Mr. High Speed member - Huang Gang
As for how much ohm impedance needs to be controlled for a high-speed differential signal, Mr. High-speed believes that most engineers will first look at the protocol document of the signal or the chip document to see if there is a recommended control impedance value. For example, for PCIE signals, the impedance after 4.0 will be explicitly required to be controlled at 85 ohms, and the USB impedance will be required to be controlled at 90 ohms. In addition to the clear impedance requirements in this part, how many ohmics will you control for other high-speed signals that are not clearly required? Just as the single-ended traces of a PCB should be controlled at 50 ohms, differential traces should be controlled at 100 ohms if there is no explicit protocol. In fact, many engineers do not necessarily know the internal theory and reasons very clearly, but they will also subconsciously control 100 ohms, which shows how deeply rooted the concept of 100 ohm differential line is!
However, what we want to explore in this article is: Is 100 ohms really the best impedance choice in any product? Of course, from the theory of reflection, if the impedance of the link from the load of the transceiver chip to each part of the PCB is perfectly 100 ohms, then the high-speed wire control 100 ohms must be the best choice! It means that the impedances are matched everywhere on the link, and there is no reflection at all, can this still be bad?
What will the real situation be? In order to be convincing, this article gives 2 simulation cases of real projects, let's taste them together!
Case 1: On-board chip-to-chip 25G signal simulation case
In the chip-to-chip PCB link, in addition to the PCB traces, there must be some impedance discontinuous structures, such as in the above example, there will be vias at both ends of the BGA, and there will generally be AC coupling capacitors at the receiving end. Friends who have a certain amount of simulation experience know that the impedance of the capacitor at the position of BGA vias is generally difficult to achieve 100 ohms, and most cases will be lower than 100 ohms no matter how they are optimized. At this time, let's verify that if the impedance of these impedance discontinuities does not reach 100 ohms, for example, in the case of about 95 ohms, the PCB traces are compared according to 100 and 95 ohms respectively.
First, let's take a look at the chip-to-chip link TDR impedance comparison, which is the difference between the default 100 ohms for PCB traces and the 95 ohms for control.
Of course, from the perspective of TDR impedance, we can't see the difference intuitively, so let's look at two other more critical indicators, that is, the indicators of insertion loss and return loss.
From the point of view of insertion loss, after optimizing several discontinuous points, although the simulation performance of 100 ohm traces is also very good, it can be seen from the simulation results that the results of 95 ohm PCB traces are more advantageous, whether from the perspective of return loss or insertion loss, it is the party with better performance.
Case 2: Chip-to-chip simulation of a backplane connector
Another case is the case of the cross-version 25G signal, and the connection relationship of the whole system is as follows:
The impedance of the BGA vias on the previous board is not described, and the high-speed connector that is connected across the board should be paid attention to. The high-speed backplane connector used in this case is a well-known manufacturer's product, and it is a connector that is very versatile and validated at this rate.
What is the nominal impedance of this connector manufacturer? 92 ohms, not 100 ohms as you think. We get the impedance of the signal of one of the pairs of connectors, and it is indeed about the same. It's in the early 90s.
Then we are still the same, the three interconnected boards of this system, we control the impedance of high-speed traces by 100 ohms and reduce them to 92 ohms, and see how the performance compares.
First of all, let's look at the comparison of the TDR impedance of the whole link, and we can clearly see the low point of the two connectors, as follows:
So what is the performance comparison between insertion loss and return loss? Still disappointing everyone, three boards to control 92 ohm routing will still be better than 100 ohm routing.
Of course, the theory is a bit complicated, so I won't expand on it here. From a design point of view, if there is no clear requirement, it is not too much of a problem for hardware engineers or design engineers to control high-speed traces according to 100 ohms, and many of our cases are designed according to 100 ohms differential signals. This article is more likely to give you an alternative way of thinking to find some better design solutions. But it's still a weak notice, if you don't go through a more accurate simulation, or don't try it casually, because you don't know how much is good, only simulation can be a good quantification of the performance of the link, design friends please be cautious to use this trick, it is still easy to overturn if you don't use it well!
Here's the problem:
List the common impedance discontinuities in PCB design of your products, and briefly describe your design optimization methods?