Fiber-optic-component integration advances

Hybrid approaches have more near-term promise than monolithic strategies, but each portends increased miniaturization.

BY JEFF D. MONTGOMERY

Over the past quarter-century, active fiber-optic components have advanced steadily in miniaturization (more gigabits transported per cubic inch), while costs correspondingly have fallen. The integration of parts within a component and the integration of single-channel components into multichannel modules have been major factors enabling these trends.

The trend toward increased miniaturization is colliding with package design challenges and the movement toward automated assembly and testing of fiber-optic components. Between 1987 and 1996, the conventional butterfly package was considered a compact option for housing an edge-emitting laser diode (particularly 155-Mbit/sec to 2.5-Gbit/sec versions), plus its monitor photodiode, manually assembled. The 1997-2000 period brought major investment by leading component producers such as Agere, JDS Uniphase, and Alcatel in semi-automated assembly of these units, along with incorporating driver ASICs and moving up to 10-Gbit/sec diodes. That, however, is still a couple of orders of magnitude larger in size and cost per gigabit than the near-term goal of data communications multifiber/multichannel transmitter designers. Packing numerous transmitters into a small package introduces serious problems of optical crosstalk and electrical crosstalk between channels as well as major heat dissipation concerns. But progress is being made.

Cost, performance, and miniaturization are the key drivers of the evolution from discrete circuits to hybrid to monolithic optoelectronic ICs (OEICs). For a production rate of 10 channels per day, discrete may be the lowest-cost solution; for 1,000 per day, it may be hybrid; for 10,000 per day, monolithic. A few transmitter and receiver units are now in the 1,000-per-day category and will hit 10,000 per day before 2010. The pressure for fiber-optic-component miniaturization comes from two main drivers:

The need to install succeeding generations of equipment-providing orders of magnitude greater throughput-within the same floor space allocated to earlier equipment.
The need to make optical-signal transport more cost-competitive with copper transport.

Active-component integration is evolving at two levels: hybrid optoelectronic ICs (HOEICs) and monolithic ICs (MOEICs). Numerous development programs have proceeded at both levels over the past 15 years.

HOEICs have pulled far ahead in volume production applications. They combine die of various functions on a substrate. The substrate typically is ceramic or silicon, with conductive patterns formed by electroplating, deposition, or other techniques. Active die typically are emitters, photodetectors, or ASICs. These die, with the substrate electronic-signal and power interconnects, constitute transmitters and/or receivers.

MOEICs, in contrast, use substrates of semiconductor material such as silicon, gallium arsenide, or indium phosphide, and the active functions are processed directly into the substrate. That eliminates the time-consuming placement and attachment of individual die and achieves greater density. Hundreds of VCSEL transmitters might be processed on a single wafer.

The challenge of MOEICs is that different functions of the transmitter or receiver often achieve optimum performance and/or cost when fabricated in different materials and by different processes. Apart from production-cost considerations, an argument for MOEICs is that the concept permits shorter electronic interconnects-thus, lower parasitic reactance, along with being able to obtain higher-dimensional precision of line or matrix arrays.

With the progression of gigabit VCSELs to high-volume production and acceptable yield (less than five defects per wafer), a compromise solution has emerged: dicing emitter, photodetector, driver, and amplifier ASIC arrays from their respective wafers and assembling them hybrid fashion onto interconnect substrates to achieve greatly (per-channel) miniaturized, potentially lowest-cost transmit and receiver modules. Xanoptix chains commercial shipment of its 36-channel transmitter/receiver module, and TeraConnect is shipping evaluation units of its 48-channel module. These and other VCSEL module vendors expect 256-channel evaluation units to become available in 2003 and first commercial shipments in 2004.

Monolithic integration, meanwhile, is advancing, without fanfare, in devices of relatively compatible materials and processes. Photodiodes and their amplifier(s) in III-V compound substrates have been available, with an increasing range of performance alternatives, from Vitesse Semiconductor, Tri-Quint, Anadigics, and other vendors. Substantial R&D is now proceeding on integration of driver ASICs with VCSELs, particularly for 10 Gbits/sec and higher, where III-V materials offer performance advantages.

The ultimate in monolithic integration is the integration of passive optical functions such as couplers, filters, and photonic switches on the substrate with active devices. Unfortunately, such an approach would be prohibitively expensive for commercial use in the foreseeable future. It is probable, however, that this concept will phase into selected military/aerospace optoelectronics within the next 15 years.

The term "component integration" has been extended in the optoelectronics industry beyond the original concept that was based on solid-state elements. As production volumes have increased, economic advantages as well as miniaturization have supported progress in combining several passive components into a single "supercomponent." E-TEK Dynamics, acquired by JDS Uniphase in 2000, was a leader in such integration. A more recent example is Lightwave Microsystems' integrated planar optical add/drop multiplexer now being evaluated by several OEMs.

The integration of active and passive fiber-optic components, if compared to the semiconductor IC industry, is still in its infancy. Impressive progress, however, has been achieved over the past decade. Economics, competition, and customer pressure for miniaturization will drive major advancement in integration over the next decade.