Nonlinear Termination Techniques for Electronic Systems

Jeffrey C. Kalb

New termination methods reduce noise and decrease electromagnetic interference in high-speed, low-voltage applications.

High-speed electronics often require signal termination to maintain signal integrity. Signal termination reduces ringing and is also a significant factor in electromagnetic interference (EMI) control. This article discusses the termination issue and the benefits of new methods for termination.

The theory behind transmission-line behavior and termination techniques is straightforward and works well for cases in which the signal path is well controlled, such as in coaxial cables. Unfortunately, modern electronics are filled with nonideal signal paths, poorly controlled impedance paths, multiple paths, etc. Nonlinear termination techniques help to overcome these adverse conditions, improving signal quality and reducing EMI.

As clock and bus frequencies soar, providing proper transmission-line termination has become a major issue in electronic systems. A transmission line in this context is essentially any signal path—whether a trace on a printed circuit board (PCB) or a discrete wire—where the length is long relative to the wavelength of the frequency at which it is being operated. When signals propagate down such a transmission line and encounter impedance mismatches along or at the end of the line, portions of the signal are reflected back and forth along the line, causing ringing or noise generation. This ringing, in turn, causes additional radiation at a frequency related to the length of the transmission line.

Ringing is caused when more energy is introduced onto a wire than is necessary to achieve the desired transition. The classical response to this problem is to terminate a transmission line with either a series resistor at the driving end (to limit the energy introduced) or with a parallel termination at the end of the line to absorb the excess energy. Such approaches, however, either slow the signals because of the reduced drive, or greatly increase the power dissipation of the system by requiring a constant load current at the receiving end.

These termination approaches also produce secondary effects such as changes in the required drive capability of the circuits, wiring restrictions, or the need for extra components. In addition, such approaches require costly, design-limiting control of the transmission-line impedances. Consequently, designers prefer wiring systems that do not require terminations.

One way to avoid termination problems has been to use clamp diodes, a form of nonlinear termination that was first introduced in the late 1960s. Nonlinear termination devices provide termination outside of the normal-signal operating range to retain a high characteristic input impedance (and thus a low power consumption) within the operating range without slowing the signal. The idea with this system is to allow for some slight overshoot on a signal transmission before absorbing the excess energy. This solution entails slightly longer settling times on the transmission line, but it is a system engineers have accepted. In some ways, it can be thought of as an ease-of-use concept. As operating voltages have dropped, however, clamp diodes have become ineffective because the voltage drop of the diodes doesn't scale with the system operating voltage.

The Theory

A fundamental concept behind terminating transmission lines properly is the notion that the current and voltages involved must reach equilibrium on both ends of the line. Achieving this can require one or more current-voltage (I-V) transitions up and down the line that evidence themselves as voltage and current changes or ringing. No method is specified for achieving this required stability. Stability could be attained through traditional impedance matching or some other technique. Achieving a balance through the use of nonlinear devices outside of the normal operating range is a viable alternative.

 

Figure 1. General circuit configuration.

Figure 1 shows the general circuit configuration applicable to all of the following discussions. It shows a circuit driving a transmission line with a receiving device on the other end. The dotted box shows the application of a clamp or another form of termination at the receiving end.

Transmission lines—with all of the parasitic effects of attached devices, connectors, and so on—can be extremely difficult to model or simulate. However, a fairly simple technique can be used to approximate the results graphically. Such a system can effectively address the nonlinear impedances sometimes represented by the input and output of devices.

 

Figure 2. Typical x-y plot of voltage and current for a traditional CMOS or TTL device without special termination.

Figure 2 shows a typical x-y plot of voltage and current for a traditional complementary metal oxide semiconductor (CMOS) or transistor-transistor logic (TTL) device without special termination. The voltage scale was chosen to coincide with the voltage of the supplies being used, and the current scale was consistent with the magnitude of current that would flow if a resistor equal to the nominal characteristic impedance of the transmission line were put across the supply terminals. In this case, a 5-V supply and a 50-W transmission line were used. Hence, the current scale goes to 100 mA. On this chart, the output characteristic of the driver in the output low state and the input characteristic of the receiver gate have been drawn (both within and outside the normal operating range).

An analysis and discussion is presented here that relates specifically to a waveform falling from a higher voltage to a lower voltage. But an analogous discussion could address a rising waveform using the relevant impedance characteristics. It is important to note that voltages are shown as traditional plus or minus figures. Currents are defined as positive if they flow from the receiving end to the driving end of the transmission line.

 

Figure 3. Electrical activity on the transmission line.

The upper portion of Figure 3 shows a voltage-current diagram that allows an analysis of the electrical activity on the transmission line. The lower portion of the diagram shows the resulting voltages on the driver and receiver ends as the waveforms reflect back and forth along the transmission line. Each square along the time (t) axis on the bottom graph represents one trip of a current-voltage waveform down or back on the transmission line.

For a 50-W transmission line, if the output is initially high at 5 V and the output of the driver turns on (output going from high to low) with the characteristic impedance shown, the resulting voltage and the current waveform that will propagate down the line can be derived graphically. Drawing a 50-W impedance line from the start point to intersect with the output impedance curve of the driver (point A) gives the voltage and current waveform. In this case, the output voltage initially pulls down to about 0.9 V, and a current waveform of ~80 mA is sent down the transmission line. That transition can be viewed in the lower portion of Figure 2 when the drawing is turned on its side. On the initiation of the driving pulse, the output voltage of the driver drops from 5 V to the voltage of point A. One transmission time later (one square), the wave will have propagated down the line, and the receiving end will change in response to this stimulation.

However, the equilibrium point—the balance between the voltage and current—at the receiving end is not the same as that on the transmission line or at the driving end because the impedances are different. The equilibrium point at the receiving end can be derived by drawing a –50-W impedance line from point A to intersect with the impedance curve of the receiving gate. So, the transition from 5 to 0.9 V as seen at the driving end will be seen as a transition from 5 to –2.5 V (point B) on the receiving end. Of course, this equilibrium point is now unstable relative to the driving end.

Because two different voltages cannot exist on the same wire, they must equalize. This stabilization process is initiated by the generation of a reflected wave, which is sent back toward the driving end. Drawing another 50-W transmission line from point B to intersect once more with the driving impedance at point C shows that the output of the driver has been driven to about –0.3 V. This change still does not produce an equilibrium point for the whole system, so another reflected wave is sent back down the line to the receiving end. Each transition is shown in the drawing at the bottom of Figure 2 at the appropriate time interval.

When the reflection from the output (point C) gets back to the receiving end (draw another –50-W line to the intersection of the impedance curves), a problem arises because the new reflection takes the receiving end up into 0.9-V range (point D). This result exceeds the worst-case noise tolerance level of the receiving device and could cause an erroneous signal to be generated. In the presence of small inductances in series with the various lines, or with a poorer clamping voltage on the receiver, these effects can even be worse.

Figure 3 illustrates the normal input characteristics of the receiving device supplemented by placing a Schottky diode parallel with it to clamp the undershoot. Notice that although the current waveform in the first transition is the same as in Figure 2, the clamping action of the diode changes the equilibrium point at the receiving end so that the input voltage does not become as negative. Hence, the energy reflected back down the line is reduced. This reduction enables the receiving device to avoid the damaging positive-voltage reflection seen in the previous example.

National Semiconductor introduced the transmission-line control feature into TTL circuits in 1968, which fundamentally changed the wirability of these circuits. This control made large systems of TTL circuits practical. Since then, such diodes have been part of bipolar logic circuits. They are implemented partially by the ESD protection diodes in CMOS circuits.

 

Figure 4. Electrical activity on the transmission line, in which a resistor is inserted in series with the Schottky diode.

Figure 4 represents a special case of the characteristics in Figure 3, in which a resistor is inserted in series with the Schottky diode. The resistor value is selected so that the equivalent impedance at the point of equilibrium matches the impedance of the 50-W transmission line. In this situation, the current reflected down the transmission line from the receiving end is exactly the amount necessary to settle the transmission line in two transition times—one back to the driver—and one down to the receiver.

In Practice

Although this alternative may not be practical in a real-life situation (it requires that the full characteristics of both the driver and transmission line be understood), it leads to an interesting realization. If the combined characteristic impedance of the devices on the receiving line outside operating range is greater than the transmission-line impedance at the initial point of equilibrium, the system will sustain reflections that ultimately cause a positive reflection into the operating range of the receiving device. These reflections could cause noise problems.

However, if the combined characteristic impedance at the receiving end at the initial equilibrium point is lower than that of the transmission line, no positive reflections will result. The closer the match between the transmission-line impedance and the termination at the point of initial equilibrium, the faster the system will settle.

 

Figure 5. CMOS with Schottky diode.

Achieving this match is more complicated than it may seem. First, the equilibrium point is a function of the all the conditions that went before it, especially the initial starting voltage of the system. Perhaps more importantly, the effective impedance of the diode varies with current. And because the operating range of the Schottky diode below 0.5 V shows extremely high equivalent impedance, equilibrium points in that region will cause undesirable reflections (see Figure 5).

In this case, the operating voltage has been dropped to 1 V (3.3-V operation is the approximate crossover point for the effectiveness of good Schottky diodes). The equilibrium point at the receiving end of the transmission line occurs in the high equivalent impedance region of the diode, leading to a proportionately high level of reflection down the transmission line. Such a situation could be catastrophic. Because the diode impedance is so high in the low-voltage region, the reflection problems are proportionately higher than before the introduction of clamping diodes on TTL circuits.

The semiconductor industry has recognized this problem, and designers have been moving toward resistor-terminated systems. Using such systems comes at an expense. Many components must be added. They also require a lot of additional power dissipation, which is particularly troublesome in portable or other small systems. Achieving precise termination is also difficult to engineer at the board level. This is especially problematic when the bus lines in question involve multiple drops along the line and uncontrolled line impedance, as well as sockets with varying numbers of loads that change the effective impedance layout constraints.

 

Figure 6. A nonlinear termination method with low-power diode-termination benefits at low voltages.

Figure 6 shows what happens with a new nonlinear termination method, which has low-power diode-termination benefits and still works at low voltages. With this method, a CMOS circuit is used to create the nonlinear termination (see Figure 7 for the equivalent diagram). The resulting I-V plot is shown as the active clamp in Figure 6. Within the active region of bus operation, these devices may be somewhat conductive, perhaps as much as 2 or 3 mA. However, compared with the ±20 mA of current that a terminating resistor would consume (even in a low-voltage situation), this is extremely low. As soon as the transmission-line voltage reflects outside of the operating range, the termination circuit turns on and provides the necessary low impedance, protecting the low-voltage region of operation. As a result, the damaging positive reflections are eliminated.

This approach works at essentially any supply voltage, but is most effective below 3 V. The technology used to implement the termination circuit, which is the same used in an equivalent level of CMOS technology, scales along with the traditional semiconductor scaling.

 

Figure 7. A CMOS circuit used to create nonlinear termination.

Such systems save a lot of power, and are relatively insensitive to the absolute impedance level of the bus in question. This minimizes costs by reducing the number of boards and decreasing engineering design time. This termination technology can be placed directly on the integrated circuits, which eliminates the space, cost, and reliability issues associated with additional external components. The simplicity of system design associated with earlier technologies is recaptured.

Nonlinear terminations can be extremely useful in applications such as memory buses, where a number of cards might be inserted onto the bus. It is difficult to determine the correct termination impedance in such applications, and the impedance varies with the number of memory chips or modules. Nonlinear termination is essentially self-tuning. The nonlinear terminator adjusts as other loads on the transmission line do more or less clamping, loading, and terminating. Because there is no need to terminate in a power-consuming resistor network, the reduced load enables chip designers to use smaller drive circuits and full rail-to-rail voltage swings to get greater inherent noise margins.

An often-overlooked aspect of the ringing and noise problem is the generation of EMI. The frequency of ringing on a transmission line is determined by the length of the transmission line rather than by the system's operating frequency. Therefore, every wire in a system could radiate at different frequencies. Controlling the ringing—either through traditional resistor terminations or through nonlinear terminators—minimizes this noise generation.

In theory, however, it could take nonlinear termination a bit longer to stabilize the signal. For designs with tight-tolerance termination resistors, tight impedance control on the transmission lines, and a clean layout with no impedance discontinuities, designers can achieve a cleaner signal using resistor termination. But, in practice, this may not be the case. Impedance-controlled PCBs could vary by 10 or 20%, even with well-designed layout requirements. In the best case, resistor termination would be better. For worst-case analysis, a nonlinear terminator may actually be a better option.

One of the most effective uses for this terminator would probably be found on memory buses, such as those used on double-data-rate dynamic random access memory. For these buses, the loading characteristics vary depending on the number of devices inserted into sockets, particularly where the voltage swings are likely to be high for the standard parts. If robust, easy-to-use designs are required for all but the most tightly tuned transmission lines, nonlinear termination could be the optimal solution.

Conclusion

The concept of using nonlinear terminators on high-speed logic circuits has been around since the late 1960s. They were the termination option of choice for all bipolar digital circuits. With the advent of lower-voltage CMOS circuits, these diode techniques became impractical. As a result, higher-performance systems have come to rely on the use of termination resistors to improve signal integrity, which has led to increased system complexity, design complexity, power consumption, and cost.

The availability of a new method of nonlinear termination makes it possible to regain simplicity and cost-effectiveness in the higher-speed, lower-voltage systems that are now common.

Jeffrey C. Kalb is the former president and chief executive officer of California Micro Devices (Milpitas, CA). He currently serves as a technical advisor and is a member of the company's board of directors. He has held many senior-level positions managing complex semiconductor and system environments. He can be reached at 408-263-3214.

Back to July/August Table of Contents