Electronic Design

Accurate Measurements On High-Speed Rambus Traces Present Challenges

Using time-domain-reflectometry normalization and fast oscilloscopes, measurements can be made on very short traces.

Making highly sensitive measurements on high-speed computer motherboards and memory boards is tough, especially when the traces are short. Such short traces (e.g., 1 to 2 in. in length) are common in the Rambus memory architecture, which uses 250-ps rise-time signals. They also can be found in the smaller-form-factor pc boards used in laptop computers and network equipment applications.

The issues brought about by short traces can be conquered using the test procedure called time-domain-reflectometry (TDR) normalization. Through this process, high-speed oscilloscopes can accurately measure the characteristic impedance of short traces by removing sources of error.

Three basic measuring tools are needed to measure the characteristic impedance of those traces: a high-speed digitizing oscilloscope with a modular TDR plug-in (Fig. 1), a 10-GHz TDR probe, and a TDR calibration substrate (see R&D Equipment List). The only caveat is that the TDR oscilloscope must have normalization capability.* Ideally, the versatile TDR equipment is suited for an R&D environment. If a high-volume manufacturing capability is required, additional test equipment is recommended (see Manufacturing Equipment List).

A TDR Calibration Substrate is used to remove the errors that are usually introduced by test fixtures. Probe-tip ground-lead inductance, reflections from connectors, and cable loss are typical examples of these errors. The normalization process demands that the user perform a two-point calibration at the probe tip. An appropriate digital filter can then be automatically configured in the oscilloscope firmware. It's recommended that this calibration process be performed while probing Rambus In-line Memory Modules (RIMMs) and Continuity RIMMs (CRIMMs). When probing short traces on motherboards and Small-Outline RIMMs (SO-RIMMs), however, the process is mandatory.

The calibration substrate has precision, thick-film resistors that are laser trimmed to yield the utmost in accuracy (Fig. 2). To ensure a long life with various probe-tip configurations, durable gold electrodes are fired onto the alumina substrate. That substrate is a non-reactive standard. Due to the lower-frequency content of time-domain reflectometers, it's favored for Agilent TDR measurements. A 28-Ω airline isn't recommended for the company's TDR calibration, but it can be used effectively with the vector network analyzer.

Calibration And Verification
Using calibration and verification together will enhance the confidence of Rambus board designs. To minimize measurement error, calibrate a 50-Ω output-impedance instrument in a 50-Ω environment. Next, verify the calibration by measuring a well-known impedance value that's close to the characteristic impedance of the device under test (DUT). This two-step process guarantees that the measurement equipment is highly accurate in the region of interest.

The two-step process is as follows: Using one TDR probe with suitable bandwidth and an oscilloscope with normalization capability, probe the short standard on the calibration substrate. This is the electrode area indicated on the left side of Figure 2. Next, probe the 50-Ω precision resistor. Be sure to place the ground pin of the TDR probe on the larger electrode pad, because the excess capacitance of the larger electrode effectively compensates for the excess inductance associated with most probe ground tips. After confirmation from the TDR scope, the calibration is complete.

The next step is verification. Start out by probing one of the 28-Ω, ±0.25% thick-film resistors located on the calibration substrate. For the 28-Ω standard, the ±0.25% resistor tolerance translates into ±70 mΩ. This resolution is important for Rambus applications, in which the ±2.8-Ω tolerance is essential for proper operation of the signal channel. Some Asian pc-board manufacturers have recently tightened the specification to ±1.4 Ω.

Activate the normalized waveform with the following button-press sequence: plug-in setup (hard key), normalize response (soft key), TDR normalize on (soft key). When performing verification, use the normalized TDR waveform. The standard TDR waveform will include secondary reflections that will introduce significant inaccuracies at the front end of the DUT (up to 3 Ω of error for Rambus SO-RIMMs).

On pc boards, TDR provides useful insights for the digital design engineer trying to understand signal-integrity issues. When no connectors are conveniently located near a circuit trace on a board, it's common to launch the TDR step into the trace with a high-bandwidth TDR probe. But finding the optimal location on the board to probe isn't always easy, especially since the results obtained will vary depending on the launch point chosen.

As seen in the memory architecture of the Rambus circuit topology, shown in Figure 3, the first microstrip segment under test is located between the Rambus memory-controller chip (RMC) and the first RIMM connector. Figure 4 reveals the impedance profile of this motherboard microstrip trace when probing from the RIMM #1 connector side.

Sometimes, it's useful for the designer to troubleshoot the whole Rambus channel with all of its components, although most TDR measurements are performed on bare pc boards. The motherboard with the test waveform in Figure 4 has already been populated with the various components associated with the Rambus physical layer. Check out the higher impedance of the RMC package's ball-grid-array (BGA) structure. This particular BGA adds excess inductance to the 28-Ω characteristic impedance of the motherboard microstrip. It thereby pulls the impedance higher.

Capacitance-Pad Compensation
To enhance signal integrity, various high-speed digital design techniques are employed during layout. An example being used on Rambus motherboards is the capacitance pad, or "C-Pad." At about 0.075 in. long, this trace structure can be visibly seen on the underside of the motherboard as a short stub. Its purpose is to compensate for the excess inductance of the RIMM connector lead, which penetrates the via after board population.

Capacitance-pad compensation is a well-known technique that enhances signal integrity at high speeds by closely controlling the impedance environment. When a fast-rise-time edge from a Rambus signal encounters this structure, it will travel in both directions and essentially see two 28-Ω transmission lines in parallel for a very short distance. This creates an area of effective low impedance. It also compensates for the higher impedance of the excessively inductive connector wire.

An interesting design flaw in a motherboard C-Pad has been discovered, as seen in Figure 5. The impedance profile shows a low dip in the front side of the waveform. The designed geometry of the C-Pad has overcompensated for the inductive connector lead by pulling the impedance too far down. In this case, the correct geometry for proper impedance compensation would be a reduction in trace width. The normalized yellow TDR waveform reveals the impedance profile after the stub is removed with an X-Acto knife.

Choosing the proper point to launch the TDR step can make measuring easier. Figure 6 shows the results achieved from using the recommended procedure for measuring Rambus motherboard traces—probing a trace from the memory-controller side of the microstrip segment.

The motherboard must be unpopulated, allowing the easiest access to most of the Rambus traces of interest. The benefits of this probing location can be seen in the flat and uncluttered impedance profile of the trace. Plus, many ground vias are available in this motherboard, and the smaller via size doesn't require a stub for compensation.

The Small-Outline RIMM is a new- form-factor memory module that will be used for laptop-computer memory. Thus, it's much smaller than a standard RIMM module. The challenging part of measuring the characteristic impedance of an SO-RIMM is the very short 28-Ω microstrip segment, typically only 1 to 1.5 in. long.

The characteristic-impedance profile of an SO-RIMM has three distinct segments (Fig. 7). The first is a 28-Ω segment. The second, which is the higher impedance at typically 56 Ω, is located in the middle. Lastly, there's another 28-Ω segment. The high-impedance segment in the middle is where the silicon memory devices are loaded onto the board.

At Rambus signal speeds, the silicon is modeled as an RLC equivalent circuit with an effective 56-Ω impedance shunted to ground. After that silicon is loaded onto the board, the high-speed data sees two 56-Ω microstrips in parallel (28 Ω). The post-silicon controlled-impedance environment creates minimal reflections. It's actually the basis for choosing the 28-Ω characteristic-impedance value for the Rambus memory architecture. This principle of bare-board impedance modulation also is used in standard RIMMs.

The enhanced accuracy of TDR normalization is easily seen in the SO-RIMM measurement. Secondary reflections would ordinarily obscure the correct impedance reading in this short trace. But those reflections are nonexistent on the normalized TDR waveform. The calibration substrate was used with a high-frequency TDR probe to set the reference plane at the probe tip. The standard TDR waveform shows how secondary reflections produce measurement error.

A notable phenomenon visible in this TDR measurement is the lossy nature of the particular SO-RIMM used in this experiment. The upward slant of the middle 56-Ω segment would normally indicate a microstrip with increasing impedance. Because the geometry of the known test device was well controlled, however, this cannot be the case. So, this upward slope indicates either dielectric loss or skin-effect loss of the microstrip. Though the second 28-Ω segment near the end of the SO-RIMM seems to have a higher impedance, it does not.

Like the upward slope in the 56-Ω middle segment, the losses encountered in the pc-board construction create a distributed impedance effect. This makes it seem like the trace has a higher impedance. The point to remember is that the trace's characteristic impedance hasn't changed. Just reverse the device under test and launch the TDR step from the opposite end. A similar upward slope will be observed, confirming this effect. Therefore, probing should always be placed nearest to the trace area of interest.

Coupon Measurements Validate It
An experiment was performed to validate the accuracy of the TDR measurement on progressively shorter Rambus traces. A 6-in.-long, 28-Ù microstrip test coupon fabricated from FR4 material was used. The characteristic impedance was measured with the 10-GHz TDR probe. The resultant TDR measurement can be seen in Figure 8 as the longest trace waveform.

The board was then cut in half to measure the remaining 3-in. section (middle waveform), and cut one more time to yield a 1.5-in. section. A measurement marker was placed 26.4 mm from the probe-tip reference plane, marking the physical location within the microstrip structure where the measurements were taken. At the 26.4-mm marker location, the three measurements were within 300 mÙ of each other.

When performing characteristic-impedance measurements of Rambus motherboards and SO-RIMM modules, the errors that are introduced by test fixturing and probes must be removed. The critical nature of Rambus signal-integrity measurements calls for calibration procedures beyond the standard TDR test methodologies. Effects of ground-lead inductance, cable attenuation, and connector reflection increase measurement error outside of the acceptable limits for short Rambus traces. When working with Rambus motherboard and SO-RIMM signals, get the necessary confidence with time-domain-reflectometry normalization and verification with a precision calibration substrate.

*Normalization is firmware built into the HP 54750A that's based on the Bracewell transform originally licensed from Stanford University. A detailed normalization measurement technique is described on Intel's web site at the URL http://developer.intel.com/design/chipsets/memory/pcbtest.htm.

R&D Equipment List:
HP 54750A digitizing oscilloscope
HP 54754A differential TDR plug-in module
HP N1020A TDR probe kit
HP 54006 6GHz resistive divider probe
HP 54701A 2.5 GHz Active probe
HP 54121-68701 RF accessory kit

Manufacturing Equipment List:
HP 83480A-K16 Rambus RF switch matrix
HP 83480A-K17 Rambus RIMM test fixture

Recommended Reading:
Resso, M., "The Right Technique Yields Critical Direct Rambus Signal-Integrity Measurements," Engineering Design News, Aug. 5, 1999.
"Evaluating Microstrip with Time Domain Reflectometry," Hewlett-Packard Application Note, HP Publication 5968-0007E, 1998.
Dascher, D., "Measuring Parasitic Capacitance and Inductance Using TDR," HP Journal, April, 1996; available on the web at www.hp.com/hpj/apr96/ap96a11.htm.

Hide comments


  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.