Although fiber-optic links at telecom wavelengths rely on InP-based optoelectronic devices such as lasers and photodetectors, electronic ICs made with InP have yet to reach the commercial mainstream. However, as network equipment makers start to look more seriously at 40 Gbit/s systems, InP-based electronic and optoelectronic ICs are poised to make the crucial breakthrough.

Companies such as TRW and HRL have a long history of developing InP ICs for military and space applications. Now these companies (TRW through its Velocium subsidiary) are addressing commercial markets. They have been joined by firms such as Vitesse and Alpha in the US, NTT Electronics (NEL) and Hitachi in Japan, and by foundries such as Global Communication Semiconductors (GCS).

The key factor is performance. In a paper at this year's GaAs Mantech conference, Mike Sun of Alpha Industries' Sunnyvale Operations summarized the performance advantages of InP HBTs. "The most attractive features of InP DHBTs are low power consumption due to their low turn-on voltage, and a high Johnson limit due to the high breakdown of the InP collector," he says. "Also they allow moderate circuit integration levels up to a few thousand transistors, and offer the potential to integrate optoelectronic devices such as photodetectors."

So why is this happening now? A key factor is the growing demand for high-speed devices, which has led IC manufacturers to invest in the commercialization of their technology. The article by Velocium in this issue (see "The path to InP production") describes some of the crucial steps in transferring technology from the lab to the fab. In many cases it has proved possible to leverage the process steps and equipment used in GaAs IC manufacturing, but the availability of materials has been a different matter. A key factor in the birth of the InP IC industry has been the recent availability of 4 inch InP wafers with suitable properties (see "Substrate makers tackle the growing challenges of InP"). For companies lacking internal epitaxy capabilities, the availability of 4 inch InP epiwafers has also allowed them to establish their own processes.

Velocium

In May 2001 TRW formed a commercial subsidiary, Velocium, to manufacture and sell high-speed GaAs- and InP-based ICs for fiber-optic and wireless communication systems. The company built the world's first 4 inch InP line, and has now introduced a range of InP-based ICs, including a 40 Gbit/s photoreceiver, a 12.7 Gbit/s NRZ-to-RZ converter and a divide-by-2 frequency divider targeted at down-conversion or phase-locked loop applications. The latter product (DV2401) is rated at 27-43 GHz, although performance extends up to 50 GHz. Features include a dynamic divider design, a single -4 V power supply, on-chip termination (50 Ω to ground), single-ended input and differential output, and a die size of 1.1 x 2.0 mm2.

HRL Laboratories

One indication of the level of interest in InP technology at the recent Optical Fiber Communication (OFC) conference was the appearance of HRL Laboratories of Malibu, CA. HRL's commercial activity in the fiber-optic market is complementary to the work being carried out on behalf of its owners - Boeing, General Motors and Raytheon - in the automotive, defense, commercial airframe and space businesses. According to Marko Sokolich, research department manager of IC Process Engineering at HRL, the cumulative demand for ultra-high-performance circuits in these five broad business areas is modest but important, and all the areas benefit from InP. "By serving these markets with our flexible, high-performance InP processes we are able to enhance yield, reproducibility and reliability in an efficient manner," he says. "In this way we are able to reduce all of our customers' costs to access this technology."

Sokolich says that HRL exhibited at OFC in order to showcase its InP foundry and design service capability. "We were particularly interested in educating the fiber-optics market about HRL's history of InP development, dating back to 1988," he says. "Our InP technologies have demonstrated a number of benchmark high-performance millimeter-wave and mixed-signal circuits over the past decade."

Recently, however, HRL has been engaged in the design and demonstration of high-speed digital circuits for 40 Gbit/s and higher fiber-optic communication applications. "Circuits built in our technology have gone into several OC-768 transceiver prototypes for various customers," says Sokolich. "In 2001 we shipped the wafer equivalent of 10,000-20,000 ICs for 40 Gbit/s systems."

HRL offers a "modest volume" fabrication capability in state-of-the-art InP SHBT, DHBT and HEMT processes, as well as design services for fiber-optic communication building-block circuits. The company has developed a first-generation (G1) InP SHBT process with ft and fmax values of 80 and 120 GHz, respectively, and a G2 InP SHBT process with ft and fmax values of >150 and >180 GHz, respectively (see Compound Semiconductor October 2001, p51). During 2002, HRL plans to release a G2+ InP SHBT process with ft and fmax values of >190 and 250 GHz, respectively, as well as a G2+ InP DHBT process with ft and fmax values of 150 and 300 GHz, respectively, and a breakdown voltage (BVceo) of >9 V.

However, Sokolich believes that process parameters do not tell the whole story. "We believe that discrete transistor metrics are necessary but not sufficient for an IC technology," he says. "We have also demonstrated a number of important circuit-based metrics." For example, HRL has fabricated a static divide-by-2 IC operating at >65 GHz, and demonstrated a 60 GHz packaged version in its booth at OFC. Divide-by-4 ICs operating at >65 GHz and divide-by-8 ICs operating at >70 GHz have also been fabricated.

Other ICs include: a 2:1 mux and a 1:2 demux, both operating at >50 Gbit/s; a 40 Gbit/s modulator driver with Vdiff = 8 Vpp and Vse = 4 Vp; and 4:1 mux and 1:4 demux ICs operating at 40 Gbit/s.

Vitesse converts 4 inch fab to InP

Vitesse was one of the first manufacturers to announce its intention to use InP ICs for 40 Gbit/s applications (see Compound Semiconductor April 2000, p7). Vitesse's fab in Camarillo, CA, which originally manufactured GaAs MESFET products, has been converted to a 4 inch InP wafer fab. The company has developed a series of products and is also offering foundry services using its first-generation (VIP1) single-HBT process. According to Alan Huelsman, director of Vitesse's InP program, the process uses a mesa-isolated npn structure that leaves room for further advances through device scaling. "Implementation of this transistor design allowed us to bring up a manufacturable process with adequate performance in a very short period of time," he says. "Fab yield has been excellent and typically runs at about 85%. Circuits with less than 100 transistors often show 100% yield, and the largest circuit built to date in this process - a 4:1 mux containing 2500 transistors - shows very good yield."

The MBE-grown epilayer structure used by Vitesse includes an InP emitter and a Be-doped base, and aluminum interconnect technology was adopted from Vitesse's GaAs process. The transistor performance peaks at a collector current density of about 1 mA/µm2, and the ft and fmax values at this density are both >150 GHz. The BVceo is 4.2 V for the SHBT process, and a DHBT process with a BVceo of >7 V, targeted at lithium niobate modulator drivers, is under development.

"The VIP1 process was frozen in August 2001 to allow us to work with foundry customers," says Huelsman. "However, we have a process development team of about 12 engineers who are working on a second version of the process. By moving to smaller dimensions and making other improvements we expect to reach ft and fmax values exceeding 200 GHz."

Huelsman says that Vitesse has built all the key building blocks for the OC-768 physical layer, including a transimpedance amplifier (TIA), a limiting amplifier, a 4:1 mux/ demux pair and a driver with 3.5 Vpp voltage swing. The analog parts are all being sampled to customers.

Applications at 10 Gbit/s

For most 10 Gbit/s applications the extra performance provided by InP is not required, and so InP is uneconomical when compared with GaAs, SiGe or CMOS. However, this is not always the case; Vitesse has already introduced InP parts at 10 Gbit/s, including an RZ modulator driver and an integrated pin-TIA for 10 Gigabit Ethernet. "An RZ driver requires a bandwidth of about 30 GHz, and InP is required to achieve this level of performance," says Huelsman. "On the receive side the demand is for integration; the pin-TIA replaces both a GaAs or SiGe amplifier and an InP detector."

At OFC Velocium introduced a 12.7 Gbit/s NRZ-to-RZ converter that electronically converts NRZ signals into RZ format. Standard RZ systems use two modulators, one to create a train of pulses and the other to apply the signal; Velocium's converter eliminates the need for the pulse train. The converter operates at up to 12.7 Gbit/s, providing headroom for forward error correction, and has a power consumption of less than 1 W. "By utilizing an advanced InP process for our NRZ-to-RZ converter, we can meet the higher bandwidth performance required for RZ modulation at lower power, compared with alternative solutions," says Velocium's Frank Kropschot. "From a system perspective, the converter eliminates the need for an additional optical modulator, therefore reducing overall system cost and complexity."

Optoelectronic integration

In addition to the high performance of InP-based electronic ICs, another driving force for introducing these devices is the prospect of integration with optical functions. Starting with pin diodes and TIAs, this could lead to the integration of arrayed waveguides (AWGs). Alan Huelsman says that Vitesse is already working with foundry customers to develop both active and passive InP-based optical products.

In fact InP integration is by no means a new phenomenon. Opto Speed, an optoelectronic component manufacturer based in Switzerland, first introduced monolithic InP receivers in 1999. These consist of an InGaAs pin photodiode monolithically integrated with an InP HBT TIA. At the recent OFC meeting, Opto Speed announced a new family of optical front-ends for RZ and NRZ applications at bit-rates exceeding 40 Gbit/s, which are based on these integrated devices. The modules, measuring 15 x 15 x 4 mm3, have a responsivity of 0.45 A/W, a sensitivity of -9 dBm, a -3 dB bandwidth of 35 GHz and a power dissipation of 800 mW.

Similarly Velocium has introduced the PRX401, a 40 Gbit/s photoreceiver designed for use in VSR SONET, SDH, datacom transponders, digital video, avionics and test instrumentation. The 34 GHz bandwidth product contains an InGaAs pin photodiode monolithically integrated with an InP HBT TIA, and has a 70 V/W conversion gain. The power dissipation is 90 mW, and the device is available as a bare die or a connectorized module measuring 9.4 x 22.9 x 8.9 mm3.

IPAG teams with Multilink

Another company developing optoelectronic components and ICs based on InP is Innovative Processing AG (IPAG), based in Duisburg, Germany. IPAG has extensive in-house capabilities, including MBE and MOVPE growth, and has developed both HBT and HEMT InP processes. Along with pin diodes and electroabsorption modulators, IPAG has developed a 43 Gbit/s pin diode monolithically integ-rated with a traveling wave amplifier (TWA). IPAG's front-end receiver module containing the pin-TWA circuit has a -3 dB bandwidth of 45 GHz, an optical input power of +6 dBm and a conversion gain of 50 V/W at 1550 nm.

IPAG recently entered into a collaboration with Multilink Technology, a fabless supplier of GaAs digital and mixed-signal ICs based in Somerset, NJ. The companies will work closely to develop 40 Gbit/s pin diode receivers, TWAs and monolithically integrated devices. "IPAG focuses on InP as the most attractive candidate for the monolithic integration of high-speed electronic and photonic devices," says IPAG's CEO Ralf Bertenburg. "With our extensive knowledge and access to a complete InP processing line, we are able to establish a standardized InP process technology with high yield."

A minority equity investment in IPAG will grant Multilink early access to critical optical and optoelectronic devices. Multilink will work closely with IPAG on system and component requirements and packaging, and will also act as the exclusive channel for IPAG's line of integrated optoelectronic receivers.

Japanese companies

Not surprisingly several Japanese companies are developing InP devices for 40 Gbit/s systems. At OFC, NEL demonstrated ICs based on a 0.1 µm InP HEMT process with ft and fmax values of 200 and 350 GHz, respectively. The company unveiled a 50 Gbit/s 4:1 mux module incorporating a D-type flip-flop and a clock distribution unit, and also exhibited a 2:1 selector, a 1:2 demux and a toggle-FF. OpNext, the spin-off from Hitachi, has also developed 40 Gbit/s parts, including an InP HBT modulator driver, an InP post-amplifier and a 40 GHz clock recovery circuit, which operates in conjunction with SiGe 4:1 mux and 1:4 demux ICs. In June 2001 OpNext announced a joint development program with Velocium for 40 Gbit/s GaAs and InP ICs.

GCS backs double HBTs

Foundry company GCS has developed 4 inch InP SHBT and DHBT processes that take advantage of various process modules (photolithography, thin film, etch etc) used for the company's InGaP/GaAs HBTs. The InP process uses a device structure with an InP emitter and a C-doped base, and incorporates a high-yield, low-loss planar interconnect technology. A device with a 1 x 3 µm2 emitter operating at a current density of 100 kA/ cm2, had ft and fmax values of >160 GHz and a BVceo of >3.5 V. At a higher current density of 160 kA/ cm2, an ft value of 200 GHz was extrapolated.

Reliability is vital to commercial digital communications applications. According to a paper at the GaAs Mantech conference, the reliability issues of InP HBT technology remain unresolved (Nguyen et al.). GCS has addressed reliability in a fundamental manner. In the device structure, the use of an Al-free emitter (InP rather than InAlAs) and a C-doped base (as opposed to using Be) are claimed to enhance reliability, while a non-alloyed ohmic metal contact scheme enhances the robustness of the devices under high current drive. A proprietary passivation technology has also been developed to ensure surface cleanliness.

One of GCS's customers is Inphi, a fabless company developing high-speed InP ICS (see Compound Semiconductor April 2002, p28). Another fabless design house, GTRAN, has developed its own process for InP ICs and is using external foundries, including GCS.

Xindium and UIUC

Xindium Technologies, a year-old spin-off from the University of Illinois at Urbana-Champaign (UIUC), is also developing commercial InP-based ICs. The company will undertake pilot production and seek to license its processes to commercial foundries (see "Xindium uses InP expertise to make commercial 40 Gbit/s ICs"). At this year's GaAs Mantech conference, a paper from Xindium and UIUC described an InGaAs/InP SHBT process with ft and fmax values of 160 and 252 GHz, respectively, for a 2 x 5 µm2 device (Shen et al.). These values were recorded for C-doped SHBTs grown by GSMBE on 2 inch wafers, which had a base doping level of 4 x 1019/cm3 and a collector thickness of 3000 Å. GSMBE is capable of higher base doping levels due to hydrogen passivation in MOVPE. In integrated pin/ SHBT structures, the reduction in base resistance is crucial for higher RF performance.

Alpha Industries

In fall 2001 Alpha Industries announced plans to develop a commercial InP HBT process (see Compound Semiconductor September 2001, p6). The company has implemented an InP/ InGaAs/InP double HBT structure using commercially available epitaxial material grown on 4 inch substrates. The process leverages Alpha's InGaP HBT technology and uses many of the same back-end steps, such as passive NiCr resistors, MIM capacitors, spiral inductors and metal interconnections.

As these steps are often yield-limiting for GaAs HBTs, Alpha expects the InP HBT technology to show similar circuit yields and integration levels to those of the GaAs process. The InP DHBT process is non-self-aligned and achieves critical dimension control between the emitter and base contacts. A typical 1.2 x 8 µm2 device with a 200 nm n-InP collector showed ft and fmax values of 160 and 190 GHz, respectively, and a BVceo of 7 V (Sun et al.).

An indium phosphide future

Opinions vary as to which 40 Gbit/s ICs will be made with InP, while companies such as IBM, AMCC and Infineon argue that SiGe can handle most functions for 40 Gbit/s networks. At OFC, AMCC introduced SiGe mux and demux ICs to complement its 40 Gbit/s TIA and modulator driver. Inphi, which has developed 4:1 mux and 1:4 demux ICs in InP, is using a GaAs PHEMT foundry process to manufacture its modulator driver ICs until suitable InP DHBT processes are available. Similarly, Velocium unveiled a 40 Gbit/s modulator driver IC at OFC, which was manufactured using a GaAs PHEMT process. The 50 GHz bandwidth driver gives 8 Vp-p output voltage at 43 Gbit/s.

When 10 Gbit/s components were first introduced, GaAs had little competition. Now, advanced SiGe and GaAs technologies are already challenging InP as 40 Gbit/s applications gather momentum. However, it does seem certain that the confluence of market demand, the availability of 4 inch substrates and epiwafers, and the activity of a number of major IC manufacturers, will push InP into the mainstream at last.

Further reading

N Nguyen et al. 2002 Proc. GaAs Mantech conf. (San Diego, CA) 26.
S-C Shen et al. 2002 ibid. 22.
M Sun et al. 2002 ibid. 30

All from http://compoundsemiconductor.net