One of the key limitations of all of these studies, however, is that they have been performed under nominal environmental conditions. This paper will show the impact that manufacturing variation and environmental conditions can have on the channel. Using actual test data, frequency domain analysis and pulse response theory will show the impact that these variations cause on the forward channel response, or S_DD21. This data will then be used to illustrate the inter-relations between signal encoding, channel response, manufacturing variation on channel response, and environmental conditions on channel response.

S-Parameter Characterization
Before looking a the impact of environmental conditions on backplane channel performance, let's take a look at how backplane channels are characterized. Scattering parameters (S-parameters) are used to characterize a channel's performance in the frequency domain. S_DD21, which is the differential response at the output of the channel based on a differential input, has been used as the main way of characterizing the capability of a channel.

Figure 1 illustrates the impact of the layer connection on the S_DD21 performance of 34-inch channels from Tyco Electronics HM-Zd XAUI Interoperability Platform.

There are multiple signal layers within backplanes and daughtercards. Connectors using compliant pin technology are inserted into plated through holes in the backplane and daughtercards. As the targeted signal layer connection moves upward towards the top layer, the amount of ripple in the S_DD211 profile can be seen to increase, grow, and become more "unpredictable" with frequency. This phenomenon can be attributed to the magnitude of the discontinuity associated with the plated through holes, which form resonance structures. Thus, the method of layer access will be critical, as any via barrel after the layer connection will result in these resonance structures.

From the discussion above, it's clear that the channel consists of two types of losses—predictable and unpredictable. The predictable losses are those losses associated with the transmission media itself. These are mostly skin effect losses that increase with the square root of the frequency, and dielectric losses that increase linearly with frequency. Dealing with the predictable losses by selecting better board materials will not overcome the degradation to the channel caused by the unpredictable losses. Instead the impedance discontinuity itself must be dealt with in order to minimize the unpredictable nature of the channel.¹

Ultimately, the engineer is interested in assuring that the necessary bit error rate (BER) requirements of the system are met. S_DD21 plots, however, only provide the engineer with a qualitative insight into the performance of the channel. While useful in their simplicity, S_DD21 plots are insufficient in assisting the engineer to verify that the necessary BER requirements are met. Thus, using the frequency domain data to generate pulse responses and apply convolution theory in time domain is garnering a lot of interest. In addition, when looking at a pulse response, inter-symbol interference (ISI) due to the spreading of the pulse is easily visible.

Figure 2 shows pulse responses for the channels presented in Error! Reference source not found. for operation at 3.11 Gbps. Several general observations can be made:

• The peak amplitude at Sample 0 is progressively reduced as the layer connection moves towards the top layer.
• The higher the layer connection, the greater the amount of ISI contribution at Sample -1 (pre-cursor contribution).
• At samples 1,2,3, etc, the ISI contribution (post-cursor contribution) is dependent upon the channel response itself.
• The ratio of each sampling point to Sample 0, which ultimately determines the maximum amount of eye opening possible, needs to be considered. While the absolute value of the ISI contribution at Samples-1, 1, 2, 3, etc may be similar, if the value at Sample 0 is reduced, the percent contribution at each of these sample points will be increased.
• The sampling point may not be at the maximum amplitude of the pulse.
• Receiver jitter (the receiver's determination of the sampling point) can significantly impact the amount of signal seen at each sampling point, due to the slopes of the pulse.

Figure 3 illustrates the reduction in peak amplitude and spreading of the pulse shapes at 3.11 Gbit/s. As system length increases, the impact on the amplitude at Sample 0, pre-cursor ISI contribution, and post-cursor contribution is clearly evident. So, for example, while the ISI contribution at Sample 1 has only increased from 100 to 150 mV, the amplitude at Sample 0 has decreased from 900 to 600 mV. So in going from 5 inches to 34 inches, the percent eye closure due to ISI at Sample 1 has increased from 11 to 25 percent.

Material Issues
The channel data presented previously does not include any insight into the influence of manufacturing variance or environmental factors on channel performance. Considering how the data presented so far demonstrates a roll off that increases with frequency, knowing if there is any additional factors that will increase channel loss will be critical. This will be necessary to know in order to determinine the long term viability of the backplane's ability to support higher speeds in order for the system to support higher capacities.

It has been shown that dielectric constant (d_k) and dielectric loss tangent (d_f) vary with frequency.² This, however, only scratches the surface of the real problem. Temperature also has a major impact on the d_k and d_f of any given material. To be brief, temperature increases both the d_k and d_f of most laminate materials because of the increased activation of ionic species in the resin matrix.³ Absorbed moisture also has an impact on the d_k and d_f of all materials since water has a higher d_k and d_f than the base materials.

Polarizations that exist within the dielectric material force these two parameters to be a function of frequency. Dipole polarizations due to polar molecules and interfacial polarization due to inhomogeneities in the material directly explain why the moisture level in a dielectric substrate will affect its electrical properties. The rate of polarization and the movement of free ions/electrons are strongly influenced by temperature. Thus, the dielectric constant and loss tangent will be dependent on frequency, temperature, and the material content, e.g. moisture.

Figure 4 shows that when a Glass/FR4 dielectric material was bone-dry, e.g. 0 percent moisture content, an increase in temperature (25 to 80 to 120 deg. C) increased the dielectric constant, and subsequent increases in moisture content further exagerated this increase. The range of dielectric constants over temperature and moisture varied from 4.8 to 5.8.

Figure 5 shows that when a glass/FR4 dielectric material was bone-dry, e.g. 0 percent moisture content, an increase in temperature (25 to 80 deg. C) actually improved the dissipation factor, d_f from approximately 0.015 to 0.009. However, as the moisture content increased from 0 to 0.5%, the d_f for the materials at both temperatures was nearly equivalent at 0.02. Testing at 120 deg. C yielded the opposite of both of these trends—the d_f increased with temperature while decreasing with additional moisture content.

Figure 6 shows the impact of frequency testing on dielectric constant when temperature and moisture are included. Figure 7, on the other hand, shows the impact of frequency testing on dielectric constant when temperature and moisture are included. It is important to note that the testing is only one up to 1.5 GHz.

Using the stripline test method, Park Nelco (www.parknelco.com) examined the interaction between d_k, d_f, temperature up to 19 GHz at a stable relative humidity of 52.46 percent for Nelco 4000-13.⁵ Figure 8 demonstrates the relative stability (3.6 to 3.8) of this material over temperature and frequency at the constant humidity test condition. Figure 9, on te other hand, demonstrates how the d_f will flucuate over temperature and frequency (0.013 to 0.017).

This data demonstrates that the interrelation between temperature, humidity content, and frequency will cause a variance in the performance of a channel. It needs to be pointed out at this point, however, that this is only looking at the material by itself, and not in a system environment, where the change in d_f will have an accumulative effect with overall system length.

Environmental Variance
Several good methods are available to characterize a printed circuit board (PCB) d_k and d_f performance over temperature, humidity, and frequency.^7,8,9 Unfortunately these methods do not lend themselves easily to measuring a backplane system.

Backplane system-level temperature and humidity verses frequency performance measurements require at least two items. The first is a reach in temperature/humidity chamber to accommodate the backplane. The second is a vector network analyzer (VNA) to measure the frequency response.

A balanced four-port XAUI style backplane with daughter cards was used to test environmental variance. All the PCBs were made from a lossy FR4 type material with HM-Zd connectors Figure 10.^10,11,12,13

The backplane system was placed inside a reach in temperature/humidity chamber from Cincinnati Sub-Zero Products.¹⁴ A set of test cables were routed from the backplane daughter cards to a four-port VNA outside the chamber.¹⁵ The Agilent VNA test setup (Figure 11) contained an 8720D Network Analyzer, an ATN4112A test box, and a laptop PC with mMultiport Software (v1.1649.596).

A room temperature and humidity calibration run was done with the reference plane at the end of the cables inside the chamber. This calibration file was then used for all data taken. To determine the error introduced by this calibration file, two thru cable mix-mode measurement runs were performed at 20 and 85 deg. C.¹⁶ The S_DD21 and S_DD12 magnitude data was compared and the following error recorded:

The 16-in. C0 channel (total system length is 20 inches, which includes 2 inches per line card) on a XAUI backplane was then used to collect mix-mode S-parameter data from 50 to 20050 MHz. The chamber was set at the following points:

To determine the dwell time at a particular temperature and humidity, a control XAUI backplane was placed in the chamber and removed once a day for weighing.⁷ When the weight had stabilized to within the error of the scale (0.5 mg), the S-parameter data was recorded and a new set point programmed into the chamber.

Figure 12 shows the S_DD21 performance and phase delay of the same channel from the two builds. Note the increase in variation with frequency of the throughput with the associated variance in the phase delay.

Figure 13 demonstrates the pulse response at 5.15625 Gbit/s, showing in time (left) and then normalized to its Sampling 0 point (right). Note the increased variation in amplitude reduction (600 mV best to 450 mV worst), propagation delay, and post cursor ISI contribution.

Figure 14demonstrates the pulse response for the different measurements at 10.3125 Gbit/s. Note the further exasperation of the problems noted at 5.15625 Gbit/s operation.

Signal Encoding
To understand the ramifications of both temperature and humidity, a simulated test bench was created to quantify the signal changes versus coding schemes and data rates. Two coding schemes are presented here, namely non-return-to-zero (NRZ) and PAM-4 (a multi-level coding scheme). NRZ has been a long-standing serial coding scheme used in the industry. Examples of NRZ are fibre channel, XAUI, Gigabit Ethernet, and TFI-5. The data rates where varied from 3 to 5 Gbit/s and finally 12 Gbit/s.

For the 3-Gbit/s data rate analysis, a two tap FIR transmitter and a single tap decision feedback equalization (DFE) technique, similar to current a serdes offering today, was created to demonstrate the effects of both temperature and humidity. For both the 5- and 12-Gbit/s data rate analysis, an 11-tap fast impulse response (FIR) transmitter and an 8-tap DFE were utilized.

In addition to the eye diagrams, a quantitative unit of measure was employed to evaluate the effects of the environment. This unit of measure is the signal-to-ISI ratio (SIR). To denote the effects of both the transmitter's FIR and receiver's DFE, units of measure are labeled SIRF and SIRD, respectively. The following equation was used to calculate SIRF and SIRD for subsequent analysis:

where w(k) = received NRZ or PAM-4 equalized symbols at the input of the NRZ or PAM-4 slicer (V,-V) for NRZ or (V, V/3, -V/3 and "V) for PAM-4
and a(k) = corresponding transmitted ideal NRZ or PAM-4 symbols.

SIR(D/F) measures the residual ISI. The SIR(D/F) does not include the nominal 9-dB detection signal-to-noise ratio (SNR) loss of PAM-4 relative to NRZ predicted from the reduced level separation. SIR measures the "size" of the ISI "cloud" only. Hence, when comparing SIR(D/F) between NRZ and PAM-4 encoding techniques, PAM-4 SIR numbers need to be reduced.

The exact comparison between NRZ and PAM-4 SIR values requires more details about the actual implantation (i.e. CDR loop and noise)—both internal and external. For the short simulation bit patterns used in this study (2¹⁶ bits), SIRD values for NRZ encoding above 14 dB corresponds to error-free performance (zero BER) and SIRD values for PAM-4 encoding above 23 dB corresponds to error-free performance.

As shown in Table 2, for 3-Gbit/s data rates, neither temperature nor humidity affected the SIRD values significantly. This reinforces why present day systems have not highlighted concerns due to environmental effects. At the higher 5 Gbps data rates, however, the variance of SIRD increases. Yet, by optimizing the tap settings as temperature and humidity varies, a nearly constant SIRD value of 19.5 dB can be maintained.

Table 2 also compares the NRZ and PAM-4 encoding techniques at 12 Gbit/s. For 12-Gbit/s NRZ, only the 20 and 40 deg. C case come close to the desired error free operation and optimization at higher temperatures and humidity does not provide adequate performance. For 12-Gb/s PAM-4, reasonable SIRDs can be maintained over all temperature and humidity cases.

In addition to SIRD values, eye diagrams provide valuable insight into the effects of temperature and humidity. Typical 3 Gbps systems today utilize pre-set tap values. Figure 15 shows that even with mis-matched tap settings, the eye is reasonably open for error free operation. It must be pointed out, however, that this is only for a 20 inch system channel. Channels with additional loss may exhibit more sensitivty than seen here.

Note: The indifferance for the channel shown in Figure 15 does not hold true for 5-Gbit/s data rates and higher.

In Figures 16 to 18, three eye diagrams are shown. The first eye diagram optimizes the settings of the transmitter and receiver for the channel at 40 deg. C and 20 percent RH. The second eye diagram uses the settings developed in the first eye diagram, but applies them to the channel data at 85 deg. C and 85 percent RH. The third eye diagram optimizes the settings for the channel performance at 85 deg. C and 85 percent RH. Figures 1 and 17 are for NRZ encoding at 5 and 12 Gbit/s respectively, while Figur 18 is for PAM-4 encoding at 12 Gbit/s.

From Figures 16 to 18, there are several things to note regarding the observations of performance over temperature, humidity, and frequency:

1. The decrease in SIRD when setting for the transmit and receive are not re-optimized between conditions.
2. The similarity in SIRD when the settings of the transmit and receive are re-optimized between conditions.
3. At 12 Gbpit/s the delta of the SIRD of the PAM-4 solution using reoptimized settings is minimal (-0.2 dB) as compared to the NRZ solution (-3.4 dB).

Wrap Up
This discussion above has demonstrated the sensitivity of channel performance to manufacturing and environmental influences on channels of 20 in. and 34 in. in length and their subsequent increasing impact as the frequency of operation increases. Furthermore, while data from two different board materials has been reviewed in this paper from a system perspective, no sort of mechanism has been developed that can be used to estimate what any material's sensitivity to these effects will be for system level evaluation.

From a system level perspective, once the behavior of the nominal channel has been identified, the engineer needs to worry about channel loss from manufacturing and environmental influences. It is also important to note that there was a dwell time variation with temperature, indicating a potential temperature dependant diffusion process. Thus, humidity could cause, not only, time dependant effects, but additionally, location dependant effects. Additionally, it is important to note humidity at low temperatures did not create as much of an issue. Whereas, humidity in combination with temperature 60C or higher really caused a large impact on channel performance. Thus, any residual humidity in the board, due to manufacturing or board exposure to humidity, might not show up until higher temperatures of system temperature are achieved.

Understanding the reality of the channel then, allows a more informed decision on evaluating the electrical signaling technique. NRZ signaling, which inherently has a higher Nyquist frequency, will be more sensitive to these factors, which increases with frequency. Multi-levels signaling, however, is less sensitive to these factors since its Nyquist is at least ½ of the NRZ technique, depending on the number of levels for multi-level signaling chosen.

In addition to the type electrical signaling technique chosen, devices must employ some sort of adaptive process to their settings, since the channel itself has been demonstrated to be changing. Programmable devices will not be able to meet and maintain desired BER over all the variances discussed, and their ability to do so will be limited as the frequency of operation increases.

Authors' Note: The authors would like to acknowledge Pervez Aziz of Agere Systems, Chris Wittensoldner of Agere Systems, Adam Healey of Agere Systems, Andy Marocchini of Tyco Boards and Silvio Bertling of Park Nelco for their help with this article.

References

1. "Reliable Serial Backplane Data Transmission at 10 Gbps," D'Ambrosia, Fogg, Lazaris-Brunner, Tailor, DesignCon 2002.
2. "The Impact of PWB Construction on High-Speed Signals," Helster, Morgan, DesignCon 1999.
3. "Dielectric Properties and High-Speed Electrical Characterizations", Mumby, & Schwartzkopf.
5. Comparison of the Dielectric Constant and Dissipation Factors of Non-Woven Aramid / FR4 and Glass / FR4 Laminates, Khan, DuPont Advanced Fiber Systems, September 1999.
6. Correspondence from Silvio Bertil, Park Nelco, to John D'Ambrosia, Tyco Electronics, 11/17/2003.
7. Ongoing Telephone Conversations with Silvio Bertling, Park Nelco.
8. IPC TM 650 Test Methods Manual section 2.5.5 Stripline Testing.
9. James Baker-Jarvis, Richard G. Geyer, John H. Grosvenor, Jr., Michael D. Janezic, Chris A. Jones, Bill Riddle, Claude M.Weil, Jerzy Krupka, "Dielectric Characterization of Low Loss Materials",IEEE Trans.Dielectric&Electrical Insulation,Vol.5,pp.571-577, Aug. 1998.
10. IPC TM 650 Test Methods Manual section 2.5.5.3 Two Fluid Cell Testing Method.
11. Tyco Electronics PN: B024519-011 Rev.1 S/N 60703-3
12. Tyco Electronics PN: PWB#58092 PNL#8.
13. Park Electrochemical Corporation PN: N4000-2.
14. Tyco Electronics PN: 120670-2 and 1469001-1.
15. Cincinnati Sub-Zero Products Model Number: ZHS-32-2-H/AC
16. Delaire USA PN:DC4031-00.
17. David E. Bockelman, William R. Eisenstadt; "Combined Differential and Common-Mode Scattering Parameters: Theory and Simulation"; IEEE Trans. Microwave Theory & Tech. Vol.43 No.7 July 1995.
18. Mettler scale Model AJ100.

About the Authors
Greg Sheets is the director of high-speed interfaces at Agere Systems. He received a B.S. in Chemical Engineering from Cal-Poly, Pomona and M.S. and Ph.D. degrees in Electrical Engineering from Arizona State University. Greg can be reached at sheetsg@agere.com.

John D'Ambrosia is the manager of semiconductor relations for Tyco Electronics. He received a B.S. in Electrical Engineering Technology from the Pennsylvania State University and a Master's Degree in Engineering Management from the National Technology University. John can be reached at john.dambrosia@tycoelectronics.com.