One of the great success-failure-success stories of the semiconductor era involves the gallium arsenide integrated circuit. As a laser or light-emitting device, gallium arsenide, or compounds like gallium arsenide phosphide, enjoyed steady growth. But the story was different with the IC—and very uneven.

Depending on their product orientation and emphasis, some GaAs companies saw relatively steady growth; some suffered steep declines, then recovered; and some died.

GaAs field-effect transistors first appeared in the late 1950s and 1960s. Even before then, scientists and engineers regarded GaAs as the technology of the future because of its high electron mobility, thus high-frequency operation and high switching speed, its resistance to radiation and its light emission.

All through the 1970s, large companies were moving the technology from research into commercialization. In the United States, companies like the Rockwell Science Center, Bell Labs, RCA, Hughes, Westinghouse, TRW and most defense operations had GaAs labs. In Japan, Fujitsu, NEC and Toshiba made GaAs amplifiers. Almost every major R&D center in the United States, Europe and Japan had a GaAs effort.

The emphasis was on discrete transistors, most of which found their way into power circuits or low-noise front ends in commercial communications and satellite receivers. They were replacing silicon transistors and going into frequency ranges that silicon couldn't reach.

In the 1970s and more powerfully in the 1980s, the efforts went into integration, starting with tens of transistors on a chip, then moving up. In the early- to mid-1980s, when most of today's GaAs leaders and several failures started, there was enough interest and venture capital to make a commercial GaAs industry, not just a military-contracts business.

Some of those contracts were by no means trivial. The Defense Advanced Research Projects Agency had great interest in the future of GaAs. Some of Darpa's research contracts ran upward of $10 million.

The new GaAs companies moved in two separate streams: high-speed switching and high-frequency communications. The two streams had two main and separate targets: supercomputers and military, the latter largely concerned with radar applications and communications. Within each stream, there was further division, as some companies aimed at greater integration and others focused on greater speed.

All streams grew into rivers until two events stanched the flow. The fall of the supercomputer business slammed the digital-IC business, and the fall of the Berlin Wall and Soviet Union slammed the military analog-IC business.

On the digital side, a principal aim was to make ICs that could provide faster switching speed than available with silicon-based emitter-coupled logic, the speed king of the day, and lower power consumption.

In 1968, Motorola, a leader in emitter-coupled logic, introduced its MECL III gate array. The chip could be tailored to different functions, including an oscillator, with frequencies up to 225 MHz, or a logic circuit, with rise and fall times of 1800 ps. MECL was very fast, but it chewed up power, up to 25 mW with a 5-volt power supply. Target markets were supercomputers, automatic-test equipment and disk-drive electronics - places where lots of data had to move very fast.

ECL for computers and ATE

ECL, from Motorola as MECL, and from others, was growing nicely as supercomputers and automatic-test equipment flourished. Then, in the late 1980s and early 1990s, high-end desktop computers and high-performance parallel processing began killing supercomputers.

To make matters worse, CMOS circuits were catching up in speed, and GaAs chips, despite their higher cost, were chewing away at ECL markets, resulting in a leveling off, not a business slowdown, says Raj Maini, manager of logic-market development at ON Semiconductor (formerly part of Motorola).

GaAs manufacturers claimed that they could offer the speed of ECL with one-third the power consumption, along with equivalent pricing. They replaced many ECL gate arrays at major supercomputer companies like Cray, Amdahl and Convex. But they couldn't beat ECL circuit density.

After the supercomputer effect, says Maini, ECL started growing again, thanks to newer ECL technology and rapid growth in long-distance communications and networking. ECL circuits are now flooding network routers and switches, telephone central-office equipment and cellular base stations.

ECL circuits now typically operate with 3.3-V supplies-the same voltage as GaAs ICs—rather than the original 5 V and are moving toward 1.8 V. The power dissipation of the circuits is down while speed is up. Typical rise and fall times are now 500 ps, with dissipation of 25 mW per gate.

Power dissipation does not rise with operating frequency in ECL circuits, as it does with CMOS. So ECL dissipation matches that of CMOS beyond about 400 Mbit/s. ECL now occupies a powerful niche between CMOS, which can offer millions of gates on a chip, and GaAs, which can operate at frequencies beyond 15 GHz, compared to 3 GHz for ECL.

ECL often provides the glue logic in systems using other technologies. And recent developments in silicon-germanium chips, pioneered by IBM, may give ECL a further boost in competition with GaAs. But GaAs vendors are moving into silicon-germanium, too.

A steep roller-coaster

The picture with gallium arsenide has been dramatic, with steep declines and explosive growth. Most companies focused sharply on specific markets and technology. And GaAs has seen several dropouts along the way, among them Gain Semiconductor, which died, and Microwave Semiconductor, which was acquired by Siemens.

Typical of the up-down-up scenario was the history at Vitesse Semiconductor Corp. Co-founder and CEO, Lou Tomasetta came from Rockwell Science Center, a west-coast GaAs pioneer. Vice president and general manager, Chris Gardner, came from Bell Labs, an east-coast GaAs pioneer.

Founded in 1986, Vitesse started by going against the ECL gate-array leaders, Motorola and Fujitsu, and capturing some of their business, primarily at major supercomputer companies, like Cray, Convex and Amdahl. Vitesse focused on digital GaAs ICs and on building more complex LSI chips, rather than just faster chips. The company felt that more gates on a chip lead to better manufacturability, lower power consumption and lower cost.

Then, in the early 1990s, as supercomputers collapsed, the company took its beating. But it wasn't fatal. "We had to remake ourselves," said Gardner.

Vitesse moved into the communications business - the wired cellular infrastructure—and into routers and switches. And it went after fiber-channel disk drives and RAID storage systems.

The company's concentration on greater integration led to gate arrays with as many as 200,000 gates. Some of these chips go into automatic-test equipment at companies like Teradyne and Schlumberger. Gardner said those testers are almost like supercomputers.

Vitesse started as a gate-array ASIC company. It now does 90 percent of its product development and revenue in standard products, so its average gate count is declining. It still emphasizes digital communications, with gate speeds up to 10 Gbit/s. The company is the leader in chips for fiber-optic communications.

Shifting to communications

Vitesse, in California, started in 1986 with supercomputers as the main target for its digital ICs, took its lumps when supercomputers did, then shifted very successfully into communications.

In New Jersey, Anadigics started in 1985 with analog ICs, mainly for power amplifiers and low-noise receivers. The company stayed in the analog business, took its lumps when the defense business tanked after the Soviet Union went away, then made a stunning recovery when it switched from defense applications to mass-market ICs in the early '90s.

Anadigics made the first GaAs IC to sell for less than $5. It was a low-noise block converter that receives a block of frequencies in the Ku band (11 to 12 GHz in those days), amplifies the block, then downconverts to an i-f frequency of about 1 GHz, which travels through a cable to the set-top box in somebody's living room.

Chairman and cofounder Ron Rosenzweig said that in the 1992-1993 time frame, Anadigics was shipping about 3 million block-converter chips for the European satellite market. That alerted everybody to the fact that GaAs ICs could be sold to the commercial and consumer markets.

The company moved down the cable to the set-top box, with tuners that enjoy 60 percent of the market. When the cellular market began to explode in 1993 and 1994, Anadigics went after it with receivers and amplifiers for cell phones. GaAs really shines in this area because of its low power consumption.

To reach a telephone central office, the cell phone must radiate one watt and, no matter how you slice it, says Rosenzweig, that watt must come from the cell phone's battery. So the more efficiently you can convert dc power to rf power, the more talk time you can get.

Impact of the U.S.S.R.

While the decline of the Soviet Union was a scary setback for GaAs communications companies serving mainly the military, it turned out to be a fabulous blessing. It opened cell-phone markets in places like China and Russia that were closed during the Cold War. The countries destroyed by the Cold War are now rebuilding their infrastructures, and they need cell phones.

TriQuint Semiconductor, a spinoff of Tektronix in 1985, became a special situation where three money losers combined to form one winner. To start, TriQuint focused on digital telecom ASICs, military ASICs, with developments fueled by Darpa, and on microwave ASICs, primarily for satellites and global positioning systems. These activities laid the groundwork for the cellular market, though nobody knew it at the time.

In 1991, TriQuint acquired Gigabit Logic, which concentrated on digital ASICs for computer manufacturers like Prime Computer and Cray Research, and Gazelle, which produced off-the-shelf communications and timing products aimed at high-end PCs and workstations.

Prior to the merger, all three companies were losing money mainly because too many companies were chasing too few opportunities, said Don Larson, director of marketing for the Telecom and Computing Division.

Before the acquisition, Gigabit sold much of its fab facility to supercomputer manufacturer, Cray, which sold it to M/A-Com in 1995.

The 3-in-1 business helped TriQuint achieve critical business mass, said Larson. So when military cutbacks came along, there was only a temporary setback because the company had already begun to focus on commercial business.

After the mergers, TriQuint focused on the cellular business and quickly became profitable. The company has been very successful with chips for fiber optics and recently acquired a fab from Texas Instruments, which added strength in millimeter-wave chips for satellite and military communications.

Like many others in the mid-1980s, M/A-Com, which started life as Microwave Associates, focused mainly on the defense business. Finbarr McGrath, product-line manager for multifunction ICs at the company's Semiconductor Unit, points out that the volume in military phased-array radars was so great that most defense contractors invested in their own GaAs fabs or worked with companies like M/A-Com that already had GaAs expertise.

In the early 1990s, the fall of the Berlin Wall and Soviet Union led to cutbacks in defense spending, resulting in cutbacks in M/A Com's revenue. At the same time, cell phones began to proliferate, so the company switched to the cell-phone business.

It started with discrete GaAs FETs and simple, single-function ICs, then moved up to highly integrated multiband, multimode chip sets, both for cell phones themselves and for base stations.

In 1995, M/A Com bought the GaAs fab from bankrupt Cray, which Cray had bought from Gigabit Logic before Gigabit was merged into Tri/Quint. That was fortunate, said McGrath, as M/A-Com was able to convert a fab originally aimed at supercomputers into one aimed at cellular handsets. It's now converting to 6-inch wafers.

GaAs is so powerful at high speeds that many traditional silicon-semiconductor companies have been converting parts of their output to gallium arsenide. But it's not a one-way street. In recent months, leading GaAs houses have been moving into silicon.

Some 60 percent of Vitesse's designers are now working on CMOS products. In the past year, Vitesse acquired three silicon companies. RF Micro Devices, another important GaAs house, has working parts in silicon-germanium. Anadigics has shipped a frequency-synthesizer silicon IC, and TriQuint has begun internal silicon developments.

Why the switch? It's the low cost of CMOS and the low-noise attraction of SiGe, particularly in high-frequency power amplifiers.

And considering its staying power, don't count ECL out either.

The Century of the Engineer: Misunderstood Milestones