It is designed with high bandwidth, which is almost processed into an open circuit?

author ： Mr. Yibo technology Expressway member David

Do you feel that you have only seen Dongge's articles recently , Did you ever think that we ** What about it . No more , Here, I still want to solemnly explain with the majority of fans , Mr. Gao has been accumulating some interesting materials these days , Learn some new technologies , In order to show you more enlightening articles , We are here again ！
 

Back to the point, huh , As our main business of Mr. Gao, that is SI, In fact, we have always been concerned about the differences between design and processing , Especially for high-speed signals , Mr. Gao only when the machining error is relatively small , In order to complete some simulation and test fitting . It happens that Mr. expressway is studying some suitable 112G Design of ultra-high broadband transmission ,PCB After design and processing , Mr. Gao immediately threw himself into the intense test , The result is the emotion of the first scene of this article ...

When we test a pair of belts 1.85 When wiring the inner layer of coaxial connector , We found unexpected results . Design is a pair of difference lines , We mainly look at the loss of a high-speed plate and add coaxial connectors , How high frequency can the bandwidth and linearity of the link go .
 
We first focus on the impedance of the coaxial connector and the impedance of the differential line , First reaction , I think it's good , Especially the wiring , The impedance of the entire link is ±5% within , Although not particularly excellent , But at least psychologically acceptable .
 
However, when we measure the insertion loss curve , The whole person is not good , What is this? This , Surprisingly, a pair of ordinary differential wiring will have such a large resonance point , stay 28GHz The position of is no different from that of the open circuit ！
 
So the first thing we suspect is design , See if it's PCB Some very low-level mistakes have been made in the design , Results the design documents were checked , No problem at all , Line width and line spacing , The same length is no problem ！
 
This is strange , At first, Mr. Gao was wondering if he had missed the back drill , So after all 3mm The board of , Our difference lines have reached a very high level , But from the difference TDR Resistance has ruled out our thoughts , Because there are no low impedance points on the whole link , If it is a missing back drill , You will definitely see the problem in the impedance ！
 
What's the problem ？ So we focus on the difference line P and N The problem of single terminal delay time , So we put this pair of differential P and N Give a step signal respectively , Let's see if there is any difference in the time domain . The result surprised Mr. Gao ,P and N There is a delay difference between 0.17？ incorrect ,1.7？ It's not right either , yes 17.5ps！！！
 
If you are right about ps If this unit has no concept , Mr. Gao changed to mil You know how exaggerated it is , It's about 100mil The difference in length , in other words , It is designed as two wires of exactly equal length , After processing, there are 100mil The difference in length , The most important thing is that the routing length is less than 3inch.
 
Through the above test and analysis , In fact, Mr. Expressway probably thought that it was caused by the glass fiber effect , What exactly is the glass fiber effect , Not here , You can read the articles in many previous issues . So we took the board to the board factory and sliced it , In order to better fit our expectations . In a moment , The slice chart comes out , Sure enough , That's the problem .
 
Maybe some fans haven't seen it PCB Slice , In fact, it is cut in the section perpendicular to the line , Use a high-power microscope to view the cut surface . Like this slice in this article , It can be seen that one of the differential lines just runs in the overlapping area of the glass fiber （ Right ）, At the other wiring window （ Left ）, And I don't know how much chance it is , There are two up and down pp（core Two, too pp form ）, and pp Also in the case of flat cloth , Still can not stop the difference P and N In a relatively consistent state . In addition, as long as we know the dielectric constant difference between glass cloth and glue , You will see how serious this situation is .
 
Or it will end like this , Mr. Gao is crying in the laboratory ...

The problem is coming.
From design to processing , Do you think there are any detailed methods to improve the glass fiber effect ？