| ??? 11/16/07 17:10 Modified: 11/16/07 17:18 Read: times |
#147080 - Layers, layouts... Responding to: ???'s previous message |
Richard said:
Just one small point however, it is a common misconception that using a star point for power supply connections will actually improve things, I don't remember the details, the the headline I certainly did. I recently went on a course that was about signal integrity and EMI control, and along with my colleagues were suprised at how many "accepted" practices are either a no no, or just don't work. There really is no subsitute for proper power/ground planes, and they must be placed in the correct layers to work effectively. It always depends! You are right about the ground planes. Today we want to have a ground plane instead of a star ground point. But this does not mean, that everything should be connected to it in a haphazard way. The related ground currents, if they are noisy, should flow in such a way, that they do not interfere with each other, so that the idea of the star ground point should be kept in mind, although using a ground plane. Some applications even need a split of ground plane to reserve an extra path for super noisy or high currents, so that they cannot erode the ground potential. Better than splitting the ground plane in this case is to use a separate layer for this current only, of course. Another issue is, that if you use separate signal layers, then you might need additional ground planes just for shielding of these signal layers, so that no two signal layers are in direct contact, in order to avoid capacitive cross coupling. This is the case for instance, if you want to use a ground plane on top and bottom to form a "Faraday cage" for the whole board. Then, with two signal layers, you would need a third ground layer in the middle of board: "G S G S G", where "G" means "ground" and "S" means "signal". Normally, the arrangement "S G G S", or "G S G S" with ground fills on the signal layers is enough. High frequent clock signals, or more concretely spoken clock signals with sharp edges should also run in a separate layer, if the signal trace must be rather long for any reason. In extreme cases they might need their own ground planes nearby. With the supply planes I disagree a bit. You don't need to have a supply plane, if you extensively filter the individual Vcc pins of chips. This is what I do in mixed analog digital designs, where ADC/DAC can share the same supply line. Only in extreme demanding high frequency applications a solid supply plane is of benefit. A solid ground plane is far more important than a supply plane, proper supply voltage decoupling DIRECTLY at EACH chip provided. Richard said:
Another is to believe the manufacturers regarding layout, they are reguarly wrong, however if you do deviate ant it does not work then they say sorry you did not use our design, catch 22. Hhm, I wouldn't say they are "reguarly wrong", but just the opposite, they are too perfect to work in a real world. Maybe we mean the same? What you often see in datasheets of ADCs/DACs is a layout that is optimized for highest performance. But look where they get the supply voltages and related grounds from! These seem to come from all directions, from different parts of the world, carefully prepared to have absolutely no connection to each other, no galvanic and no due to stray capacitance. But in a standard product you have only ONE supply unit, and the ADC/DAC's supply voltages must also supply other parts of your circuitry. More, EMI filtering guarantees, that they "see" each other far before they hit your ADCs/DACs. So, you can forget the supply and ground routing of such a recommended layout. But you can learn from it... Also, they can easily connect the several grounds under their ADC/DAC to provide best performance. But in a real project you have to connect the grounds somewhere else normally. So, these example layouts are not wrong, they are too perfect. Kai |



