Good PCB (Printed Circuit Board) layout is pretty much the same for all applications, whether the signals of concern are DC, a few Megahertz, or in the Gigahertz area. Having said that, there are some things that can change as we go up or down in frequency. For example, separation of DC signals is not as critical as is separation of high frequency signals. Nevertheless, following some general rules will help maintain signal integrity (SI) and minimize electromagnetic interference (EMI), resulting in successful connections and minimal noise, consistent with the choice of adequate components.

Unfortunately, this subject is so large that complete answers would require a small book, so are beyond the scope of this writing. I am sorry that this is necessarily the case.

I consider the first rule in maintaining signal integrity to be considering all traces to be transmission lines. This is regardless of whether that line carries DC or a dynamic signal. The truth is that most lines have AC components on them whether we realize it or not. There are exceptions, of course, but if we consider all traces to be transmission lines we can not go wrong. What is involved with treating traces as transmission lines?

The first thing is that the output impedance must equal the impedance of the transmission line it is driving and the impedance of that transmission line must equal the input impedance of the receiver. Low impedance lines means a lot of current is required to drive them, putting demands upon the line driver and the power supply. High impedance lines are more sensitive to the system capacitances, including the capacitance from the transmission line to ground and the input capacitance of the receiver. The latter provides for a maximum input impedance of the receiver and can make the receiver input impedance very dependant upon input frequency; not a desirable situation for wide band signals. Transmission line impedances of 50 to 200 Ohms turns out to be pretty reasonable for most systems, with 50 to 75 Ohms being common for single-ended signals and 100 to 150 Ohms being common for differential signals.

The best transmission lines have only two connections to it: one at the source end and one at the receiving end. Admittedly, this is not always feasible without the use of special circuits like clock buffers for clocking circuits, for example. With care, however, it is possible to use a single line to drive more than one signal receiver. The scope of that is also beyond our scope here.

Maintaining a constant transmission line impedance means that it is best to keep the line entirely in one board layer, but it can go through a properly designed via (hole through the board) close to a device pin to which the line is connected. Defining a properly designed via is beyond our scope here.

Setting the impedance of the transmission line involves consideration of trace to return plane capacitance, which are determined by trace width and trace to return plane distance, as well as the dielectric constant of the board material. For differential signals, the separation of the two lines also affects the differential impedance. Be aware that there most certainly is ground return current associated with differential signal lines. It is a myth that the only signal current in differential lines are in the lines themselves.

Generally, we decide the impedance of the transmission line, and then we match the impedance of the driver and receiver to that impedance. If the driver output impedance is lower than that of the transmission line, add a resistor in series with the driver such that the driver output impedance plus that of the resistor is equal to that of the transmission line. If the output impedance of the driver is greater than that of the transmission line, a resistor to ground should be placed at the output of the driver. In either case, the resistor should be as close as possible (not just as close as reasonable) to the driver output pin.

Terminating the transmission line for low to moderate frequencies is just as simple as matching the driver impedance to the transmission line and uses the same method. For high frequencies we should consider impedance of the receiver input capacitance.

Remember that signal currents follow the path of least impedance (not necessarily the same as the least resistance).This means that sometimes the return current will flow in a supply plane, which might not be desirable as this could add high frequency noise to the supply line. Active devices have very poor power supply rejection (PSR) at high frequencies (PSR or PSRR specs are generally at relatively low frequencies). Having a solid, unbroken ground plane with no traces in it will help direct return currents to that plane. Other techniques, beyond our scope, can be used when return currents must sometimes flow through power planes and the signal line crosses more than one power plane, or when separate analog and digital ground planes are used and the signal line must pass over both of these planes.

Remember that signals propagate. That is, they do not have an instantaneous jump from one place to another on the board, so there is a time associated with stimulating a line until it is “seen” at the other end of that line. This time is generally on the order of nanoseconds, but it is a time delay. Accordingly, it is important that differential signals have both lines that are of the same length. The higher the frequency, the more important this is. What we need is that the time difference for a signal to traverse the distance of the two lines is very short compared to the wavelength of the highest frequency on the board.
To avoid signal feedback problems, we should keep the signal path as straight as possible. That is, avoid such things as the signal farther down the path coming close to the signal earlier in the path. Also, always keep higher level signals away from lower level signals.
Keep in mind that all paths have an impedance. Currents flowing through this impedance will cause voltage drops. Also realize that currents follow the path of least impedance, meaning that currents in a return plane will be concentrated below (or above) the signal line, If two traces are above and below each other in different board layers, their return signals could flow together, resulting in crosstalk or noise. This is less of a problem at very high frequencies because of what is known as the “skin effect”, which is, again, beyond our scope.
Finally, realize that copper fill that is grounded at a single point can form a folded monopole antenna, picking up signals and feeding them into the ground plane, resulting in circuit noise. This antenna can also radiate signals that flow in the ground plane at or near the ground via connected to this “antenna”. The solution is to ground all copper fill at more than one point and at all corners. It is a good idea to ground such fill with a grid of vias that are separated by less than 1/4 the wavelength of the highest frequency of the system.
Current loops also define an antenn a that can radiate. These loops are formed when the return path must divert from close proximity with the outgoing signal path. Therefore, preventing radiation is another reason to ensure that the return signal can remain close to its outgoing signal counterpart.
Good PCB design is not a trivial matter, but keeping in mind a few good rules can go a long way toward successful board layout with good performance and the elimination of signal integrity and EMI issues.