While modulation is nothing new, one newer technique that is gaining attention and being adopted in new components is PAM-4, or 4-level pulse-amplitude modulation. PAM has been used traditionally in Ethernet, but now it is being used in high speed digital components and computer peripherals to increase data rates beyond what is practical with existing differential signaling protocols. Today, you can find PAM-4 being used in some of the following areas:

• GDDR6x used in GPUs
• PCIe 6.0 (upcoming)
• 56G, 112G, and higher channel speeds for Ethernet links
• High gigabit SerDes

Other protocols like lower speed Ethernet and other IEEE standards use some variant of PAM with a different number of signal levels. For systems like high speed backplanes that need to transfer large amounts of data between system components, PAM-4 and similar signaling standards will be more important, especially once more GPU-based and PCIe 5-based components are incorporated into aerospace and defense electronics systems.

PAM-4 Bandwidth and Power Spectrum

Compared to a binary state signaling standard like non-return to zero (NRZ) or LVDS, PAM-4 has a broader power spectrum, with more power concentrated at higher frequencies. NRZ is a good standard for comparison as PAM-4 is meant to be a higher data rate replacement for NRZ. This is despite the fact that the rise time and UI of signals in the two standards may be identical. The image below shows how far the power spectrum can span into high frequencies for signals with 4 ps rise time. The Nyquist frequency is rather low, but the power spectrum is still significant up to very high frequencies.

 

 

This difference in bandwidth is what creates new signal integrity challenges for digital designers who now need to work with PAM-4. When I write "bandwidth" in this context, I'm referring to the power spectrum of a stream of PAM-4 pulses, i.e., in a pseudorandom bit stream. In terms of the data rate, the bandwidth of PAM-4 matches that of NRZ as they can have the same Nyquist frequency. This is because, in terms of clock rate, the two types of signals can have the same UI, but PAM-4 carries double the number of bits because it has double the number of signal levels. In addition, a PAM-4 channel will experience the same band-limiting due to receiver sampling rate limitations (Nyquist frequency).

This may seem to be an apparent contradiction, but measurements of the power spectrum carried in a PAM-4 bit stream reveal this difference clearly. The higher frequency content in PAM-4 signals creates some design challenges in the following areas and signal integrity metrics:

• Crosstalk and capacitive coupling. Crosstalk at higher frequencies can be stronger in the victim interconnect. Capacitive coupling is also stronger as higher frequency signals can see lower impedance to other conductors in the system.


• Losses. Any losses due to copper roughness and skin effect can be larger, which is reflected in the power spectrum shown above. This will be reflected in S-parameter measurements as well.


• Jitter and ISI. With smaller eye opening between signal levels, a receiver will have a more difficult time recovering the received signal. This requires minimizing losses, as well as designing transmission lines to account for dispersion in the PCB substrate.

If transmission lines for PAM-4 signals can be designed to accommodate these factors, it's possible for the design to maintain signal integrity and have signals that are recoverable at the receiver.

PAM-4 Routing Guidelines

Routing in PAM-4 channels on the PCB should follow the typical high speed PCB routing guidelines used in any other modern high-speed digital system. Because the power spectrum in PAM-4 is confined at higher frequencies, there can be significant attenuation along an interconnect, leading to distortion at the receiver. This is reflected in an eye diagram, as shown in the example below.

 

 

From an eye diagram, we see that any rounding off in the signal transition and variation in the signal levels will cause the eye diagram to start to close. The issue in PAM-4 is that the variation in the signal levels is small, only about 300 mV. This should illustrate the need to control losses to the greatest extent possible and prevent distortion in PAM-4 interconnects. If you look at the upper-right corner of the first graph, you'll see how extreme attenuation totally masks clear signal levels in the eye diagram.

Although the Nyquist frequency does not increase as the number of PAM-4 signal levels increases, this only affects how the receiver interprets received signals and recovers the correct signal level. In other words, a channel that is bandlimited by the Nyquist frequency is only bandlimited in terms of how it is received by the receiver. However, any frequency can create crosstalk and EMI if the board is not laid out properly. Therefore, don't just use the standard spacing rule (e.g., 5W rule between differential pairs). Instead, test your layout early with an idealized channel to assess crosstalk. This takes a multiport S-parameter approach, where the crosstalk S-parameter is measured and is compared with your signaling standard or component receiver requirements. Depending on your trace width, coupling distance, ground plane distance, and dielectric constant, you may be able to break the standard high speed routing rules, which are often only applicable in 20-year-old designs anyways.

 

If your company is pushing the limits of telecom, data center, and embedded design, it pays to work with an experienced electronics design firm. NWES helps private companies, aerospace OEMs, and defense primes design modern PCBs and create cutting-edge embedded technology. We've also partnered directly with EDA companies and advanced ITAR-compliant PCB manufacturers, and we'll make sure your next high speed digital system is fully manufacturable at scale. Contact NWES for a consultation.

 

---

---

---

---