Though analysts predict that indium phosphide-based products will become a billion-dollar business over the next decade, indium phosphide's transition from the research labs to the market place comes during a telecom business environment that would try the patience of Job.

Not only are investments in fiber optic-based communications being stunted by overcapacity, corporate debt and accounting scandals, but a potentially lower-cost technology, silicon germanium (SiGe), will compete with InP in the market for very high-speed electronics.

While carriers are cutting back investments, the good news is that to remain in business telecom equipment vendors must develop new equipment that will reduce costs as they add channels to 10 Gbit/second OC-192 multiwave-division systems, and develop OC-768 systems that hit 40 Gbit/s.

Behind the scenes
Gary Franzosa, director of marketing at Inphi Corp. (Westlake Village, Calif.), said, "If you look at 40G by itself, the design cycle for these OC-768 systems is very long. By the time you back up from systems deployment to module manufacturing to IC vendor, you are looking at two to three years. In 2002 there are system vendors deploying trial systems, and a lot of the design sockets for 40G are going to be won this year. There's a lot more activity going on in 40G than meets the public eye."

For the 10-Gbit/s systems, system vendors may adopt InP-based components in their second-generation OC-192 cards, he said. "People want to take advantage of existing systems, make their OC -192 systems more robust, and to do that they are looking for specific components that make their existing solutions more cost effective. There are a lot of requirements by the large system houses right now for specific components that can take advantage of indium phosphide," Franzosa said.

Technically, there is really nothing quite like indium phosphide. Carriers (holes and electrons) arguably move through it faster than through any semiconducting material known to man, at lower per-transistor power consumption. And it is among the few semiconductor materials that can be fashioned into photonic devices that transmit and receive light.

Aaron Bond is the chief technology officer at T-Networks (Allentown, Pa.), a startup developing optical modulators. "One of the many interesting things about indium phosphide is that it is the only material that exists that can generate, modulate, amplify and receive light at telco wavelengths — 1.55 and 1.3 microns — the very wavelengths that the telecommunications industry uses in its single-mode optical fibers. And it has one of the highest bandwidths for transistors," Bond said.

Cost, followed by power consumption, are the primary concerns of telecom system vendors. The traditional weapon in reducing cost — higher levels of integration — is particularly complex in the InP industry, where both photonic and electronic functions can be brought to bear.

"We are leveraging the semiconductor integration possible in indium phosphide, trying to push the limits of distance, how far InP photonics can go," Bond said.

The modulators from T-Networks, for example, integrate the amplification, modulation, shutter and TAP functions, which turn a continuous-wave laser into a signalling system.

"We will replace a lithium niobate modulator, an external TAP, and an amplifier. We put those in one package, throwing away the cost of those extra packages for three components that don't have to be bought separately. Even if the parts alone would cost the same amount of money, the customer will save on manufacturing: There is no need to place three parts on a board, with an external splice, extra fiber handling — all those things that result in extra costs," Bond said.

For photonics, Bond argues that gallium arsenide-based photonics devices don't work very well for signals that must travel for more than 10 kilometers (six miles). Unlike the difficult-to-handle lithium niobate, indium phosphide is a semiconductor material that takes advantage of batch processing.

Limits to integration
Integration — which follows Moore's Law in the silicon industry — is more complex in indium phosphide. Companies must decide which functions are best handled by CMOS (or SiGe BiCMOS).

And they must decide which photonic functions can best be integrated with the electronics — and which are best kept separate for cost reasons.

Marko Sokolich, a research manager at HRL Labs (Malibu, Calif.), said, "For SiGe, most things are pretty well understood, but indium phosphide still has so much potential. There are so many things we can do with bandgap engineering that indium phosphide has a long way to go. A lot of engineering work on the materials is ahead of us, and when you fiddle with the layers you come up with ways to solve the barriers to speed and breakdown voltage."

Sokolich cautioned that there are practical limits to the combination of high-performance photonics with the high-speed electronics. "You can integrate the photodiode for short-range photonics. But the reason the communications people keep coming back to indium phosphide is for performance reasons — we can get 50-GHz clock rates. And it is going to get cheaper. We can already multiplex a few thousand transistors to do the main functions on the same die."

However, to try to force a high-performance optical function — other than a short-range PIN detector — on the same die with the digital functions would be "wasteful," he said, because the material stack would become too complex, too expensive.

"A 1.55-micron VCSEL (vertical-cavity surface-emitting laser) requires islands of growth four to five layers thick. The process to achieve that is very expensive — 50 percent more to perhaps double the cost. For that reason, technology may move toward more advanced packaging rather than doing all of the photonic and electronic functions on the same substrate," Sokolich said.

"Taking an SoC approach could result in more expensive fabrication, lower performance and a more inflexible choice of components. What you really want is to optimize the components, so you may not even want to add a PIN," he argued.

Lester Eastman, a professor at Cornell University, is considered by many to be the father of the indium phosphide industry. Many of the 110 or so Eastman students who have earned their doctorates at Cornell have gone on to achieve distinction in the compound semiconductor field. More recently, Eastman has focused on gallium nitride, but in the late 1970s and 1980s, his lab was the first to use an molecular beam epitaxy (MBE) system to create InP structures.

Eastman has a practical bent as well. "In industry, people must be concerned with production reproducibility, and people with experience came to realize that to stack up a laser structure along with the transistors would require a pretty fancy process," he said. "The laser is tall, with a nonplanarity that makes it difficult to do the machining. In our experience when we tried it, at the lower levels the spin-on resist would drift in funny patterns.

"It might be physically practical, but not affordable. Of course, people are taking very inventive approaches to this problem, but will it be profitable? That is the question," Eastman said.

Even within the electronics realm, high levels of integration are several years away. For the near term, companies may use simple V-connectors to link discrete components. But as signals cross chip boundaries over bond wires, degradation occurs that calls for higher levels of integration.

Ray Milano, analog and optoelectronics manager at Vitesse Corp. (Camarillo, Calif.), said, "It's obviously tough, but, long term, the integration of the optical and electronics functions for communication system functions is the challenge. The more chip boundaries you cross the more you corrupt the signal, and you have to correct for that. The first to a killer app, to put a number of the key pieces in place in a single device over the next two to three years, will be a winner.

"We will achieve that in steps, but ultimately the monolithic approach will win out because of these significant packaging issues. Certainly within a decade there will be sophisticated PIN detectors integrated with the transimpedance amplifier, with the modulation and control functions on the same substrate," Milano said.

For optical communications Vitesse has fielded a five-chip set, including the TIA, the limiting amplifier, and a 5,000-transistor heterojunction bipolar (HBT) 16:1 MUX digital circuit with an on-chip clock management unit.

Vitesse's strategy is to work with optics customers that complement its electronics products. It is developing planar optics processing capability in InP — for example, optical wave guides — that it offers through its InP foundry service.

Process advances needed
Even so, the InP community faces severe cost hurdles, with process costs several times higher than GaAs. InP foundries can put 5,000 or more transistors on a single device, but Franzosa said that to keep yields high and costs down, some of the most complex functions — such as the demux — are created at Inphi with only 1,500 transistors.

Integration takes on different meanings in the optical communications world, where adding functions — such as wavelength selection and switching, as well as detection or generation of optical signals — can be integrated with electronics.

Milton Feng, a veteran researcher in the InP, is one of several founders of Xindium Inc. (Champaign, Ill.), a startup closely affiliated with the University of Illinois' world-class compound semiconductor research facility.

"We have spent a long time working on it, and we definitely believe that in a year or two indium phosphide will be the low-cost solution. At the 1.5-micron wavelength, we can field a monolithic device that integrates the PIN detector with the TIA (transimpedance amplifier). That kind of thing is where the advantage of indium phosphide lies," said Feng.

Xindium CEO Cindana Turkette, a veteran executive in the networking industry, said Xindium plans to demonstrate working 40-gig InP HBTs, including discrete pin detectors and integrated PIN-TIAs. "We'll be shipping devices for customer development in the next couple of months," she said.

In the low-frequency domain of 2.5-Gbit and 10-Gbit transmission systems, "separate devices may be OK, but in 40-G, I don't think you can use separate ICs. For cost reasons, a monolithic device is the real solution. When you have high-speed chips a lot of the RF issues can be managed within the chip. Separate chips have a lot of packaging and power problems," Turkette said.

Loi Nguyen, the CEO at Inphi, said many telecom system vendors are considering both SiGe- and InP-based components, but he believes power considerations will tip the balance in favor of indium phosphide.

"One of the major challenges in going to OC-768 is that every component burns a lot of power, and in our estimation SiGe burns three times as much power as indium phosphide. The system OEMs have a lot of problems with power consumption."

Nguyen said parts fabbed in SiGe-based BiCMOS also require as many as 35 masks, with a mask set costing as much as $700,000. "There are hidden costs to SiGe. Especially in the early days of low unit volumes, SiGe is a very costly solution. An indium phosphide mask set can cost $26,000, versus $700,000 for SiGe."

One strategy for keeping costs low is to partition the solution into two chip sets, the front end made on InP, the second set in bulk CMOS. Inphi and Broadcom Corp. have worked together to define partitioning issues, as well as the Q40 interface between the parts. The approach was presented to Optical Internet Working Forum, and has been published, so that "in principle" other companies could design parts to the Q40 interface, he said.

Nguyen said Inphi's philosophy is "not to use indium phosphide for any speed lower than 10 Gbit."

The TIA, modulator drivers, front-end mux and demux are in the sampling stage from Inphi, and Nguyen said the CMOS parts "will be available shortly."

Sheng, at Xindium, said his company has a similar InP-CMOS partitioning strategy, though he declined to name which company Xindium will partner with. "We have the same strategy. Anything that can be done in CMOS should be done in CMOS. The true advantage of indium phosphide is in the front end, so we'd like to integrate the parts that handle the input from the optical signal and then connect to CMOS parts" that handle the demultiplexed 2.5 -Gbit/s electronic signals. For OC-768, "ideally we'd like to have a single chip solution — input output signals at 40 G, and electrical signals at 10 G," he said.

David Caruth, a design engineer at Xindium, said that for 2.5- and 10-Gbit/s systems, a separate TIA and PIN "may be suitable to do in a hybrid (package) but at 40G the bond wires between the PIN and the TIA are too large to maintain performance — you end up with an inferior TIA."

Companies that can demonstrate bonded discrete components at 40 G "maybe can do it once, but they can't do it repeatably. And then if they try to do with multiple channels, it's going to be pretty tough. The people who do try to do it directly have a PIN with very high responsitivity — 0.8 to 1 amp per watt — out of the photodiode. Once you package it with a TIA the parasitics cause an effective response drop of a couple tenths of a watt. So for very short reach, some customers are content with slightly less responsitivity as long as the PIN is integrated with the TIA," said Caruth.

Xindium has developed an integrated TIA-PIN with a responsivity of 0.4 to 0.5 amps per Watt that is acceptable for short-reach applications that have sensitivity requirements of -7 dbm, he said.

Others in the industry plan to use wave guide photodectors that are edge coupled, which Caruth said require difficult packaging for the optical components. "Our detectors are on an integrated structure, which can be top or rear illuminated. That is good because that is the same approach used by the detectors in the 10 G and 2.5 G systems. By being compatible with the previous product, the customer's assembly costs and assembly times are much lower."

Sheng claims that Xindium derives its integration advantage partly from its InP process technology — developed at the III-V fab at the University of Illinois' Microelectronics Center. Xindium has a carbon-doped emitter process that differs from the beryllium-doped process used at several other companies. "Most foundries are set up to do a TIA, not an integrated PIN-TIA. We believe we have the ability to market an integrated PIN-TIA for the same cost as others can provide a TIA. The reason is that we have perfected an epitaxial structure that allows an integrated PIN-TIA — that is the Xindium advantage."

The secret sauce
Milano noted that most of the InP material structure, etch and metallization, and high-quality dielectrics are generic to the IC industry. "The secret sauce is to make it work with the particular chemistry of indium phosphide and its related alloys."

Vitesse is moving its GaAs production to a 6-inch line at Santa Rosa, Calif., and concentrating its 4-inch line — with a weekly capacity of 720 wafers — on indium phosphide production.

"The keys for patent and IP innovation is figuring out the simplest epilayer stack that will get you the performance and function that you want. That is why the materials are still the key thing: how many layers can you stack up, what compositions are they, how to process that particular stack. There are a lot of brute-force ways to do that, which involve materials regrowth and interrupting the standard IC process flow. What people are looking for is a simpler process, a single stack that can be grown once and processed using nonplanar techniques to achieve the ultimate function they are interested in," Milano said.

Vitesse and Velocium, a TRW group company, both operate InP fabs and offer foundry services as well. Several dedicated InP foundries, such as Global Communications Semiconductor (Torrance, Calif.), are working with the fabless startups, such as Inphi.

Dwight Streit, president of Velocium, said the push to higher integration levels is well underway at Velocium. "We have published circuits with 3,000 transistor counts, and we need several thousand transistors to do some of the A/D converters, frequency synthesizers, and HBTs. We've demonstrated 3,000 and get reasonable yields to those specifications. We have a new process under development that will get us to 10,000 transistors," Streit said.

Streit said Velocium also can leverage TRW's extensive photonics expertise, which includes lasers developed for the military.

"We believe a very good business opportunity is to integrate the PIN diode with the the TIA for 40 gig. It is less expensive and the performance is better. The reason it is cost effective is that all of the signal matching is on the same die, and the more parts you have to assemble the more costly the assembly process becomes. We are aiming at the metro market, because there is no volume right now in the long haul market."

Beyond the fiber optic-communications market, Streit said Velocium will use its indium phosphide expertise to develop power amplifiers for WCDMA handsets and monolithic millimeter ICs (MMIC) for other wireless systems.

Prototypes of the power amplifier parts will be ready by the end of the year, with one half of the noise figure of competing technologies. Volume production may come as soon as 2004. "The W-band CDMA margins are going to be very, very tough, and indium phosphide can meet those very hard specifications," he said.

Streif said his view is that today's electronics industry "is all cost driven. The market opportunity for OC-768 has been moving steadily to the right (i.e., farther out in time). That market is in dire straits, because the view today is that 10-Gbit systems are just fine, thank you. But we will have 40-Gbit systems, and whether they are accepted is all going to be cost driven."

Power consumption
After cost, power is the main concern of the telecom operators. An ongoing debate continues over whether silicon germanium, with its ability to add digital CMOS gates to bipolar HBTs, can meet the power-delay requirements of the front-end electronics for OC-768.

"For the equivalent speed, SiGe has four times higher power consumption than indium phosphide, and the dominant constraint at 50-70 GHz is power. For the same power, we can operate 1,000 gates in indium phosphide and 100 gates in SiGe," Sokolich said.

InP also has an advantage in low-power applications because the turn-on voltage is .75 V, compared with .95 V for silicon, he added.

At the recent Optical Fiber Conference (OFC), held in Anaheim last March, Infineon, IBM and other companies working on SiGe-based parts for OC-768 demonstrated their IC products. At that meeting, a technology manager from AMCC gained headlines by claiming that SiGe burns too much power for OC-768.

Nguyen, at Inphi, said that in an optical system "the laser requires a lot of cooling, so if the electronics also burn too many watts, it requires a far better heatsink than the OEMs are providing today to make sure the transceivers last the lifetime of 20 years or so. Some of these prototype systems have very large fans, and to cram as many cards as possible in the rack, the system companies do not want to change their entire form factor."

Milano said the argument is not so simple. As a company with CMOS, GaAs, SiGe and InP experience, he said Vitesse is "agnostic" about which process is best for a particular application. (It operates GaAs and InP fab lines, and uses a foundry for its SiGe products.)

"Choices will be driven by what early subsystem implementations look like. People focused on low-complexity transceiver functions might favor indium phosphide, rather than the higher-complexity transponder functions. Those considerations could drive your technology choice to one or the other material," Milano said.

The comparison is muddied by SiGe's progress to 180-nm design rules, close to the state of the art, while Vitesse, for example, currently uses 1-micron design rules for its current InP products, though it has a 0.5-micron process ready for production.

Milano said that if you look simply at the transistor figures of merit for InP, "you could argue that indium phosphide should lead to a low power circuit. The difficulty is that you can't treat that independently of the architecture. The old method of just looking at gates doesn't work, because where you can exploit parallelism you can really exploit power consumption. It is nice at a conference to argue about a specific device metric, but you have to look at a specific applicaton."

Nevertheless, Milano noted that in HBTs the power consumption scales directly with the emitter area, and parasitics improve with finer geometries as well. One argument is that since InP manufacturing is far from mature, it will catch up more quickly with SiGe, which is fairly close to the design rules used in leading-edge CMOS.

Streit argues that indium phosphide manufacturing will improve rapidly because it leverages many of the same techniques used by the GaAs industry.

Other opportunities?
With the telecom market so muddled by scandal and corporate debt now, will the InP startups be able to invest and survive?

Franzosa, of Inphi, notes that in a difficult funding environment his company received $24 million in new funding in June, following the initial round of $12 million in December 2000.

"OC-768 is definitely not happening next year. There will be trial systems, and we do get some revenue from that, but overall it is definitely delayed. Other system opportunities exist, such as test and measurement, wireless and satellite communications, where you have the combination of high performance and low power," Franzosa said.

Though Streif, at Velocium, counts W-band wireless power amplifiers as a major opportunity, Franzosa — a former Motorola wireless engineer — said he has his doubts.

"We are always looking at new markets that could take advantage of performance, and potentially the cost advantage, of InP. But in my opinion, for wireless power amplifiers. . .in terms of utilizing InP versus SiGe or GaAs, its tough, because you have a cost issue. And you have to have a significant performance advantage to offset that. Even if you have one, if you look at it from the macroscopic view, its not enough. So when you start looking at the performance advantage of InP, you have to back this out and look at the whole system."

Cornell professsor Eastman agreed. From his vantage point as an advisor to several InP startups, Eastman said he believes "it would not be wise to push indium phosphide into the wirless space" for power amplifiers. "Silicon is eating the lunch of the III-V people in the wireless space," Eastman said.

Where indium phosphide wins out over SiGe, Eastman said, is in applications, such as amplifiers operating in the 100 GHz range. "Silicon germanium may run out of steam, because in many of these applications you don't just require speed, you need volts, and you can never get enough volts out of SiGe."

Sokolich said HRL Labs is investigating the use of InP for the voltage control oscillator for automotive collision avoidance radar operating in the 77-GHz range, a potentially new and large market.

But Sokolich acknowledges that OC-768 is the best opportunity for the InP companies, and argues that "that is a market that is small and will stay small, so it is best suited for the startups."

Despite today's calamitous telecom industry situation, real opportunities exist, according to Strategies Unlimited, a market research firm based in Mountain View, Calif., which recently issued a report on the market prospects for indium phosphide.

The report, based on an expectation that the communications device market will produce revenues of $2.6 billion in the next four years, estimates that InP devices, including advanced laser diodes and detector ICs, will garner revenues exceeding $500 million in 2006, with the potential to reach several billion dollars in annual revenues within the next 10 years.

With delays expectable ahead for both 3G wireless — where Velocium and perhaps others see great promise — and for OC-768, which companies will survive and prosper?

Perhaps the business challenge is best summed up by Nguyen, who studied under Eastman at Cornell and then spent a dozen years at HRL Labs before founding Inphi: "Clearly the market environment is difficult. There is no question about that.

"For OC-768, no one is in mass production anyway, so this is the development phase, with volumes really hitting in '04. This is a shakeout period, and only the best company, whether it is a component company or a box company, only the best will be able to raise money and remain viable."