To optimize performance or accommodate legacy platforms, system designers must make trade-offs between the passive and active transmission system components.
By Bill Keller
As 10 Gbps copper transmission becomes a reality, a number of electrical transmission protocols are emerging towards that goal. When transmission failures occur in such a system, the signal integrity engineer has to identify whether the failure mode is due to the passive or active layer, essentially because no universally adopted method currently defines passive channel interconnect performance in the physical layer (PHY). Data transmission in copper backplanes at 3.125 Gbps is becoming a routine occurrence in contemporary systems. Data transmission at 5 Gbps in copper appears to be the next benchmark for mass adoption. However, the emergence of the enabling technologies for 10 Gbps interconnect applications is changing the definition of what an interconnection system comprises.
Transceivers, printed circuit boards (PCBs), and connector technologies have integrated to meet the demands for 10 Gbps copper data transmission in the PHY. This has happened for two reasons. First, transmission standards are defined for application to active devices that communicate across a physical interconnect medium. By defining standards in this way, the model is seeking to treat the physical layer as a "black box."
Second, by defining performance at the active level, active techniques can compensate for performance deficiencies in copper interconnection systems. Specifying the PHY in this way has been convenient and completely appropriate in defining architectures where the physical layer has been operating at 2.5 Gbps and silicon algorithms are not required to compensate for poor performance of the copper interconnect. But for transmission at 10 Gbps, this approach constitutes a major over-simplification that conceals the significant engineering issues involved in high-speed interconnect design, and ignores the important cost vs. performance trade-offs that are considered in the design process.
Defining the PHY
For 10 Gbps transmission, the "black box" of the PHY needs to be redefined as a "two-color" box, inside which active and passive channel options can be traded off in order to reach optimum performance and cost. To understand the need for redefining how the physical layer is specified, the PHY must be split into two further levels defined here as the active interconnect system, which consists of the passive interconnect platform (PCB and connectors) and the active transceiver technology (see Figure 1).
 Figure 1. Active interconnect system model. |
null
The signal integrity engineer wishes to design a system PHY that can transmit a data stream through the active transmitter and passive interconnect platform and deliver error-free data at the output of the receiver. The performance of an active system transmission channel depends on three variables: the characteristics of the driver, the channel and the receiver.
To help determine the performance of the active system, a number of data patterns have been defined. By applying a pattern to the input of the active transmission channel and monitoring the bit error rate (BER) at the receiver output, engineers can determine if the system is performing as expected.
This is fine if there are no fault-finding diagnostics required. But if the BER test fails, how does the engineer determine the source of the problem — is it a function of the pattern, the driver, the receiver, the connector or the PCB?
 Figure 2. Passive interconnect platform. |
null
Consider applying the same pattern directly to a passive system (see Figure 2). Because there are no transceivers, the pattern is injected from the BER tester to the passive system through a launcher, and the output eye-pattern is captured on a BER tester. The bit error pass or fail is determined by applying a mask. But which is the best mask to use? This depends on which transmission protocol is required. Each protocol defines a pattern and a mask to be applied.
 Figure 3. Output eye-patterns from six different patterns applied at 10 Gbps to the same transmission channel. |
null
On the surface this seems a reasonable approach. By applying a mask to a pattern, it is possible to determine whether a passive channel passes or fails. If there is a BER failure, however, to which element of the passive channel is the failure attributable? This type of diagnostic information is difficult to determine from the eye-pattern. Figure 3 shows the output eye-patterns from six different patterns applied at 10 Gbps to the same transmission channel. Clearly, some of the patterns applied are more aggressive than others. The mask applied to pattern 6 may result in a pass while the same mask applied to pattern 1 would result in a fail.
S-Parameters
The electrical performance characteristics of the passive channel remain constant. So trying to characterize the electrical performance of a passive channel using patterns is inappropriate. Some other method is required. One popular method currently being adopted is "scattering parameters" or S-parameters.
As data rates have increased, the limitations of lumped modeling methods, such as RLGC (a mathematical equation that computes resistance, inductance, conductance, and capacitance), have become apparent. Lumped models have to make assumptions concerning the impact of physical structures on the overall electrical performance. At higher frequencies, even small physical structures can have a dramatic impact on the electrical performance. In these cases, a distributed model is required to characterize the system. S-parameters model the system as a "black box" multiport network (see Figure 4).
 Figure 4. S-Parameters model the system as a 'black box' multiport network. |
null
One major advantage of expressing the system in this way is that it fully defines the system over a range of frequencies. The use of a transfer function approach allows the system to be easily partitioned for modeling and measurement. S-parameters are relatively simple to simulate and measure. To accurately predict the performance of an active interconnect system, the industry has recognized the need for a complete S-parameter-based model for the entire passive channel. With this model and accurate driver/receiver models, system designers can model and predict active system performance.
S-parameters are a function of frequency and are expressed as complex numbers that can be difficult to interpret. A simple method of interpretation based on S-parameters needs to be established. One such method that is being adopted in certain sectors of the industry is the concept of the compliance channel.
Compliance Channel
The compliance channel refers to the passive elements of an interconnect system and a set of limits or masks that can be applied to the S-parameters for that passive system. These limits are analogous to masks that are applied to eye-diagrams in the time domain. One compliance channel currently being defined is the XFI (pronounced "ziffy") electrical specification, which is part of the XFP 10-Gbps pluggable optical transceiver specification. The XFI compliance channel consists of PCB traces, vias and the XFP connector. The XFI specification controls the reflections at the host end of the channel by requiring the compliance channel to meet a differential return loss of 10 dB from 1 to 7,500 MHz.
The compliance channel concept offers significant value to designers and manufacturers. Most importantly, it recognizes that S-parameters are the most complete and accurate method for describing the passive interconnect system. By setting limits on these parameters, it allows designers to understand exactly what performance to expect from the passive interconnect system. It also allows manufacturers to know what limits they are to meet with their products.
This method of specification also allows some flexibility in the channel implementation. For example, if the designer wanted a longer transmission line than could be supported in FR4 given the 6.5 dB maximum loss, they could choose a lower loss laminate to meet the longer distance goal. Similarly, a lower loss connector could be chosen to allow additional loss budget for trace length.
While the XFI specification is groundbreaking in its goal to define a compliance channel for 10 Gbps in copper, it is addressing a relatively simple topology compared to a typical backplane environment. The XFI channel is relatively short, has a specified surface-mount connector, has limited via stub effects and has little worry about crosstalk.
A broader 10 Gbps specification will need to encompass backplane environments where transmission distances are longer; connector options are varied and their models are more complex; the number of impedance transition points is increased; the stub effect of vias may be greatly increased; crosstalk concerns may be greatly increased and differential skew can be a significant problem.
Additionally, sophisticated techniques are being developed by transceiver manufacturers to overcome less-than-perfect channels. The range of techniques is growing. At the same time, there is a great deal of variation in the performance of high-speed channels. Some groups are attempting to use existing infrastructure (backplanes and connectors) to reach 10 Gbps data rates. Others are calling for new, higher performing technologies. A new framework is necessary to define future system designs of over 10 Gbps.
The new framework needs to include S-parameter-based limits or controls on the issues that impact passive system performance, namely loss, reflections, crosstalk and differential skew. The new framework also needs to meet the needs of varied users who are working with a wide range of different PCB and connector infrastructures and performance constraints.
These "tools" can be used to define an acceptable channel. To meet the needs of users, the framework should embrace those who want to use existing infrastructure as well as those who want to use improved technologies. Designers who want to use existing infrastructure will be working with channels that have high loss. To achieve satisfactory performance, these channels will need to be matched with transceiver technologies that possess sophisticated equalization or signal processing techniques. Other designers have the luxury of being able to choose to use low-loss connectors and backplanes and, therefore, can accept simpler and lower cost transceivers. There may also be a middle-ground performance with a moderate-loss channel and basic equalization techniques.
Conclusion
In implementing a 10 Gbps transmission system in copper, the system designer must trade off between the passive and active transmission system components to optimize performance and cost or accommodate legacy platforms. The emerging tools to enable this analysis are S-parameters and compliance channels. By adopting a framework based on these tools, it is possible to classify passive and active transmission system elements to facilitate the trade-off process.
Designers who must use existing channels that have relatively poor signal quality (Class C) may need to use silicon technologies such as adaptive equalization or pulse amplitude modulation (PAM) to compensate for channel issues. Channels with better signal quality (Class B) would need less sophisticated techniques such as pre-emphasis or basic receiver equalization. Channels with the best signal quality (Class A) may need no special silicon techniques. By adopting this type of framework, suppliers of transceivers and interconnect products have the opportunity to refine their product offerings for varying market requirements, while system designers can make design decisions early in the design process.
Adapted from a white paper presentation at DesignCon, January 27-30, 2003, Santa Clara, CA. The full paper can be viewed at www.winchesterelectronics.com.
BILL KELLER is Director of Business Development, Winchester Electronics, 62 Barnes Industrial Rd. N., Wallingford, CT 06492; (203) 741-5400; Fax: (203) 741-5500; Email b.keller@winchesterelectronics.com.
