While some manufacturers have identified indium phosphide as the material of choice for future OC-768 parts, AMCC has already demonstrated silicon germanium products for this market. Bob Metzger and Tim Whitaker report.

AMCC eyes SiGe for 40Gbit/s parts Although the optical networking industry is currently migrating from 2.5 to 10 Gbit/s, device manufacturers are already beginning to develop the components that will be needed for the OC-768 (40 Gbit/s) systems that are expected to be deployed in the next 23 years. While many have conceded that Si-bipolar and CMOS can handle 2.5 Gbit/s applications, there is no clear technology winner for 10 Gbit/s devices. Early 10 Gbit/s parts were implemented using GaAs, while SiGe HBT and even CMOS parts are now available at this data rate. When it comes to 40 Gbit/s devices, many assume that the technologies of choice will be HEMTs or HBTs based on either GaAs or InP. Several manufacturers have already announced plans to implement 40 Gbit/s components in InP. These include such compound semiconductor heavyweights as Vitesse, RF Micro Devices and the newly formed TRW spin-off Velocium (see page 7). The indication is that the intrinsic high speed of InP devices, coupled with the ability to integrate ICs with optical devices (e.g. telecom lasers and detectors) on the same wafer, make this material the best choice for 40 Gbit/s components. Additionally, many people within the compound semiconductor community believe that silicon simply will not be able to perform at 40 Gbit/s. Applied Micro Circuits Corporation (AMCC) does not share this belief, and is backing SiGe for 40 Gbit/s applications in the most emphatic way possible by introducing commercial 40 Gbit/s parts. "Our philosophy is that if you can do it in silicon, you should," says Ken Prentiss, AMCC s director of marketing for telecom products, who cites the benefits associated with high-volume manufacturing in the silicon industry and many years of investment. Prentiss also mentions AMCC s experience with SiGe as an important factor in choosing this technology for next generation parts. "We have eight 10 Gbit/s parts in SiGe that have been shipping for a year or more," he says. "Although we don t ignore other technologies [such as GaAs or InP], we haven t found a reason to use processes other than those based on silicon in the markets we address. Our customers care more about whether you meet their price, performance and delivery targets than what technology is used, and we believe that SiGe is the optimum technology to meet these needs." Why not use indium phosphide? Like many other manufacturers considering the 40 Gbit/s market, AMCC evaluated the possibility of using InP. "One major reason that we rejected InP was the need to introduce a high level of integration in our serdes (multiplexer and demultiplexer) ICs we need many thousands of gates," says Prentiss. As manufacturers such as Velocium and Vitesse begin to commercialize InP-based electronics and move up the manufacturing curve, the ability to increase transistor counts and enhance yields will undoubtedly improve. Today, however, the ability to achieve significant yield at the 10 000 transistor level does not exist for InP-based devices. While AMCC also appears to be skeptical about issues such as the manufacturability of InP devices, and the availability of InP wafers, other companies are equally concerned as to whether SiGe HBTs have the performance capability to be used in the implementation of 40 Gbit/s circuits. Proof of 40 Gbit/s performance Rather than claiming what should be possible, AMCC has answered the question of SiGe s capabilities by using the technology to produce 40 Gbit/s circuits. shows a schematic of an OC-768 (40 Gbit/s) electronics portion of an optical network that AMCC hopes to implement using a more advanced version of IBM s commercially-available SiGe process (see sidebar). The first part of this system to be implemented in SiGe is the transimpedance amplifier (TIA part #S76800) which provides high-speed amplification of current received by an external photodetector in the receiver frontend. The TIA can successfully handle data rates up to 48 Gbit/s and accept return-to-zero (RZ) and non-return-to-zero (NRZ) data streams for enhanced performance. The 45 GHz bandwidth of the TIA allows the implementation of OC-768 operation without having to resort to using a PIN-diode to optically amplify the incoming data signal. The TIA features a transimpedance gain of 220 ohms, and with a low input noise of 4 A, this allows signals as low as 50 A to be detected with a BER of only 1E-10. Using a 5.2 V power supply, the TIA dissipates 0.6 W. In addition, AMCC has also just announced a 40 Gbit/s MUX, another key element in the implementation of the system detailed in . The 40 Gbit/s eye-diagram is shown in . It is anticipated that remaining 40 Gbit/s building blocks will be available by the end of the year. Performance trade-offs In light of the parts produced by AMCC, it is clearly false to claim that SiGe cannot be used to make 40 Gbit/s circuits. However, there are certain inherent limitations to a silicon-based bipolar process which make this a significant challenge. When considering a high-speed bipolar process, the issue of breakdown is of primary concern. Speed in these devices is improved by decreasing the electron transit time, which is accomplished primarily by thinning the base and increasing the doping levels at the collectorbase junction. As a consequence, there is an accompanying reduction in breakdown voltage. While the supply range for the 40 Gbit/s TIA is 8.5 V (VCC = 3.3 V and VEE = 5.2 V), AMCC is able to utilize IBM s advanced SiGe process even though its BVCEO is less than the 8.5 V supply range. The reason that the low breakdown SiGe process can be used for this application is that BVCEO is a measure of the collector-to-emitter breakdown when the base electrode is open-circuited. However, the collector breakdown can be increased when the base looks into any finite impedance. By realizing this limitation, AMCC circuit designers optimize the impedance that the base sees in order to enhance the device breakdown. In an InP-based process (for a double heterostructure device in which InP is used in both the emitter and collector) a BVCEO of 812 V is possible, therefore eliminating the need for any special impedance optimizing circuitry. In addition, InP-based HBTs operate at a significantly lower turn-on voltage (0.7 V as compared to 1.21.4 V) and at lower current densities, therefore consuming less power than SiGe parts. While AMCC considers the use of SiGe to be a critical element in their deployment of OC-768 circuitry, it is not sufficient in and of itself to meet the performance criteria. There are several additional factors that allow them to reach the 40 Gbit/s milestone. For example, the availability of low resistance interconnects is extremely important. The AMCC SiGe chips contain up to seven layers of metallization, five of which are made from copper. This results in lower sheet resistance, and also reduces on-chip voltage drops and raises parasitic pole frequencies. In addition, due to the use of a dielectric trench for isolation and multiple layers of dielectrics to isolate the various metal layers, the available dielectric thickness is several times greater than in most VLSI processes. This can be utilized to greatly reduce parasitic capacitances, which in turn improves speed performance. The performance of inductors is also improved by increasing Q. In addition, because IBM uses a BiCMOS process, the multiple device types on a single substrate allow greater flexibility in design. It is these elements - when added together with custom design approaches - in which the designers know how to optimize the circuitry to enhance SiGe transistor performance, that has allowed AMCC to enter into the 40 Gbit/s marketplace using SiGe. No one disputes the fact that InP-based devices have the fundamental characteristics to be used in the fabrication of 40 Gbit/s devices. With the circuits developed by AMCC, it is now clear that no one can dispute that SiGe is also capable of producing 40 Gbit/s devices. So what will be the material/device of choice? It is difficult to tell at this point, but it might be best to remember that a wide range of materials and device structures are currently being used to manufacture 10 Gbit/s components. With some manufacturers backing InP, while others such as AMCC are committed to SiGe, it seems likely that, at least at the outset, a diverse range of technologies will be used to manufacture 40 Gbit/s circuitry.