Taking this somewhat ambiguous subtitle and putting some parameters and definers around it, the topic that this article addresses is the manner in which differential signaling has evolved into being the principal way in which electronic components are connected and some of the things that can degrade a signal path. At this particular juncture in time, the entire Internet relies on differential signaling to move data from one place to another. Without this method of signaling, the Internet and the plethora of communication products upon which we have all come to rely would not be viable or, even more accurately, couldn’t exist.
The good news is that the latest versions of ICs make it possible to send data over differential pairs in PCBs at rates as high as 32Gb/S (and higher). The bad news or, perhaps more accurately, the gauntlet that has been thrown down to the industry is that at these speeds, very small variations in the laminates used to fabricate PCBs can destroy a data path if care is not taken in how these paths are designed and manufactured. And, one of the variations that destroys data links in a path is skew.
Starting with the basics of the problem, skew is the misalignment of the two signal edges of a differential pair as they arrive at the terminals of the differential receiver. The outcome of skew is pretty straight forward. When the aforementioned misalignment is bad enough, the links in the data paths may no longer work.
At this point, it’s useful to define differential pairs and review how they operate:
- A differential pair is a signal path that has two equal and opposite signals travelling on the two paths.
- The receiver detects the “difference” voltage at its input and determines whether the data bit is a one or a zero.
- The receiver is designed to ignore the “common mode” component of the signal, usually a voltage offset. This is the primary benefit of this signaling protocol.
- Figure 1 is a graphical representation of a differential pair.
It’s certainly true that matching the physical lengths of two different signals is not a problem. And this is the result of the following factors:
Based on these factors, it would seem that accounting for and managing skew should pretty much be an engineering slam dunk. This would certainly be the case if it weren’t for some less obvious sources of skew.
- Modern IC technology is capable of maintaining alignment at the package level to as little as 1 ps.
- Modern PCB layout tools and connector manufacturing are capable of matching lengths to as little as 1 ps.
The Hidden “Gotchas” in Skew
As noted previously, differences in the lengths of differential signaling paths, or misalignments, are contributors to skew. In addition, skew can be the result of differences in the velocity of the two differential signaling paths. In both instances, the contributors to skew are a result of the glass weaves used in PCB laminates. Specifically:
First, some basics:
- Differences in path lengths as large as 37 pS have been measured over a 14 inch path created by the lack of uniformity of the glass weaves in laminates. (This is more than one bit period at 28Gb/S.)
- Differences in the velocity of differential signaling pairs occur when one side of the pair travels over the resin while the other side travels over glass in a PCB laminate. The side that travels over the resin will travel much faster.
The laminate in a PCB is a composite of glass and resin. The dielectric constant of the glass is around 6 and the dielectric constant of the resin is less than 3.