Signal integrity, at a 'glance': eye diagrams can be the window into crosstalk and jitter problems. An eye-opening look at the process. (Eye Diagrams).

Many types of faults can affect signal integrity. Single-net faults such as improper terminations can cause reflections. Crosstalk can occur from one net to another, and power and ground faults can introduce noise into signals. Jitter can be caused by external sources modulating the signal or by data-dependent delays.

Eye diagrams can help identify many of these high-speed issues. By examining the eye it can be quickly verified that all the critical data sheet parameters of the latch have been met: upper and lower noise margins, setup time and hold time.

Accurate measurements, including eye measurements, start at the probe tips. This is often a problem area. No matter the scope, it can't make measurements more accurate than the probe permits.

In the real world, scope probes do two things. First, they become part of the circuit. As such, they can change its signal integrity. This is because the scope probe, now matter how well designed, presents reactive loading and resonance to the circuit under test. FIGURE 1 shows an eye with a probe designed for minimal loading at the probe tip attached. Without going into detail on probes and loading, I'd recommend reading Reference 1.

[FIGURE 1 OMITTED]

Probes also add noise and distortion, and have their own bandwidth limit. The measurement system's signal-to-noise ratio, bandwidth and response flatness can't be any better than the probe, and are often limited by the probe.

The next important parameter is jitter in the scope. Jitter in the scope comes in two flavors: trigger litter and time base jitter. For conventional eye measurements, triggering on the clock and running the scope in repetitive mode with infinite persistence, trigger jitter tends to be dominant.

FIGURES 2a and 2b show an eye measurement on a signal traveling 3 Gb/sec. Figure 2a shows what it looks like with 1 psec trigger jitter. Figure 2b is what it would look like with 7 psec of trigger jitter. The difference in the apparent eye opening is easy to spot.

[FIGURE 2 OMITTED]

Trigger jitter is easy identify and measure. Trigger the scope on a reasonably fast-rising signal. Set the time base so the rising edge is at an ~45[degrees] angle (or as low an angle as can achieved with the sweep speeds available in the scope). Set the trigger point to the middle of the screen with the time base offset (horizontal position) control (T=0). Set the scope to infinite persistence. Note where the signal passes through the trigger level--most scopes have a trigger level indicator or marker that makes this easy.

Ideally the signal would cross over with zero width at that point, since that is supposed to be the zero-time reference for the time base. If the scope has a histogram function, use it to measure the trigger jitter. Take a histogram of the signal at the trigger level crossing. The 1 sigma value of the histogram is the RMS trigger litter.

On the other hand, if a single-shot capture is used along with the scope's internal software clock recovery to create the eye diagram, then trigger jitter doesn't matter. What matters is the time base jitter and wander in the scope. Look in the data sheet for both trigger jitter and time base jitter (which may be expressed as jitter measurement noise floor). To characterize the scope's time base jitter is a bit more complicated; it requires a very stable signal source, so the source jitter is less than the scope's jitter.

An often-overlooked source of apparent jitter is vertical (voltage) noise introduced by the scope and probe. On an edge with non-zero rise time, vertical noise translates to noise in time, because the time of crossing the trigger level varies from what it is in the real world.

Many newer serial interfaces use double-data-rate (DDR) signaling, in which data are transferred on both the rising and falling edges of the clock. Since the scope only triggers on rising or falling edges (but not both), the eye observed when eye measurements are conventionally made corresponds only to the rising or falling edge. To see the opposite eye, switch the trigger polarity on the scope. Software clock recovery doesn't have this limitation: it specifies rising edges, falling edges, or both.

Many oscilloscopes come with a number of built-in automatic eye measurements such as eye height and width, mask violation testing, crossing percent, Q-factor and duty cycle distortion. In many cases, standard mask tests are available for pre-compliance testing to various industry standards.

Sometimes you need a source of signals for eye measurements. Look for pulse generators that don't contribute jitter themselves. To see how well a circuit rejects jitter may require a pulse generator whose output can be modulated with known amounts of jitter, supplied by an external source. References 2 and 3 discuss stressed-eye measurements using a pulse/pattern generator and a scope.

Some logic analyzers can make eye measurements on many signals at once. This technique offers considerable time saving over probing tens or hundreds of nodes, a few at a time, with scope probes. Reference 4 discusses this approach.

In the design phase, many simulation tools are now available to predict eye openings based on circuit models. Checking eye openings before committing to boards and silicon can save a lot of time by revealing problems as early as possible in the project.

REFERENCES

(1.) "Restoring Confidence in Your High-Bandwidth Probe Measurements," Agilent publication no. 5988-7951EN, http://cp.literature. agilent.com/litweb/pdf/5988-7951EN.pdf.

(2.) "Jitter Analysis Techniques for High Data Rates," Agilent publication no. 5988-8425EN, http://cp.literature.agilent.com/litweb/pdf/5988-8425EN.pdf.

(3.) Characterizing Jitter On High-Speed Communications Signals, www.netseminar.com/index.cgi?sem_num=570.

(4.) "Saving Time with Multiple-channel Signal Integrity Measurements," Agilent publication no. 5988-5409EN, http://cp.literature.agilent.com/litweb/pdf/5988-5409EN.pdf.

ART PORTER is a product manager at AgilentTechnologies (agilent.com).