Terminator Crazy
Have you ever had to add terminators to a high-speed board in order to get it to work? If your answer is "yes,'' do not despair—you are not alone. In my high-speed design workshops, I hear engineers constantly gripe about terminators. Why are terminators so frequently needed? The root cause is the fantastic improvements in rise/fall times we are now enjoying in modern logic families.When I started working with digital logic, a 10-ns rise time was considered fast. Today, the situation is completely different. Now there are logic families that can transition ten or twenty times within 10 ns. For example, the output rise time specified on the PCI bus can be as fast as 500 ps. When working with such fast drivers, a rising-edge waveform has plenty of time to finish its business, roll over, and smoke a cigarette long before it smashes into the far end of a printed-circuit board trace. If the far end is unterminated, the signal will overshoot, reverse course, and then come roaring back toward the driver. At the driver end, the returning signal will bounce around some more, settling down only after making several round-trip tours of the neighborhood. In other words, an unterminated trace with a fast driver will ring like crazy.
Because the drivers available today are getting so much faster, the percentage of traces susceptible to ringing is skyrocketing. To cure ringing problems, most fast boards are sprinkled with terminating resistors. The terminators are selected to have an impedance in the 10 to 100 ohm range, something that (hopefully) matches the actual transmission characteristics of the printed-circuit board. When a fast signal hits a terminator, it acts like a sponge, absorbing just enough of the signal energy to prevent reflections. Traces with properly engineered terminators don't ring, overshoot, or undershoot.
How do you know when a terminator is needed? The ratio of trace delay to rise time is the first clue. Terminations are almost always required when the trace delay exceeds the logic rise time. Many people take an even more conservative approach, installing terminations when trace delay exceeds one-fourth, or even one-sixth, of the logic rise time.
The second most important factor in the ringing equation is capacitive loading. A capacitively loaded line, even when short, sometimes requires a terminator. This happens because the capacitance gives the line an apparent delay longer than the bare unloaded line. When the apparent delay goes up, so does the need for a terminator. At the same time, capacitive loading lowers the apparent characteristic impedance of a short line. Sometimes as little as 10 or 20 pF can make a noticeable difference. The best terminating value is a function of line length, line impedance, and capacitive load.
Every designer should establish clear rules for when terminators are required. These rules put the designer in control, actively managing signal integrity instead of the other way around. The rules should take into account the trace impedance, length, and topology, as well as the driver rise time, its output impedance, the position and size of capacitive loads, and some measure of the tolerance for overshoot and undershoot. Use a trace simulator, or actual laboratory measurements, to double-check all calculations. With a little planning, ringing need not become a serious problem.