Value of End Terminator
Doug Brooks writes:
I heard an interesting theory the other day that I would like your comment on:
- Parallel [end] termination of a trace whose characteristic impedance is 50 Ohms
- It is known that any terminating resistor in the range of 40 to 70 Ohms will give acceptable circuits results.
The optimum termination is 70 Ohms because that will result in the lowest current around the loop area and therefore the best EMI performance.
On the surface, this seems to make sense. Do you have any feelings about this pro or con?
Thanks for your input,
Thanks for your interest in High-Speed Digital Design.
Let's look at the effect of the end terminator on a truly long transmission line, a line for which the round-trip delay exceeds the driver rise time.
First, a small quibble about the numbers, and then on to your main point. The theory about "acceptable results" depends on the details of your voltage margin budget, etc., for the trace. A mismatch of 70/50 ohms at the end of the line, for example, would generate an initial reflection of one part in six (0.16), which means that for most logic families you may not get first incident wave switching. You'll have to wait at least one extra round trip time before you begin the next bit in order for the residual reflection (0.16) to dissipate. The extra round trip time might be OK, or it might not, depending on where your receiver is located along the line and when you need to sample the results. If you are trying to clock really fast (more than one bit per line delay) then the one-sixth reflection may cause difficulties and you may need a match better than 70/50.
Anyway, let's not worry too much about the exact numbers, because your concept is still interesting.
Here's what happens. Let the driver be LOW for a long period of time. The line is now at zero volts everywhere, and the current in the line is defined by the end termination. If the split terminator at the end is composed of two resistors (suppose 140-ohms each), then the steady-state current in the low state will be -Vcc/140 (sinking).
Now suppose the driver wishes to impress a FULL-SIZED positive edge into the transmission line. If it's a long line (round-trip delay greater than the driver risetime), the initial delta-I current step required has NOTHING TO DO WITH THE LOAD. If you draw out a V-I curve for the driver, at this point in time the driver perceives that it is driving a 50-ohm resistive load with a DC current offset of negative Vcc/140 amps at zero volts. A voltage step of Vcc requires a current step of +Vcc/50. The initial-state current is -Vcc/140, so the current just after the driver finishes it's rising edge will be -Vcc/140 + Vcc/50 = .0128*Vcc.
OK, now re-do the numbers assuming a 50-ohm split termination (100-ohms each to Vcc and 100 ohms to ground). The initial current is now -Vcc/100, and the current step required for a full-sized initial step is Vcc/50, so the current just after then first rising edge is -Vcc/100 + Vcc/50 = .0100*Vcc.
Notice that the delta-I is defined by the delta-V required at the front end of the line, and the impedance at that point, which is 50-ohms in both cases. The delta-I in both cases is therefore the SAME (Vcc/50). The short-term EMI effects will therefore be the SAME.
Notice that the PEAK current required in the 70-ohm case is 28 percent WORSE than the PEAK current required in the 50-ohm case. Because the PEAK current has to do with the sizing of the driver I/O cell required to achieve a full-sized output voltage on the first stroke, mis-terminated lines can actually require bigger drivers than properly terminated ones (assuming you have a very long line and you still want first-incident-wave switching at every point along the line).
On a long transmission line, I view the continuous sinking current required by a split terminator as a very helpful thing. After holding the line low for a protracted period, if the driver just "lets go" of the line (i.e., the bottom half of the totem pole stops driving), the line pops instantly up to 1/2 value. From that state the driver need only lift the line up the other 1/2 way to produce a full voltage swing.
The use of a proper end termination serves to minimize the worst-case PEAK current requirements on the driver. A larger end-termination value, while it provides less AVERAGE current (and thus saves power) has the disadvantage of INCREASING the PEAK current required from the driver during the first round-trip time. If your driver can't pump out the extra current, you won't get a full-sized first incident wave on a long line. (note: it may look OK at the end where you see some initial overshoot, but it won't be OK at the front end or in the middle -- this property leaves open some wiggle room for implementing "weak terminations" if all you need is a point-to-point line).
The same general conclusions hold for the transition from HIGH to LOW.
You can see the effect I'm talking about on any signal integrity simulator. Just set up a long line end-terminated with a 50-ohm split terminator (100/100). Carefully measure the V[oh] and V[ol] levels right out of the driver. Now change the terminator to a 70-ohm split (140/140) and re-take the measurements. During the first round-trip time the driver in the second case won't perform quite as well.
OK, now let's look at the short-line case. On a shorter transmission line, one in which the un- terminated line displays only a mild amount of ringing, the driver and end-terminator are more intimately coupled. During the period of the rising edge information returns from the end termination that modifies the preceding argument. In the short- line case you will find that the use of a higher termination value will indeed provide the general benefits you describe.
In conclusion, for short lines I have to agree it's best to pick the largest value of termination that guarantees adequate signal quality.
For long lines, however, you need to check your voltage margins carefully at every point of interest, and with worst-case component values, before concluding that 70 ohms is "always OK". Even then, it might not reduce your EMI.
There is a related article I worked on with Bruce Archambeault for my regular EDN column on Signal Integrity. It appears in the Mar. 01, 2001 issue, titled "Reducing Emissions". Bruce's article concerns this same general subject, but addresses the issue of series terminations.
Dr. Howard Johnson