Above Board - Archive II

This issue of Above Board begins a series of discussions concerning design issues that were written by the designers at Ultracad Design in Bellvue, Washington. 1(206)450-9708

Contents

Frequency vs Rise Times
Transmission Line Issues I
Transmission Line Issues II

Above Board - Autumn 95

FREQUENCY VERSUS RISE TIMES

A service bureau once told a client it wasn't necessary to worry about 'good design practices' because their frequencies were not high enough to create any problems. If your service bureau tells you that, you're headed for serious problems! The issue is not frequency - it is wave shape and rise time. A 5 volt peak-to-peak 10 Mhz clock line, for example, has many harmonics, one of which is a 450 mv 110 Mhz signal! Another is a 225 mv 220 Mhz signal(and there are many others). A pulse with a 1 nsec rise time has a strong 300 Mhz frequency component. It may not be obvious that these high frequency components are there. But we've seen many unfortunate situation where companies ignored them and then couldn't understand why their boards were so noisy and why they were having so much trouble with FCC compliance.

This article discusses some basic relationships between wave shapes, rise times and frequency harmonics and why it is critically important to design your boards with them in mind.

Square Waves

A square wave can be thought of as a combination of a series of sinusoidal waveforms that are numbered harmonics of the square wave fundamental. They are related in frequency and magnitude by the following relationship : cos( t) - cos(3 t)/3 + cos(5 t)/5 - cos(7 t)/7+... (etc)

Thus the 5th harmonic is one-fifth the magnitude of the fundamental, etc. A relatively low frequency can have some very high, strong, harmonics that need to be dealt with.

Rise Times

We don't often think that the rise time of such a pulse can cause special problems. But rise time follows almost exactly the rising edge of a sinusoidal wave-form. It follows that a pulse with a 1 nsec rise time might generate a brief 300 Mhz transient of the same peak-to-peak amplitude.

Effects

Seemingly low frequency signals can generate powerful harmonics that are surprisingly high in frequency. Most ICs designed to work with such waveforms can handle them. In fact we often go to some extent to preserve and increase these harmonics (in an attempt to keep clean waveforms! It is ironic that 'clean' wave- forms generate dirty noise problems). So one of our design problems is how to get the signal from one IC to the next without radiating these harmonics .. first to the other signals we want to protect and secondly to FCC compliance measuring devices outside your system.

These issues are important for all circuits as radiated energy is a function of power which is a square function of current.

Above Board - Winter 96

Transmission Line Issues (Pt 1)

Circuit Board Structures

Transmission lines consist of wires placed in a defined relationship with a ground plane. Generally, there are five types of structures that need to be considered :

Microstrip - a typical trace on the surface of a board that is placed over a power or ground plane.

Buried or Embedded Microstrip - Similar to a Microstrip but is covered with a coating such as a conformal coating. The effect of the coating is hard to estimate, but it is the function of the thickness of the coating and the relative dielectric coefficient of the coating material.

Stripline - A trace placed symmetrically between two power planes.

Dual Stripline -Similar to a Stripline but with two trace layers between the planes.

Unbalanced Stripline - similar to a Stripline but where there are two or more signal layers that may not be symmetrical.

Formulas for intrinsic impedance and propagation delay as a function of the type of structure are reasonably accurate for Microstrip and Stripline, but are much less accurate for the other structures. Some resort to using "rules of thumb" for adjusting the formulas in the other configurations.

Typical guidelines

Many designers are uncertain about applying high speed design rules. These rules become important when the high frequency components of the signals reach sufficiently high levels. The problem is that "how high is high" is not clearly defined. Some argue that all frequencies for logic families used in today's systems are high enough. In the last Above Board, we noted that the thresholds can be surprisingly low, since the critical issue is RISE TIME, not frequency.

The rise (and fall) time is one of the primary factors in determining when it is necessary to treat traces as transmission lines. Standard guidelines suggest that one should use transmission line design factors if "the two way delay of the line is less than the rise time of the pulse" or if line length > tr/2tpd where tr is the rise time of the pulse and tpd is the propagation delay of one inch of line. But this rule is not as safe as it may sound.

Guideline Problems

1. This is not the length of line at which there is no effect. It is the length at which the effect is tolerable under some definition and some set of conditions.

2. Propagation time is a complex function of many factors including Er , loading, impedance and board construction.

3. Designers may assume that this issue only arises for ECL circuits, Other types of logic may also need special consideration when routing.

Many designs perform well until suppliers are changed or the chip was improved. It can take a lot of time to figure out why a good design suddenly turned marginal.

Above Board - Summer 96

Transmission Lines (Pt 2) Design Steps

In designing for transmission lines, the following steps are recommended:

1) Determine if the routing seems prudent for the circuit and components. Problems with the standard design rule has resulted in the use of one that is more conservative by a factor of five:

If the line length > .1*(tr/tpd)

2) Address line termination and load. Ignoring termination and load can result in poor reliability later.

3) If using series termination, exercise extreme care with stub and distributed load routing. This is due to the effect of the reflected signal from the far end of the transmission line.

4) Parallel routing also requires attention to stub and distributed load routing. Each load adds some capacitance which;

a) lowers intrinsic impedance of the line

b) slows down the signal propagation delay down the line

c) causes some reflection down the line.

If transmission line routing seems prudent, the general design rule for maximum stub length is the same as described above. A good practice is to design for the absolute minimum or zero stub length. Even in the tightest circumstances it should be possible to rout to 3/16" stub length.

This can be achieved by spending time in planning the design and especially in planning for component placement and manual routing of all critical clock and logic paths.

5) The intrinsic impedance of a line can be adjusted by changing the width, and is often done to adjust for uneven loading . Design splits can also be incorporated into the line (i.e. stubbing two 100 parallel lines onto the end of a 50 line)

6) Dead End Stubs : Never allow a stub trace to exist without a terminating point. Such a trace is an antenna and its uncontrolled impedance can cause signal reflections whose results will be absolutely unpredictable (but will never be positive).

7) Adding trace delay times: It is not uncommon to adjust the timing delay by adding to a trace. Care must be taken in how the excess trace length is routed. A trace that extends beyond the pad and then doubles back, one that turns at a right angle or "T" junctions can act as antennas. Probing a board with an EMI detector will indicate that almost invariably the strongest radiation will come from 90 degree corners and "T" junctions. Remember, antennas work both ways, if a stub or corner emits well, it also receives well. So these are the points where noise can be injected into the board.

8) Signal Return Paths : Every signal has a return path. The higher the frequency (including harmonics), the closer the return path will be to the signal trace. If you do not provide for a return path, the signal will return by some path anyway - controlled or uncontrolled.

9)Power/Ground Planes : The best provision is a power distribution plane directly under the signal. With very high frequencies where there is a ground plane under a trace, studies have shown that the signal will return DIRECTLY under the trace. Anything that breaks the continuity of the ground plane under the trace will cause the return signal to deviate from its course. Thus the path that the return signal is taking could act as an antenna where you might least expect it to do so - on the ground plane.