Probing High-Speed Digital Designs
Have you ever encountered a broken signal that worked only when your probe was touching it? Join the crowd. It's a badge of honor. It means you work on really fast systems. Then again, it may just mean you need a better probe. The one that you are using just isn't up to par, or the way it is being used is inappropriate for the task at hand. To help solve such problems, this article will explore some ways to characterize their behavior, and the trade-offs inherent in various probe styles. It will even describe how to make a resistive-input probe that performs well into the gigahertz range.
How Probes Work
Basically, all probes work the same way. When applied to a logic trace, a probe "siphons" off a portion of the signal energy and conveys it to the scope's vertical amplifier input. From there, the scope amplifies the signal and then displays it on the instrument's screen.
The siphoning process always distorts the signal being measured, because any probe loads down the circuit to which it is connected. Even with a 1-pF probe, the loading can be substantial. A 1-pF probe looks like a 160-ohm load at 1 GHz, which is the highest frequency associated with a 0.5-ns rise or fall time. How did I know that? Given any ordinary digital logic 10-90% rise or fall time T, the effective upper band edge associated with that rise time equals approximately 0.5/T Hz (see High-Speed Digital Design by H.W. Johnson and M. Graham, Prentice-Hall, 1993.) Technically, the complex impedance equals negative j times 160 ohms, but that's splitting hairs because the phase doesn't matter as much for simple digital loading problems nearly as much as the simple fact that a 160-ohm magnitude appears very noticeable to a 50-ohm circuit.
Think about it. If you connected a 160-ohm load to your circuit, it's going to change the termination conditions. Wouldn't the levels shift? Wouldn't the signals change shape slightly? Might it not ring, or overshoot differently, or cross the switching threshold at a different point in time? These same effects occur when probes are connected.
Room For Improvement
Some engineers assume that these effects are a manifestation of the Heisenberg Uncertainty Principle, but that is not the case. For ordinary digital problems, probe performance is nowhere near its fundamental physical limits. The problems are simply a manifestation of the rather crude state of the art of probe design. Better probes will do less damage to the signal under test. The industry can anticipate several more generations of improved probe designs before encountering limitations due to the immutable laws of physics.
You may be interested to know that electrical engineers in many other fields of study are also concerned with the general effect of probes on the device under measurement. A good general reference on the subject is, Electrical Measurements, by Frank A. Laws, first published by McGraw-Hill in 1938. We are not dealing here with any fundamentally new problems.
Besides the loading problem, a probe can introduce its own distortion, often in the form of additional ringing or overshoot. Even if it doesn't load down the circuit under test, a probe whose internal workings are ringy will fail to convey to the oscilloscope a faithful reproduction of the incoming signal. The actual waveforms in the circuit under test may look ideal, but what is conveyed to the scope looks completely different.
I can't count the number of times I have seen engineers chase down "ringing" problems in a circuit, trying every termination trick in the book, only to discover that the ringing was not present in the system at all, it was only a ghost image created by poor probing.
Three Probe Styles
There are three popular oscilloscope probe styles in use today:
- 10:1 Capacitive-input probes
- FET-input probes
- Resistive-input probes (also called Z0 probes)
The capacitive-input probe was originally developed for use on vacuum-tube equipment (Figure 1).
It provides a very high input impedance at dc (about 10 M-ohm), which was a nice feature when engineers spent a lot of time probing grid-bias circuits on vacuum tube systems. Nowadays digital applications don't require a 10 M-ohm input impedance at dc. For digital applications, the probe's impedance at very high frequencies is much more important.
Proper operation of the capacitive-input probe hinges on the assumption that the center conductor of the connecting cable has an aggregate capacitance to ground of 50 pF. At frequencies for which the cable begins to act like a transmission line (that is, at the frequencies you care about in fast digital design), the probe no longer performs correctly. A little box of compensating components at the end often includes a circuit to help ameliorate this effect, but because of the fundamental limitations of the connecting cable few probes of this style are rated for more than 500 MHz.
The FET-input probe (Figure 2) has an active amplifier built right in to its tip. This circuit, which incorporates an FET-input buffer stage, amplifies the incoming signal and prepares it for its journey down the 50-ohm connecting cable to the scope. To use this probe, the scope must be equipped with a 50-ohm-terminated input circuit, and a power connection to feed bias power to the FET amplifier. Always check to make sure the power from your scope is compatible with the FET probe you are planning to use.
The resistive-input probe, also called a Z0 probe, combines characteristics of both of the other types (Fig 3). Like the 10:1 capacitive-input probe, the resistive-input probe is an entirely passive device. That means that it will work with practically any scope. Like the FET-input probe, the resistive-input probe makes optimal use of its 50-ohm connecting cable. Once the input signal is coupled into the cable, it flows in a linear, time-invariant, almost lossless, and practically distortionless fashion all the way to the scope input termination, where reflections are damped. The scope must be set for a 50-ohm termination.
The resistive-input probe is cheap, it has a terrific bandwidth, and it is more tolerant of long ground wires than the other probes. These advantages come at the cost of higher IOH required in your digital circuits in order to drive the 1K resistor. In modern high-speed systems, the extra drive current is almost always readily available.
How to Characterize a Probe
Probes are available in many different styles, shapes and sizes to suit a wide variety of applications. Not all styles are appropriate for digital use. As an aid to choosing probes for a digital design lab, this section examines how to characterize those aspects of probe performance most relevant to high-speed digital logic applications.
Probes can load down a circuit, substantially distorting the signal under test. This happens when the input impedance of the probe is comparable to (or less than) the driving impedance of the device under test.
Figure 4 depicts the effects of probe loading. The figure shows three traces, all measured with a high-quality reference probe installed at the end of a long, source-terminated trace. In all three cases, the trace impedance is 50 ohms, and the signal rise-fall time as measured on the scope is about 2 nS. In all three cases, in addition to the reference probe used to make this picture, the signal is loaded with one additional scope probe. The difference between the traces is that in one case, the additional loading probe is a 1-pF FET-input probe, in one case it is an 8-pF capacitive-input style probe, and in the third case it is a 1K-ohm resistive-input probe. A separate trigger circuit is used to maintain time-synchronism between the three measurements, which have been superimposed onto this figure.
Even at the rather pedestrian signal speed of 2 nS, the loading effect of the 8-pF probe shows clearly (blue dotted line). When the 8-pF probe is attached, the rising edge is delayed by about 200 pS. In systems with little or no timing margin, this can easily be enough to cause a noticeable change in system behavior. At the frequency associated with this rising edge (250 MHz), the input impedance of the 8-pF probe is a mere 80 ohms, hardly good enough for fast digital work.
In contrast, the 1-pF FET probe and the 1K-ohm resistive-input probe do not materially affect the transition time, although the 1K-ohm probe does have the effect of scaling the signal amplitude to 95% of its nominal open-circuit value ( 1K/(1K+50) = 95% ). The input impedance of both these probes, at the frequency of interest (250 MHz), is much higher than 80 ohms.
As we go higher in frequency, eventually the 1-pF probe will run into difficulties. At signaling rates faster than about 300-pS rise-fall, only a resistive-input style probe can maintain a high enough input impedance to remain useful.
Bandwidth and Gain
Four classic criteria for evaluating an oscilloscope measuring system are sensitivity, linearity, gain flatness, and bandwidth. In modern high-performance oscilloscopes, problems with sensitivity, non-linear distortion and ringing internal to the vertical amplifier and display circuits have largely been conquered. These issues are no longer a factor. The primary limiting factor that remains, for digital applications, is bandwidth.
For very fast input signals an inadequate bandwidth in your measurement instruments will, at the minimum, distort measured rise-fall times, skew timing measurements, and under-represent the extent of ringing problems. At worst, it can cause you to miss important features of the signals under test. Without adequate bandwidth, narrow pulses, glitches and other effects can go unnoticed and untreated.
Given a scope's rated bandwidth, you can estimate it's characteristic 10-90% rise/fall time of the scope (Table 1). If the rise/fall time of your scope is at least three times faster than the rise/fall time of your logic, expect to see little measurable distortion in any observed waveform. If the rise-fall time of your scope is comparable with the rise/fall time of your logic, expect to see a substantial deterioration of observed rise/fall times, but few other deleterious effects. Don't use a scope with a rise/fall time slower than the rise/fall time of your logic.
|3-dB Bandwidth||6-dB Bandwidth||RMS Bandwidth|
|.339 / BW3dB||.429 / BW6dB||.361 / BWRMS|
All commercial probes come with a bandwidth rating. The conversion from bandwidth to 10-90% rise-fall time is, depending on the form of bandwidth specification the same as for an oscilloscope (see table 1). On a high-end scope (one for which you purchase the scope and probes separately) you must then combine the scope rise-fall time and the probe rise-fall time to get an accurate picture of how the whole instrument will perform. The formula for this combination is:
toverall = sqrt(tscope2 + tprobe2)
As you can see, a 500-MHz scope and a 500-MHz probe does not a 500-MHz instrument make. For best results, plan for a combined overall rise-fall time from your measuring instrument that is 3 times faster than the signal you wish to observe.
When purchasing probes, you will note that, due to the transmission-line effects inherent in the capacitive-input style probe, they are generally not made with a bandwidth rating higher than about 500 MHz. The FET-input probes are limited today to around 1 GHz. Resisitve-input probes are available with bandwidths as high as 10 GHz.
If you are interested in working with very low-level signals (for instance, in fiber-optic receivers) then the probe gain is going to become important. All three probe styles introduce signal loss.
The capacitive-input probe, as depicted in figure 1, has an attenuation ratio of 10:1 (-20 dB). If your scope has a minimum input sensitivity of 1 mV/div, then with this probe your effective minimum input sensitivity will be 10 mV/div. Popular FET probes have an attenuation ratio of about 20:1 (-26 dB). It's fairly straightforward to build a tiny FET amplifier this way and then boost the signal back up at the scope. Insisting on 1:1 performance at the probe level would require additional stages of amplification. The 1K resistive-input probe also has an attenuation ratio of about 20:1 (depending on the exact resistor values used).
Sensitivity To The Probe Ground Wire
Capacitive-input probes, and to a lesser extent FET-input probes, sometimes perform poorly when connected to drivers with low source impedances. This effect is greatly exacerbated by the presence of any significant length of ground wire between the sensing end of the probe and the board.
This effect can be described analytically by looking at the driver source impedance, the probe input capacitance, and the ground-wire inductance as an R-C-L series resonant circuit. Let's analyze all three probe types this way, assuming use of a 6" ground wire (about 200 nH).
For the 10-pF capacitive-input probe (with a six-inch ground wire), as the drive impedance drops below 100 ohms, the probe develops a nasty resonance at about 110 MHz. This resonance is right in the heart of digital territory, and is the primary reason why ground wires are not used in conjunction with 10-pF style probes when attempting to make accurate measurements.
The resonance in the 1-pF FET-input probe (with a six-inch ground wire) becomes evident at an even higher impedance level, which is a worse problem for low-impedance digital circuits. The resonance in FET-input probes begins developing at a drive impedance of 300 ohms, but fortunately it is shifted up to about 350 MHz. You won't notice it unless your circuit rise-fall times are 3-nS or faster.
The resistive-input probe (with a six-inch ground wire) doesn't have a resonance. It's first-order circuit parameters form an R-L network, which doesn't ring. To first order, this circuit is always damped. That's one of the nice things about it: a resistive-input probe is less susceptible to ground wire length than any other probe style.
Figure 5 shows how this information translates into the time domain. In figure 5, we show the same signal, measured four different ways. The probes were applied one at a time, and the results stored, scaled and time-shifted to fit the display. In the figure, all four waveforms clearly show a 37-MHz clock. If that's all the detail you need, then the waveforms are essentially identical. If, on the other hand, you have been chasing glitchy bus ringing problems and need to quantify the undershoot , the differences are substantial.
The top trace was taken using the FET probe, with no ground wire. In the absence of a ground wire (that is, with the shiny metal probe ground barrel directly connected to the PCB ground using a wire not longer than 0.100"), all three probes gave the same result. In that sense they all performed reasonably well (except for the 200-pS timing shift noted above under "probe loading"). Since they are all practically the same, only one, non-ground-wire picture is shown.
The bottom trace shows a capacitive-input style probe with a six inch ground wire, rated at 8-pF and 500 MHz. This configuration has a resonance at 125 MHz, which shows up clearly in the figure as an 8-nS ripple. With a 6" ground wire, this probe is not suitable for fast digital work.
The second trace from the top shows the FET probe with a 6" ground wire. The resonance in this case lies at about 350 MHz, which shows up as a noticeable, but smaller, 3-nS ripple.
The third trace from the top shows the resistive-input probe with a 6" ground wire. This probe is clearly the least sensitive to ground wire distortion.
When probing low-impedance circuits, the capacitive-input probe is highly sensitive to ground-wire length, an FET probe less so, and the resistive-input probe performs best of all.
How to Make A 1K-Ohm Probe
The 1K-ohm resistive-input probe cheap, easily constructed, and remarkably effective up to 1-GHz. If you want to build some yourself, here are a few tips.
For reasonable performance up to 1 GHz, use a 1-m piece of RG-174 cable for the connecting cable. Terminate the scope end of the cable with a BNC connector, and solder a 1/8W-1K carbon-composition or carbon-film resistor to the center conductor of the sensing end. Dress the braid at the sensing end for soldering directly to the PCB ground plane.
I have seen some engineers who like to solder a dozen or so resistive-input probes on a board, and then connect them to the scope in various combinations as needed. They like this approach because the probes stay put and can be operated hands-free. Alternately, you can adapt this probe for free-roving operation by tacking a solid ground wire onto the end of the RG-174 ground braid. A number of ground-wire attachments made for other probes can be adapted for use with a resistive-input probe. On the end of the 1K-resistor, try applying the crimp-on center-conductor contact from a male BNC connector. It makes an excellent permanent plated tip. In this form, the shop-built probe works well up to 1 GHz.
As you move toward 10-GHz, you will find that the resistive-input probe is still an excellent choice, but it requires more care in its construction. For example, the 10-GHz probes offered by Tektronix use an exquisitely crafted multi-braided low-loss coax, gold-plated SMA connectors and very nice, long, skinny 1K resistors. These features extend the useful range of the probe easily into the 10-GHz region.
The resistive-input probe presents a flat 1K-ohm impedance across the band, all the way up to about 1 GHz. Above that, the input impedance begins to roll off, dominated by the approximately 1/6 pF unavoidable parasitic capacitance that shunts end-to-end across the 1K sense resistor. Using two 1/8-watt, 470-ohm resistors in series instead of a single 1K will reduce the parasitic capacitance, improving the roll-off characteristics by a factor of two. Also, pay attention to the positioning of the sense resistor. It should be kept up off the board, away from the ground plane. When pressed down near a solid ground plane you will pick up another 1/2 pF of parasitic capacitance to ground, substantially affecting the probe's performance. Kept 1/2" or more away from ground this effect will be negligible.
The resistive-input style probe incorporates a fixed degree of signal attenuation. This is not usually a problem, assuming that your scope has adequate vertical sensitivity to make up the difference. As described above, the resistive-input probe provides a 21:1 attenuation ratio.
If you need to make exact measurements, calibrate your resistive-input probes. Being made from carbon-composition or carbon-film resistors, they may not be too accurate. If you order up a batch of custom-select 950-ohm carbon composition resistors, you can tune in a more precise 20:1 ratio. Beware the temptation to use a 1% MF resistor at the tip unless you are certain of its construction—many MF resistors incorporate an internal serpentine pattern in the metal film that will destroy the probe's high-frequency properties.
Other Practical Issues
Now we get down to some of the issues that can make or break your day. Things like flexibility of the connecting cable, size of the probe head, and cost. Here are some practical factors to think about:
- Will the probe fit between the cards in your chassis? It had better, because most truly fast bus systems won't function with extender cards, which add too much bus capacitance and screw up critical clock timing. Probes need to be squeezed between cards, with a right-angle bend at the tip. The shop-built resistive-input probe is a good candidate for this type of abuse.
- Will it stay on your bench (or get stolen)? If you have invested in something nice, consider taking defensive actions to protect your property. I've seen more than one really good probe with a little tag on it saying: flaky connector—do not use. In this respect, the shop-built 1K probe takes the cake; it's truly ugly.
- Will the probe help you meet higher-ups in the organization? Only the FET-input probe meets this requirement. Try ordering fifty of these, and you'll get to meet plenty of higher-level executives while they grill you about the cost.
In high-speed system developments, the ubiquitous 10-pF 10:1 capacitive-input probe is no longer adequate. The two alternatives are the FET-input probe and the resistive-input probe.
Of the two, the resistive-input probe is cheaper, it has as good or better bandwidth, and it is more tolerant of long ground wires. These advantages come at the cost of higher IOH required in your digital circuits in order to drive the 1K resistor. In modern high-speed systems, because the extra drive current is almost always readily available, the resistive-input probe makes a lot of sense.
As we go higher in frequency, the FET-input probes will run into increasing difficulties. At signaling rates faster than about 300-pS rise-fall, only a resistive-input style probe can maintain a high enough input impedance to remain useful.
Comments invited: email@example.com