Gigabit Ethernet

In case you haven't noticed, the Ethernet Standards Committee (IEEE 802.3) is in full swing once again.Just one year after the introduction of Fast Ethernet 100-Mb/s systems, the LAN industry is about to make another ten-fold leap in performance and throughput. The next version of Ethernet will called Gigabit Ethernet, and it will run at a delivered data rate of one billion bits per second.

If you thought Fast Ethernet 100BASE-T layouts were tricky, think again. Gigabit Ethernet is going to be faster, with more parallel signals, and tighter layout constraints. It will work, and it will work reliably, but you will have to follow the rules. There is not going to be much room for error.

In early implementations, the basic Gigabit Ethernet physical layer hardware will be comprised of six pieces:

  1. MAC/coding sublayer chip
  2. ANSI "10-bit" serializer chip
  3. (optional) connector
  4. (optional) fiber optic electrical/optical converter
  5. media connector
  6. serial media (fiber optic or copper)

Over time, the MAC/coding chip and the serializer function will be merged together onto a single substrate. That will cut out the 10-bit serializer interface, simplifying designs. Until then, pay close attention to the termination and layout of all the lines running between the MAC/coding chip and the serializer, especially the clock. This is one part of the system where a tenth of a nanosecond will make a noticeable difference. Pay particular attention to the quality of power filtering applied to your transmit clock source. The 10-bit serializer at the far end of the link does not respond kindly to frequency wander, or phase jitter, in the transmitted data.

Other architectures are under consideration for possible transmission over category-5, unshielded, twisted-pair cabling. Yes, that's right, unshielded, twisted-pair (UTP) cabling.

It turns out that category-5 UTP has just enough Shannon information-theoretic bandwidth to carry a Gigabit datastream at distances up to 100 meters. It's not a matter of whether the cable can do it, it's more a matter of whether we can build chips fast enough to recover and process the received signal. Communications engineers know how to do this type of processing for satellite communications, data modems, and other applications today, but not at Gigabit rates. As chips shrink, and consequently go faster, many people assume that the technology required to run at a Gigabit will surely emerge. When and how a final design will shake out for UTP is unknown at this time.

So when will Gigabit Ethernet designs become commonplace? At present, 76 companies have indicated their desire to promote and support the Gigabit Ethernet Standards process by joining a private industry group known as the Gigabit Ethernet Alliance. It's a good bet that all of them will be heavily involved in Gigabit Ethernet by year's end. That means at least 76 of you (probably a lot more) will be doing Gigabit Ethernet design in the near future. Market predictions for Gigabit Ethernet system-level sales are running as high as $600 million a year around the turn of the century.

Key ingredients in any Gigabit Ethernet design will include:

  • Careful control of clock skew on the 125-MHz 10-bit bus.
  • Attention to crosstalk between massive, 125 MHz TTL-level parallel buses and critical high-speed serial circuits.
  • EMI considerations on the serial cables (if you are not an expert in this area, I suggest choose fiber)
  • Huge pin counts, especially on complex switching or routing ASICS, some of which will be using 512-bit bus interconnections
  • More than usual emphasis on time-to-market; everybody will want their products ASAP

If you are planning a Gigabit Ethernet design, take the time to get the latest information from the Gigabit Ethernet Alliance web site:

  • URL: [Ed. note—As of 2015 the allicance is no longer operative]
  • Chairman: Tony C. Lee , Extreme Networks
  • Vice Chairman: Nathan Walker, Cisco Systems

The Gigabit Ethernet Alliance web site includes a list of chip, connector, and cable vendors that can supply useful raw information for your design process.Here is a list of the companies currently involved in the Gigabit Ethernet Alliance (as of late 1996):

Table 1—Gigabit Ethernet Alliance Members

3Com Corp Ancor Communications Bay Networks
Cabletron Cisco Systems Compaq Computer
Digital Equipment Corporation Emulex Extreme Networks
Fujitsu Microelectronic Hewlett Packard IBM
Integrated Circuit Systems Intel Corporation Ipsilon Networks
LSI Logic Lucent Technologies Madge Networks
National Semiconductor NeoNetworks, Inc. NetStar
Packet Engines PlainTree Systems Prominet Corporation
Rapid City Communications S-MOS Systems Siemens AG
Sun Microsystems Texas Instruments UB Networks
VLSI Technology WideBand Corporation Xircom
XLNT Designs Acacia Networks Accton Technology Corporation
Adaptec Alliance Semiconductor Allied Telesyn International
Alteon Networks AMD Apple Computer
Asante Technologies Auspex Systems, Inc. Brooks Technical Group
CellSwitch Networks, Inc. Cray Communications Digi International
Essential Communications Corporation FORE Systems, Inc. GEC Plessey Semiconductors
GigaLabs, Inc. Hitachi Cable, Ltd. Hitachi Internetworking
LanOptics Macronix America, Inc. Mammoth Networks
MicroOptical Devices, Inc. MMC Networks Motorola.
Myricom NBase Communications NBase Fibronics Ltd., an Elbit Company
NEC Electronics, Inc. Network Peripherals ORNET Data Communication Technologies
PMC-Sierra SEEQ Spike Technologies, Inc.
Standard Microsystems Corp. Sumitomo Electric Industries, Ltd. Synergy Semiconductor
UNI Vitesse Semiconductor Corporation Xylan

Please don't let the tone of this article dissuade you from engaging in Gigabit Ethernet design. The technology works. The component vendors are there to help with specific layout tips at every level of the system. You can get it to work. Just remember to listen, and listen carefully when your component vendors give you layout advice. Strict adherence to their guidelines will usually help keep you out of trouble on this type of super-fast design.