DFSI: Overcoming Channel Loss
Table of Contents
In a previous article, we established that loss is one of the primary signal integrity challenges to overcome. In this article, we’ll talk about the sources of loss, what you can do about them, and an often neglected aspect of loss.
There are various options for reducing loss, in this article I’ll discuss how to:
- Change your architecture to orthogonal or cables
- Use better PCB materials with lower DF resin and smoother copper
- Add active repeaters to the channel
- Reduce ambient operating temperature
Loss of power in a passive channel comes from one of two primary sources: conductor losses or dielectric losses. Conductor loss is related to skin effect and surface roughness. Skin effect is the phenomenon where current gathers closer to the skin of a conductor as frequency gets higher. At high frequencies, the attenuation is higher due to skin effect since there is less surface area for the current to travel through. Surface roughness (figure 1) is the texture of copper in the PCB and further increases loss. Depending on the model, surface roughness may or may not be interrelated to skin effect. Dielectric loss is related to the complex relative permittivity of the PCB laminate. Like both skin effect and surface roughness, dielectric losses increase with frequency.
Figure 1. Cross-section of a PCB highlighting the surface roughness of copper.
The first trick I’d like to share is, consider changing architecture. The first of these changes is going from a backplane architecture to an orthogonal architecture. In a backplane architecture, there are two daughter cards and a backplane. Orthogonal has two cards, thus eliminating loss from one connector. In addition, PCB traces are shorter in Orthogonal architectures as well.
Figure 2. Traditional backplane architecture.
Figure 3. Orthogonal architecture of switch and compute.
If that is too extreme, you could reduce loss by using internal cable assemblies. There are specialized low-profile solutions that replace the PCB traces with thin gauge wire that will greatly reduce the loss. Scott McMorrow posted a really good picture here  that shows what I’m talking about. Here is a link to some internal cable assemblies for this application:
Figure 4. Internal cabled architecture.
If architecture change isn’t on the table, then you need to figure something else out. Try to find materials with lower loss tangets or dissipation factors (DF). These premium materials require processes that control surface roughness. Therefore, you not only decrease dielectric loss when moving to these materials, but you’ll also reduce loss related to surface roughness. Fun fact: DK and DF are related through the Kramer Kronig relationship. That means reducing one will reduce the other. I’ll get into that in another blog. Anyway, since loss is related to the surface area, you’ll want to make your traces as large as possible. Of course, there are limitations due to impedance control, but you want to go as large as you can without causing pair-to-pair crosstalk. When increasing cross-sectional area, do not do it by increasing the copper weight of the signal layers. While it may seem tempting to double the area by going from half-ounce to one-ounce copper, but the reduced loss comes at the expense of reduced impedance control.
If you are still not achieving your loss goals, you’re not out of luck. You can use an active device to repeat the signal within the channel. These devices are usually referred to as called repeaters, signal conditioners or lane extenders. These devices are available in two general flavors: limiting and linear. Limiting devices typically compensate up to 35 dB of loss at Nyquist, and they are the more power hungry. They also require some external circuitry to set up the device. Sometimes it could be as simple as a resistor, and the more complicated devices require a master SPI or IIC controller. Linear, on the other hand, typically compensates for about 20 dB of loss at Nyquist, require minimal setup and control, and draw about half of the power of limiting devices. The major downside with linear devices is that they also amplify the crosstalk as well as the signal. This typically isn’t a problem since the signal to noise ratio stays approximately the same. If you’re interested in checking out some repeaters, visit these two sites:
Finally, here is a scenario to be aware of that has bitten many system designers. You bring up the system in a lab, and it works. Awesome! The part undergoes system testing while varying the ambient temperature, and the link drops at high temperature. What happened? Conductivity changes with temperature, and the change is not negligible.
While I haven’t found any peer-reviewed journals on how to predict its impact, there are many conference publications available that demonstrate this effect. The most recent that I’ve seen was at DesignCon 2019 by Nanya Plastics . They monitored loss over frequency of four different materials while varying temperature and relative humidity. The materials are standard, middle, low, and ultra-low loss. They found moisture exposure increases loss insignificantly (exact numbers were not given). However, temperature had a dramatic impact. What is more interesting is the increased loss was less with lower loss material. The exact reason wasn’t explored in the study, but I suspect surface roughness could be what was causing higher losses with lower grade material. Therefore, this article suggests there is even more gains to be had from higher grade PCB resins than just lower DF and the resulting lower surface roughness. It begs the question of what is more cost effective, higher grade material or extreme cooling strategies like immersion cooling ?
 M. Rowe, “Reducing loss in high-speed signals: Cables vs PCB traces,” March 13, 2017. [Online]. Available: https://www.embedded.com/print/4458119 [Accessed March 23, 2019]
 C. W. Huang, “Thermoelectric Performance of Copper Clad Laminate,” presented at Designcon 2019 Santa Clara, CA, USA, 2019
 S. Fulton III, “How Practical is Dunking Servers in Mineral Oil Exactly?,” May 1, 2017. [Online]. Available: https://www.datacenterknowledge.com/archives/2017/05/01/data-center-cooling-how-practical-is-dunking-servers-in-oil [Accessed March 23, 2019]