When does signal integrity become a potential issue in PCB designs, and how do we address it? Learn why signal integrity in high-speed PCB designs is important.
By Jeremy Cook
Signal Integrity for PCB Designers
When a microcontroller passes signals to other devices –via simple on/off switching, protocols like I2C and SPI, or even control pulses for WS2812B LEDs – it does so in nominally binary pulses representing 1s and 0s. In reality, though, nothing in the electronics world works in a purely binary format. This may or may not be a problem, as receiving devices are designed to treat signals above a specific range as high and below a certain degree as low to deal with ambiguity.
This inconsistent behavior can be seen on an oscilloscope, especially when set up to view signals via an eye diagram. Ambiguity only worsens the signal's journey between its origin at the transmitting device and where it's received. The effect is practically nonexistent in slowly changing signals. For example, flashing an LED once a second or sending control pulses in the millisecond range.
Signal integrity needs to be considered when signal speeds advance into the megahertz range and/or rise times are very quick.
We can't fully cover the broad subject of PCB design for signal integrity in a single article. However, as a short introduction, this article will outline a few basic concepts you'll want to consider when designing a high-speed board.
Is Data Integrity a Potential Problem for Your Design?
First, consider that as a signal ramps up to a higher and higher cyclic rate, wires don't act simply as pure conductors but as electromagnetic transmitters, too, spewing spurious signals that can be picked up by nearby wiring. This doesn't just relate to pure cyclic frequency but also the rise time -- the time a signal takes to transition between one state and another.
A good rule of thumb is that if you're dealing with signals with a rise time of under one nanosecond (ns), or clock frequencies over 100 MHz, then your PCB may need to factor in potential signal degradation. In the context of a PCB, the longer your traces are, the higher the chance of interference, so as noted below, it's best to keep things as short as possible.
Signal Integrity: Short, Straight, and Uniform
The first rule of signal integrity is to keep your signal paths as short as possible. While 1ns/100MHz is a good rule of thumb, the actual interference value relates to the length of the traces versus the wavelength of the electrical signals. If the transmission distance is less than 1/10th the wavelength, then there shouldn't be a problem.
Signal speeds here can be approximated as ½ the speed of light, or 1.5 x 10^8 meters/second, so when divided by the frequency in Hertz (AKA cycles/second), you're left with the wavelength in meters. A 100 MHz signal wavelength would be 1.5 x 10^8 meters/second divided by 100,000,000 cycles/second, which works out to 1.5 meters/cycle, or a 1.5-meter wavelength. 1/10th of this value would be 15 centimeters, a long trace length, but certainly conceivable. Keep traces significantly shorter, and even at 100 MHz or more, you still may not have a problem.
At the other end of the spectrum, consider the US Navy's Extremely Low-Frequency project, which used antennas 14 and 28 miles long to produce radio waves between 30 and 300 Hz for communication with submerged submarines. In other words, with this extreme antenna configuration, just a millionth of the stated rule of thumb's frequency (albeit using a massive amount of power) could generate radio waves that could signal undersea vessels thousands of miles away.
Besides issues with trace length, impedance variations in the signal path from transmitter to receiver can cause reflections that interfere with the signal itself. To help maintain constant impedance, signal trace widths should be uniform: Avoid vias when possible. Impedance mismatches between the transmission line and load must be considered and rectified as appropriate, potentially involving a resistor in parallel.
Of course, avoiding vias altogether on a PCB may not be possible and would certainly mean longer traces than would otherwise be necessary. Given the tradeoffs involved in PCB design, it's a good idea to simulate the board's function to analyze potential high-frequency effects. This can help you catch errors in your PCB before actually being fabricated, potentially saving significant development costs.
Ground Plane…Or Return Paths?
You may think of a ground plane, i.e., current return path, as something to formed with a sufficient amount of copper cladding and forgotten. For high-speed PCB design, it's better to consider the ground plane as a series of return paths that (ideally) work in parallel and follow the signal traces, even if it is a contiguous area of copper. The returning current will follow the signal path as closely as possible to minimize impedance at high speeds.
In many designs, you'll want to explore using stitching vias to connect disparate ground planes. This can help form an isolating Faraday cage around noisy signal lines and help keep return paths as short as possible.
Decoupling Capacitor
While somewhat tangential to signal integrity, consider that supplied DC power fluctuates and is rarely a constant 5V or 3.3V value. If clean power isn't available, this can mean the less-than-optimal operation of signal sources such as microcontrollers or addressable WS2812B LEDs that pass sequential signals onto their neighbors. To clean up the signal, one can use a decoupling capacitor.
Since a dielectric separates the layers of a PCB, one might note that this would also serve as a capacitor if power and ground planes are arranged on either side of a two-layer configuration. Unfortunately, the standard 1.6mm/.063" two-layer board only yields a capacitance of 17.1 picofarads (pF) per square inch or 143pF on the decent-sized 8.36in2 board that I've been working on recently. A typical decoupling capacitor is, in fact, 100 nanofarads (nF), or roughly 700 times greater than what's naturally available if my board was completely covered with copper. I suppose every little bit helps, but it's a massive mismatch in scale.
High-Speed Data Integrity Tradeoffs
Given the challenges of working with high-frequency designs, it's understandable that not every device uses extremely high frequencies. In many situations, i.e., a USB keyboard and mouse, comparatively slower data transfer speeds are more than acceptable. If one can potentially avoid various RF hassles, speed tradeoffs should be considered early in the design process. When necessary, a well-thought-out high-frequency PCB is a thing of beauty, but there's also something to be said for a configuration that sidesteps these issues altogether.