate as fast as 2.5Gbps, matching the data rate for GPON. The DX2100 also supports TR-101 and can map Ethernet QoS from the subscriber to GPON packets. The DX4100 targets metro access and Carrier Ethernet (CE). It supports most of the recent CE standards, including IEEE 802.1ad for Q-in-Q, 802.1ah for MAC-in-MAC, 802.1ag for CFM as well as VPLS and VPWS standards. The chip also supports IEEE 1588 for synchronous Ethernet operation. It can be used for active Ethernet or as a switch on IP DSLAMs. The DX4100 family offers several products; the flagship device includes 24xGbE and 4x10GbE ports. The DX4100 includes on-chip tables for MAC addresses, IP addresses, and ACLs. It is not possible to expand these with external memory, making the product best suited for metro access. With the DX2100 and the DX4100, Marvell has enhanced the DX product line to meet specific application and customer requirements for extending Ethernet switching into carrier applications. —Jag Additional coverage of Marvell’s Ethernet products appears in our report A Guide to Ethernet Switch and PHY Chips.
The demand for high-speed processors continues to grow as market opportunities abound for these speedy chips. These devices are ideal for handling the complex functions required by networking applications, automotive, consumer devices, industrial control, medical imaging, and much more. These applications share a need for speed but also the flexibility that a general-purpose processor provides. These chips use standard instruction sets such as MIPS, PowerPC, ARM, and x86, allowing programmers to use a wide variety of operating systems and development tools. This completely revised report provides extensive coverage of high-end embedded processors from a mix of old and new vendors, featuring AMCC, AMD, Broadcom, Cavium, Freescale, IBM, Intel, Marvell, PMC-Sierra, RMI, and Via Technologies. This edition covers all the recent announcements, such as Intel’s Atom and Tolapai, Marvell’s Discovery processors, AMCC’s PowerPC 460, and Via’s Isaiah. It also provides up-to-date product roadmaps as well as market share and market size data for the embedded segments covered. Learn how multicore and multithreading technologies are impacting the embedded world. Find out which target markets and applications bear watching. The Linley Group provides the technology analysis you need to make informed business decisions. Order A Guide to High-Speed Embedded Processors by May 31 to take advantage of the prepublication discount. For more information on this report, visit our web site.
To receive Linley Wire via e-mail, click here
|
