In March 1997 Cisco CEO John Chambers told The Red Herring that his greatest worry was that the third great tech monopoly would become "fat and happy" (see "Routing the Competition"). Given Mr. Chambers's insistence that it is "paranoia that keeps us on top," Cisco's reluctance to respond to a growing threat to its high-end router business is all the more puzzling.

That threat is Layer 3 switching, also known as wire-speed routing, which takes place at the network layer rather than the data link layer, where traditional switching occurs (see table below). It's a technology that not only increases routing performance from several hundred thousand packets per second to 4 million packets per second but also drops the price of routers to one-tenth that of Cisco's bread-and-butter products. "Cisco is in a precarious position," says Nicholas Lippis, president and founder of Strategic Networks Consulting (a subsidiary of Softbank, the parent company of Ziff-Davis, which has a minority investment in The Red Herring). "It's trying to protect its router business as it enters a fiercely competitive market with no apparent lock-in."

What's enabling this dramatic improvement in routing is ever more sophisticated semiconductor technology. Silicon is hardly new to the networking business; it has always been integral to switching. But increasing network complexity, skyrocketing demand for networking products, and standardization of the TCP/IP protocol are driving complex functions like routing and network traffic management into silicon. The reason is simple: silicon is much faster and cheaper than software.

This trend is bringing the networking and semiconductor industries closer together than ever before, and it's creating enormous business opportunities at two opposite ends of the market. At the high end, Cisco, Bay Networks, 3Com, and Cabletron Systems, plus startups like Yago Systems (acquired in January by Cabletron) and Acclaim Communications, are experimenting with feature-intensive, lower-volume productslike Layer 3 switchesthat are developed using homegrown application-specific integrated circuits, or ASICs. At the low end, companies like Galileo Technology and Texas Instruments are commodifying the market with higher-volume, basic products like Ethernet, Fast Ethernet, and token ring switches.

(For the record, ASICs are fixed-function chips in which the software element of a programmable gate array is hard-coded into silicon. ASICs can lower the overall system costs by reducing the number of components needed and also by improving performance and reliability.)

Layers upon layers The Open System Interconnection reference model.
7 ApplicationConsists of application programs that use the network.
6 PresentationStandardizes data presentation to the applications.
5 SessionManages sessions between applications.
HIGH END4 Transport Provides end-to-end error detection and correction. Network traffic management occurs here.	Key companies3ComBay NetworksCabletronCiscoAcclaim Comm.Yago Systems
3 NetworkRoutes data to different networks. Traditional routing and Layer 3 switching occur here.
LOW END2 Data linkTransmits data from node to node. Traditional switching occurs here.
1 PhysicalDefines the physical characteristics of the network media.

Burning at both endsTo reflect this convergence and to explore the implications of semiconductor technology at both ends of the networking market, this story is divided into two parts. The first examines how silicon development enables new technologies like Layer 3 switching and Gigabit Ethernet. The second, which will run in April's semiconductors briefing, will look at the opportunity for established chip companies and smaller niche firms to outsource low-end products to networking systems vendors.

The importance of ASICs to the high end of the networking industry has been well documented since the advent of switching. Cisco led a string of ASIC-related acquisitions with its December 1994 purchase of Kalpana, an Ethernet packet-switching company that first introduced ASICs into the data networking industry, according to Vish Akella, a Kalpana founder and now CEO of Acclaim Communications.

Team buildingCisco acquired Grand Junction Networks for its Fast Ethernet switching technology in September 1995 and Granite Systems for its multilayer Gigabit Ethernet switching technology in September 1996. When Bay Networks bought Fast Ethernet switching vendor NetICs in December 1996, all the startup had to offer were "some good ASICs," Mr. Lippis says. NetICs was ultimately credited with pulling Bay out of a slump that began in early 1996.

But 1998 may be the year ASICs make their biggest splash yet in networking by enabling new technologies like Layer 3 switching. As UBS Securities analyst Scott Heritage explains, "Traditional routers are much more complex devices than switches, and they're based on software, which makes them slow. But now some companies have developed ASICs that can put that routing code into hardware to speed up the process and lower the cost tenfold."

And Layer 3 switches go one step beyond merely routing data faster; they can also manage bandwidth intelligently. Mr. Akella founded Acclaim in 1995 to develop Ethernet switches that fuse LAN switching with WAN routing. Acclaim's products introduce quality-of-service features that are standard in telecom infrastructures--like prioritization and security--to the networking infrastructure.

Although International Data Corporation analyst Esmeralda Silva expects each of the Big Four vendors to roll out Layer 3 switching products in the first half of this year, Mr. Akella says Acclaim is a step ahead at this point. "Right now, our technology is superior, but obviously we're not as well known," he says.

Delayed gratification Layer 3 switching products will be used with token ring, Ethernet, and Fast Ethernet networks this year because they cost less than traditional routers, but analysts note that customers won't get to experience the new technology's true performance benefits until Gigabit Ethernet networks become common after 1999.

Because of this synergy, some companies are designing chips that have both capabilities. One of these players is Yago Systems, which Cabletron bought for $213 million. Yago cofounder and CEO Piyush Patel says that "few companies have managed to combine routing and gigabit bandwidth without cutting corners and compromising performance." Yago, he claims, is an exception. With engineering talent poached from Intel and Sun Microsystems, Yago has introduced a router that can incorporate traffic management features all the way up to the application level--innovative functionality for a device that originally was designed simply to move data from here to there.

Together, Cabletron and Yago will compete directly against Cisco and Bay Networks. Even before the acquisition, Cabletron worked closely with Yago and was a minority investor. "Cabletron is a good partner," Mr. Patel said tactfully before the acquisition was announced. "We fit well into their road map." (For a more skeptical take on Cabletron, see "Holding Pattern.")

Shake it upAs early as last April, Strategic Networks Consulting was calling Layer 3 switching "a technology revolution akin to the bridge-router transition of a decade ago, which created a new order within the networking industry and brought names like Cisco and Wellfleet to Wall Street." Though it is early to begin drawing conclusions about winners and losers, it's clear that Layer 3 switching has far-reaching implications for each of the Big Four.

Cabletron hasn't had much exposure in the routing market to date, so a successful Layer 3 switching acquisition could give it an opportunity to make an impact in that market and put itself back on the map alongside 3Com, Bay Networks, and Cisco.

Bay Networks hopes to repeat its NetICs success with its latest acquisition, Rapid City Communications (acquired in June), which makes ASIC technology for Gigabit Ethernet switches with Layer 3 switching capabilities. Bay's routing switch finally hit the market in December under the direction of Joe Kennedy, the former CEO of Rapid City and Bay's new vice president of switching. Mr. Lippis says the switch "could be a real winner," although Mr. Heritage of UBS predicts that Bay will lose some of its router sales to the new product, resulting in a neutral effect overall.

Mr. Heritage anticipates a similar predicament for 3Com, which is developing its Layer 3 switching technology internally, rather than through acquisition. Its third-generation CoreBuilder technology has received positive feedback so far, including a Best of Show honor at Interop last year, but any success with routing switches will likely be countered by 3Com's susceptibility to continuing price declines at the lower end of the switching market.

Do these developments leave Cisco in the dust? Not quite. Mr. Patel of Yago asserts, "Layer 3 switching products are orders of magnitude better than Cisco's routers," but Mr. Lippis cautions that not having these capabilities isn't hurting Cisco yet. And the routing behemoth still has a few cards left to play: its loyal installed base, strength in the routing market, and phenomenally deep pockets. Jayshree Ullal, vice president of Cisco's enterprise line, insists, "We've never been afraid to 'eat our young' as our customers demand it, and our move into LAN switching has shown we can adapt."

Lastly, Cisco doesn't have to be the fastest one out of the starting gate in order to win. "Cisco has never been first to market in the past," Ms. Silva says. "They can afford to lose some of their high margins. The real question is whether they can match the performance of Layer 3 switches."

Cisco can probably afford to arrive late with Layer 3 switches, but it certainly can't ignore the natural progression of the silicon cycle, which is to integrate functionality into hardware as standards solidify. Historically, networking vendors excelled at making different networks--like DECnet, AppleTalk, and Systems Network Architecture--talk to each other. Then they began building their own silicon to increase the performance of these multiprotocol functions. Now, as the industry standardizes on the TCP/IP protocol, companies can design ASICs that move even more of the meat of networking products to hardware.

Elias Moosa, a senior semiconductor analyst with BancAmerica Robertson Stephens, says the data communications market has reached "the steepest part of the curve, where we're migrating more and more to silicon." This means that across the market, the relationships between networking and semiconductor companies will grow more intertwined, and the companies that can both design silicon well and work well with semiconductor companies will likely succeed in the future. As Mr. Lippis puts it, for the high end of the market, "the fact that Layer 3 switches can be plopped into an existing network of Cisco routers or Cabletron switched LAN segments means open season. Incumbent vendors still hold an advantage--but not a hammerlock."