High Speed Signal Integrity in PCBs

Common High Speed Signal Integrity Problems

To get started, the only thing that separates high speed from low speed signals is the signal rise time. There is no specific frequency or signal rise time beyond which a PCB would be considered to run at high speed. The clock frequency has nothing to do with high speed signal integrity, although it is true that signals with higher repetition rates tend to have faster edge rates. In fact, all the high speed signal integrity problems I’ll discuss here happen in every PCB. The difference is that, at lower signal rise times, these signal integrity problems are so minor that they might never be noticed, and the low speed system you build will probably work fine no matter how you create your PCB layout.

There are some common problems that only become obvious once a system begins running at high speed. With that being said, the points I’ll discuss here are well-known in high speed electronics. Here are the important high speed signal integrity problems to consider in your next digital system:

1. Signal Reflection
2. Signal Distortion
3. Crosstalk and Coupling
4. Ground Bounce

Signal Reflection

Reflection at impedance mismatches can lead to ringing in an interconnect, where the signal exhibits a damped oscillatory transient response at each reflection. This term "ringing" is sometimes used to refer to a stair-step response in the signal level seen at the driver, which occurs when reflection causes a phase reversal.

The cure here is to always impedance match your driver, interconnect, and receiver in a high speed interconnect. The use of a via on an interconnect creates two sections of transmission line, each of which will need to be impedance matched to the appropriate value. A real transmission line network with a reactive load component can create a network with a complex?input impedance spectrum, which also needs to be considered when designing an interconnect for high speed signals.

Then when we consider dispersion, copper roughness, and skin effect losses, the impedance of a line can be a very complex function of frequency. The result is that it can be difficult to perfectly match the input impedance spectrum of a circuit to the transmission line impedance without careful geometric adjustments. The example impedance spectrum shown here is for a stripline matched to a 50 Ohm receiver with ~10 pF input capacitance. With careful geometry optimization, I was able to get the impedance deviation to be less than 1 Ohm throughout the signal bandwidth with ~3 dB per inch insertion loss at 1 GHz.

Signal Distortion

Signal distortion occurs when a signal travels to the end of an interconnect, and it does not have the same shape as it did when it was injected into the interconnect. This occurs due to dispersion, which causes different frequency components in the bandwidth of a digital signal to travel at different speeds and experience different levels of loss.

The end result of signal distortion is two-fold:

• Spreading of the signal in the time domain due to the velocity difference between various frequency components.
• Greater insertion loss along an interconnect at higher frequencies.

The first point above causes a signal to appear stretched at the receiver; this stretching is greater when the interconnect is longer. The second point occurs because real transmission lines act like low-pass filters, both due to dielectric losses and resistive losses. Eventually, at sufficiently high frequency, absorption and scattering can occur as a signal propagates, which increases losses in specific frequency bands (both aspects are complex mathematics problems for inhomogeneous anisotropic media like PCB substrates). Both aspects are a reason why a shorter interconnect is preferred when possible.

Crosstalk and Coupling

Crosstalk is a simple phenomenon to understand; a replica of the rising edge of a signal is superimposed in a neighboring channel. A faster signal rise time leads to more intense crosstalk. Again, this does not depend on the data rate or clock rate. When crosstalk is excessive, the signal on the victim line can exceed the noise margin at the receiver, causing the signal to fall into the undefined region. This then causes the eye diagram to close, leading to higher bit error rates in the affected channel.

Coupling is essentially the same thing as crosstalk, and it occurs via the same mechanisms (parasitic capacitive and inductive signal transfer between conductive elements in a PCB layout). When we refer to coupling, we are basically referring to signal transfer to an element in a PCB other than a victim trace.

Ground Bounce

Ground bounce can be easily confused for crosstalk as it can produce a momentary peak in the output signal level from a driver IC. Ground bounce, also known as simultaneous switching noise, occurs when transistors in an IC switch. The parasitic inductance on the bond wire leading back to ground creates a back EMF as current rushes into the IC. This causes the signal level seen by the receiver to appear smaller as the ground potential appears higher when measured between the output and the ground pin.

Because ground bounce is caused by a parasitic inductance, and because there is some parasitic capacitance at the ground pin or pins on an IC, ground bounce can have an LC resonance. This would produce the same type of transient response shown above for the failed DDR4 channel. This is one reason why it can be difficult to distinguish crosstalk from ground bounce without some precise measurements or high speed signal integrity simulations.

What About High Frequency Signal Integrity?

In some ways, high frequency signal integrity is easier than high speed signal integrity, and in some ways they are not at all comparable. In high speed signal integrity, we pay attention to what happens to signals in the time domain. Frequency content in the time domain is spread out over a large range of frequencies; if you look at a very fast digital signal (e.g., ~17 ps/V slew rate), the bandwidth of this digital signal spreads from the clock frequency up to ~20 GHz. For a 1 GHz clock rate, this gives about 5 octaves of bandwidth.

Analog signals have much lower bandwidth, which might have frequency components spread over less than an octave. At high frequency, you’ll see the same effects discussed above, but they’ll appear within a smaller frequency band. In some ways, this makes design easier, particularly transmission line design and PDN disign. However, high frequency analog signals typically run at much higher frequencies than the high end of a digital signal. This creates new signal integrity problems that are not always visible in today’s modern high speed digital systems.

As your operating frequency increases well beyond 20 GHz, the first point you’ll start to notice is higher dielectric and absorptive losses. Second, copper roughness becomes more critical due to the skin effect, where current is concentrated around the outer edge of a conductor. Another signal integrity problem that occurs at high frequencies is the fiber weave effect, where resonances in fiber cavities induce skew (phase noise or shifts), additional losses, and radiation from cavity resonances.