The glass fiber effect, as a mysterious and obscure existence, has always been a part of the high-speed signal design that engineers cannot reassure about. As a kind of existence that does not necessarily happen, but once it happens, it will affect the signal quality, design engineers and board manufacturers can be said to have tried every means to avoid it as much as possible. Because Chris thinks that the causes and solutions of the glass fiber effect he talked about are not as good as the previous article written by Mr. High Speed "The Glass Fiber Effect Explained in the DesignCon Article Limits Our Imagination of High Speed?" That's a good story, so I deliberately turned over the cabinet to find out this article, so that everyone has a certain understanding of the glass fiber effect and the conventional solutions we are using!
Yes, at present, several methods that are used more are either to use PP and Core of flat fiberglass cloth, such as 1078, 1035, etc., or the design engineer needs to work hard to make the high-speed signal line into a certain angle of routing, or waste more sheets, so that the board factory can do a certain angle of panel rotation. Of course, as mentioned above, the glass fiber effect will definitely exist in a certain sense, but it is only a matter of degree. In the selection of flat fiberglass cloth, the choice of 1078 or 1035 these good PP and Core can indeed weaken its influence with a high probability, but it is not certain, very, very occasionally you can also see a flat fiberglass cloth with a very large glass fiber effect case, just like the article "Obviously designed with high bandwidth, you almost processed it for me to open the way?" It's not 100% insured.
Chris recently suddenly wanted to take a closer look at the glass fiber effect, so he also looked through the relevant industry documents to see if there was any better solution. I accidentally saw the long-dusty design idea proposed by a large company, which is the method of the following figure!
What is this line of thinking? This method is introduced as an example of 1035 and 1078, and we know that theoretically PP and Core, which are named after a certain name, represent that the way the glass cloth is woven is fixed, and to put it bluntly, they think that the spacing between the horizontal and longitudinal fiberglass cloth is fixed. 1035 is about 15.2mil lateral spacing, 1078 is about 18.5mil lateral spacing. Therefore, this new design idea is proposed, that is, if the line width and spacing of the differential line are just the same as the spacing of the glass fiber cloth, it means that no matter where the trace is on the glass fiber, the changes felt by the two lines between the differential lines are exactly the same, so that theoretically there is no modal conversion, and in this sense, there is no glass fiber effect!!
At the first moment of seeing this method, Chris felt that his eyes lit up. As a serious engineer, Chris plans to try to test this idea through simulation. If you want to play, play with the worst PP, Chris 3D modeled according to the specifications of 1080PP, and probably built a 3D model that fits the actual PP very well.
It can be seen that the 1080PP glass fiber window is really big! Then we design the differential line to be taken on it according to the theoretical spacing of 1080 glass fiber. Judging from the data obtained, the center spacing of the glass cloth in the meridian direction of 1080PP is 17mil.
Therefore, according to the thickness of the modeled PP, we designed two sets of differential line structures under the premise that the impedance of 100 ohms is satisfied.
Case1: The line width is 5mil, the line spacing is 12mil, and the center spacing is 17mil, which is just the same structure as the glass fiber spacing; to verify the effectiveness of this new design idea.
Case2: The line width is 4.5mil, the line spacing is 7mil, and the center spacing is 11.5mil, which is not the same as the glass fiber spacing.
The final 3D model is shown below.
Why don't we take a look at the case of case2, which is not designed according to the glass fiber spacing? Then we arbitrarily move the pair of differential lines laterally so that they exist in several different positions on the PP, for example.
Oh yes, and I forgot to remind you that the difference line modeling of both cases is 2 inches in length! Let's take a look at the difference between the two cases with only a 2-inch trace that is not long!
Then let's start to perform a passive simulation of case2, the difference line in the above 4 position situations, and look at the two main indicators of modal conversion and the impact of insertion loss.
From the modal conversion, it can be seen that the modal conversion of the 4 traces is obviously affected by the glass fiber effect, and the effect in the two states of 1 and 3 is even worse!
At the same time, it can be seen that the impact on insertion loss is also very obvious. This is certainly not the case with normal differential routing.
There is no harm if there is no comparison, so let's turn our heads and take a look at the situation under this new idea design, that is, the simulation results of Case1.
In case 1, the distribution of the moving differential line is analyzed in 4 states with case 2, as follows:
First of all, let's take a look at the most important indicator of modal conversion, and I really have to believe that this new design idea really presents a perfect state in the simulation, and the results of modal conversion are as stable as an old dog in 4 different positions! All are below -40db, which is very ideal.
In this way, we can also see the curve we like in the insertion loss result, and the insertion loss curve that an ideal differential line should have!
When Chris shared the document with the group, everyone said it was great and wanted to try it out for future projects. After careful consideration, Chris couldn't help but pour a little cold water on everyone, and asked everyone lightly: Since this theoretical method has been proposed for so many years, why have you hardly seen anyone design like this in actual projects?
The problem?
What if you think this is a great method in theory, but what if it is actually processed? The questions are very open, there is no right or wrong, I hope you will answer enthusiastically!