What is the difference between high speed digital circuits and the others? The high speed ones aren't digital!

The higher the clock rates and the shorter the rise and fall times in digital circuits become the more these circuits escape the rules of the digital domain and enter the analog domain. Not just the analog world, but that one of microwave technology. Here, there are no wires any more, but transmission lines; no shields, but damping factors and antennas and of course, a completely different time scale. In other words: Physics at it's best.

This article wants to introduce the newcomer to the very basics of high speed designs (not necessarily only digital ones, but mostly them). It is definitely not intended to substitute any good textbook or training.

When designing with clock rates above app. 100MHz or with slew rates less than 5ns there are some pitfalls to avoid. Because at these high frequencies a normal wire or trace on a board behaves differently as usual. First, high speed devices need a closer specified timing. The relevant data can be found in the data books and is not subject of this article. Neither is the calculation of the loads of a driver and the associated slowdown of the signal it sources. These are all easy to understand and should pose no problem to an experienced designer. But there are some other aspects one can easily overlook:

• Wires and traces

  The higher the frequency components of a signal are the more the rules of transmission lines apply. A transmission line is a theoretical construction to describe the real effects on real lines. The transmission line approximates the reality by defining a schematic for the physical wire or trace. In this schematic you see resistors, capacitors and inductors. But one thing the transmission line theory does not address is radiation. Neither incident or emitted. But for most of the cases it is a very powerful construction to be able to understand the effects.

  Without going into much detail you can view a transmission line as a infinite chain of infinitely small links, each of them consiting of a resistor and an inductor in series to each other and in series to the flow of the signal and a capacitor from the signal line to ground. These tiny elements (links) are evenly (hopefully) distributed across the whole wire/trace. Let us make a gedankenexperiment and let us connect an ideal impedance analyzer via cables of zero length to one of the links. What we will observe is a funny behaviour (or not funny at all): The impedance of this construction varies greatly over the frequency. But when we connect the links to form a transmission line it shows a different behaviour: Without the terminating resistor at the end the impedance varies periodically between zero and infinity. But with a resistor of a specific value at both ends the impedance remains stable from DC to the highest frequency.

  But that is not all. When we apply a step impulse to the input, it "ripples" through all the links towards the output. When our terminating resistor is not there we will observe one effect, called reflection. The reflected impuls travels back to the other side where it interferes construtively with the original step impulse. Therefore we measure twice the originally supplied voltage here. So, normally transmission lines are terminated by their characteristic impedance to avoid these reflections. But sometimes this is rather cumbersome, and the reflections are tolerated, even necessary for the proper working of the circuit; as an example take the method of reflective wave switching as compared to incident wave switching (i.e. terminated line).

  So, we learned that it is normally advisable to dampen the reflections. How do we do that? With the magic resistor at the output that made the impedance so neatly stable. The same resitor will completely dampen the reflection without disturbing the original signal. It's value is determined by the characteristics of the transmission line, namely the values of the inductors and capacitors in the links. For the calculation we can use a single link or the whole transmission line - the result is always the same: The ideal value of this resistor is pure ohmic (real) and can be calculated by this formula: R = SQRT ( L / C ), where SQRT is the square root operator. The resulting value is the so called "characteristic impedance" of the transmission line. It is independent of the frequency (when we neglect secondary effects).

• Antennas

  No, we do not discuss how to best design an antenna for your wireless project, but how to not design an antenna into your wirebound system. Theoretically every AC current that flows through a wire or any AC voltage across two points in space will radiate some energy into the space around. But fortunately these effects can be made almost infinitesimally small. All that is needed is some knowledge in physics and an ideal world. While the latter is not so easy to get, the former is a more practical approach. Even with a nonideal world we can make the bad effects small enough so that we (and the others) can live with them.

  For an antenna to be able to radiate significant energy into free space it's impedance must be in the order of magnitude of the impedance of free space. The more both impedances differ the less energy is radiated or picked up. This is strictly valid only in the "far field" of the antanna, i.e. at a significant distance from it. For distances smaller than a wavelength different rules apply, see below. So, then let us make our lines on the board in a way that they differ by some magnitudes from the impedance of free space and we are done. Sorry, but this solution is not viable. Simply because we cannot make our lines to have an impedance with less than some tens of ohms or more than a couple of hundreds of ohms. The impedance of free space is 374 Ohm, just not high enough to stay well below with a low impdance line, nor low enough that we could use a high impedance line.

  We need to attack that problem from another side. A second factor that determines the amount of energy radiated or picked up is the effective area of the antenna. The attribute effective is the important point here. The effective area of an antenna is usually much larger than the physical area it uses. This is because an antenna "bends" the field surrounding it, extending it's reach beyond it's physical dimensions.

  How do we deterine the effective area of an antenna formed by traces on a PCB? Theoretically we could use a simulator to do the job, but this is a pretty complicated task and usually does not give the results in a time frame that we could live with. So let us make some educated guesses? No, let us use the middle way.

  First, let us observe one fundamental fact: Even when there is an antenna it won't radiate into free space if all of it's radiation is absorbed by some medium. Second, almost all antennas are not unidirectional, but radiate most of their energy into small sectors of space. Third, if two antennas are located very close to each other and radiate the same signal, but with opposite phase their radiation is cancelled in the far field.

  These three facts, together with the effective area give us valuable measures to remedy any radiation (or interference) problem. Let us look at the world with real eyes now:

  Practical stuff

  • Terminate your traces at both ends if it is possible. Definitely possible is it with unidirectional signals. Place a resistor with the right value (see above) in series to the output of the driver and another in parallel to the input of the receiver. If a driver drives more than one receiver then either place the receivers very close together. The distance between the furthest apart should not exceed one or two centimeters. Or run separate lines from the driver to each of the receivers. Each line has the series resistor at the driver and the parallel one at the other end. But check the available output power of the driver. All these lines are in parallel for the driver, so it must supply significantly more power than into a single line.

  • Make your traces with high frequency signals very short or design them to be shielded lines. While the best approach is surely to keep wires as short as possible it is not always feasable. Then take a step forward and succumb to the rules of microwave design. There are two variations of the transmission line here that you can use: the so called "stripline" or the "triplate line". The former is easier to implement but suffers a higher degree of radiation problems. While the basic stripline is just a trace on your board that has a ground plane on the other side, the usual form used in high density designs is to guard the line by ground traces at the sides of the signal trace (in addition to the ground plane on the other side of the PCB or layer on a multilayer). This has two positive effects: First the additional ground traces shield the signal from other signals running nearby. Second they help to control the impedance of the trace. This configuration needs more space than without the additional ground traces, but is superior in terms of crosstalk and EMC. It is always best to prevent the generation of interferences than to shield them. So design your board to be intrinsically EMC compliant. The more sweat you invest here the cheaper the whole product becomes, because you then can use almost any housing and need not to resort to expensive shielding measures.

  • For long runs of high speed signals use drivers and receivers that are either differential or use a very low voltage swing. If you choose the differential version then run both traces closely together and "twist" them a couple of times so that each of them (hopefully) is exposed to the same sources of interference. But in any case, make the area enclosed between them as small as possible. This ensures that they will not pick up excessive noise, nor radiate too much.

    Low voltage swings on signal lines improves the situation with the radiated energy and allows for longer trace runs. But lower levels mean more susceptability to incident radiation. You must exercise great care to prevent interferences from higher power sources. These can come from the power section or from the outside. Good grounding and shielding is a paramount here.

  • Make all areas that a signal and it's return path (or pathes) encloses as small as possible. While this was good design practise anyway for good analog designs, it now becomes important to digital designers, too. Do not blindly trust on your ground plane. If there are other paths for the signal to return to it's source it will use them. The larger the area the higher the radiated energy or interference (effective antenna area!).

  • Consider the application of dissipative materials on your board. These materials can be foils to glue to the board, liquids that are applied by printing or dispensing or solid parts that are soldered or glued to the board. While their application does not come for free, they sometimes save more money than they cost because the housing can be made cheaper.

  • Last, but not least, do not forget conducted interferences. This article focussed on some tricky aspects of high speed design. But the "normal" conducted interferencies are still of concern. Use ferrite beads en masse and limit the slew rates of signals leaving your device to the slowest acceptable value. Protect your device by the same means - limit the slew rate of signals entering it the same way.
  There are some really good sites with useful information on them. A good design guide for Low Voltage Differential Systems (LVDS) can be found at National Semiconductors' homepage under the following link. It points to Nationals subdir and file /appinfo/lvds/index.html . Their documentation is always very good and accurate.

  ---

  © Paul Elektronik, 1998-2002