High-speed fiberoptic links are still largely based on discrete, expensive, power-hungry components and subsystems. Meaningful improvements in cost, power dissipation, and size, without performance penalties, are essential for fueling the growth of optical networks in the metro and access market segments. At the subsystem level, integration of multiple functions, both optical and electronic, into low-cost, small, low-power, and multiprotocol transceivers is being aggressively pursued.
For optical-component suppliers the push has been twofold: to lower the cost of existing optical solutions, and to provide further cost savings by integrating multiple optical functions via hybrid or monolithic integration while lowering size and power dissipation. While a steady decline in prices of discrete optical components is evident, optical integration has not yet delivered on its promise of dramatically reducing the cost of optical links.
Conventional photonic integration techniques that offer monolithic integration of indium phosphide (InP)-based active and passive components rely on complex epitaxial growth processes to create different optical materials and thereby different optical functions on the same substrate. The specialized growth processes result in low yields, and sequential etching and regrowth steps for each additional optical function rapidly undercut the cost advantage of monolithic integration. The low yields and other performance-reducing design tradeoffs have resulted in monolithic integrated active-components being available only in limited applications.
To integrate different optical functions without the need for custom epitaxy, multiple optical materials—each optimized for a particular function—can be grown in separate vertically positioned waveguides in a single high-yielding epitaxial step. The different optical components on the same chip are then fabricated using standard processing techniques similar to those used in commercial III-V electronics foundries.
Light is transported efficiently between the different optical waveguides by lateral tapers that serve as optical vias guiding light from one layer to the other.1 With a higher-yield process, it is possible to integrate a greater variety of optical functions such as lasers, detectors, waveguides, and amplifiers on a single chip. An example of an implementation of this platform is an uncooled distributed Bragg reflector (DBR) EMLs with a gain section, grating, and modulator in each of three layers (see Fig. 1).The integration platform allows independent optimization of the data-modulation and power-generation functions in electro-absorption modulated lasers (EMLs). As a result, uncooled 1310-nm distributed feedback (DFB) EMLs developed on this platform outperform the discrete DFB directly modulated lasers (DMLs) in terms of eye quality over the entire operating temperature range (0°C to 85°C).
A key distinguishing feature of uncooled EMLs compared to DMLs is that EMLs maintain high eye margins (percentage margin over eye-mask specifications) with low jitter and high extinction ratios, even at high temperatures (>9-dB at 85°C, uncooled). This is achieved by optimizing the modulator for high speed, high extinction, and low modulator drive voltages while the single-frequency laser is optimized for output power, temperature stability, reliability, and spectral quality.
The RF characteristics of EMLs are well understood. The modulator is a reverse-biased diode with low capacitance, and the impedance presented to the driver is well characterized and stable over operating temperature and lifetime of the device. In contrast, DMLs typically feature a short laser (necessary to reduce capacitance) that is driven with a high current density to obtain operation at 10-Gbit/s.
The impedance presented by a DML to the driver varies with temperature, and the dynamics of the DML may change over its lifetime. Precise matching of the driver and DML is often required to achieve adequate performance.
DMLs usually exhibit one of two issues when operated over a wide temperature range. At the rising edge excessive overshoots reduce margins even for a filtered eye. A more severe limitation is a decrease in bandwidth of the laser that reduces the mask margin due to a slower falling edge. Often a complex balance is struck between these two limitations by tweaking laser bias current, RF modulation current, and pulse width duty cycle, all of which have to be tracked over temperature. For uncooled EMLs, however, the RF drive can be kept constant at 2.5 V peak-to-peak (Vp-p) if desired and only the DC inputs—the laser bias current and modulator bias voltage—need to be changed as a linear function of temperature.
The back-to-back eye pattern obtained from the uncooled EML operating at 85°C shows a 53% margin for the 10 Gigabit Ethernet mask and a 36% margin for the OC-92 SONET mask (see Fig. 2). In the unfiltered eyes, the rise and fall times of the uncooled EML are <25 ps even at 85°C, highlighting the high-bandwidth capability of EMLs. The margins for the filtered eye (which simulates the effect of limited receiver bandwidth, thus reducing SONET margin) after 10 km of transmission in SMF-28 fiber are 51% and 26% for 10 Gigabit Ethernet and SONET, respectively.The extinction ratio (ER) can be maintained above the required GR-253 OC-192 SR-1 specification (6 dB at 1310-nm) over the entire operating temperature range with a 2.5-Vp-p RF drive. The ER for an uncooled EML improves with increasing temperature for a constant drive voltage, whereas for an uncooled DML it typically drops with increasing temperature for a constant RF drive current. Because ER for an uncooled EML at 85°C is close to 10 dB for 2.5 Vp-p (and ~8 dB for 2.0 Vp-p), reducing the RF drive at high temperatures will lower power consumption in the driver. The average output power is greater than -3 dBm in the fiber.
For SONET applications, the transmitter eye margins achieved with EMLs are typically more than twice those obtained with DMLs (see Fig. 3). The resulting transmitter yields with the use of uncooled EMLs can increase by up to 50% over yields obtained with uncooled DMLs, because system vendors typically require at least 10% eye margin to ensure link reliability over service life. Hence the additional margin simplifies transceiver manufacturing, and leads to lower transceiver cost with materials, time, and overhead savings.The drive electronics, monitoring requirements, and other operational details for EMLs are different from those for DMLs. Two current sources, one for quiescent (DC) bias and the other for RF data input, are used to power DMLs. The amplitudes of both inputs, as well as the duty cycle of the input data pattern are typically adjusted for optimum performance over temperature range, using setpoints obtained from DC and RF characterization along with feedback from a back-facet monitor photodiode.
In the case of EML operation, two voltage inputs and one current source are needed. The modulator is reverse-biased with a DC voltage using setpoints obtained from DC and RF characterization, and an RF voltage swing of constant amplitude is applied with a polarity that is the inverse of the data. The setpoints for modulator DC bias and modulator photocurrent over the entire operating temperature range can be derived by a linear fit based on measurements at only two temperatures, say, 10°C and 85°C.
The average power from the EML can be maintained nearly constant (within 1 dB) in this operational mode (see Fig. 4). The two-temperature calibration helps reduce final test and setup time for uncooled EML-based transmitters significantly. In addition, as the laser current is controlled using the modulator photocurrent for feedback, the design engineer can reduce component count by eliminating the back-facet monitor photodiode.Swami Srinivasan is product manager and Milind Gokhale is chief technology officer at ASIP, 25 Worlds Fair Dr., Somerset, NJ 08873. Swami Srinivasan can be reached at [email protected].
P. V. Studenkov et al., IEEE Phot.Tech. Lett., 13(6) 600 (June 2001).