amk wrote:Why use a two layer board?
I have a question...
I read the posts on the Compute Module Test Board, and I'm wondering what would *actually* happen if you had stayed with a two layer design? I'm designing a compute module board that's 80mm x 100mm, has no LAN chip, and also has the host USB connector on the other side of the board (for logistic reasons). The pair of traces from the compute module to the USB switch chip are about 80mm long. I've routed them as 12 mils wide and 18 mils apart with 20 mil spacing to the ground planes on the sides.
I understand the "matched length 90R differential pair" requirement. It's reasonable, justifiable, and *expensive*. CAD tools to do 4-layer work are often hundreds or thousands of dollars more. Prototype 4-layer PC boards are at least triple the cost of 2-layer boards from my favorite fab lab. I do my best to ensure the ground planes on my top/bottom combined signal/ground plane design is as contiguous and non-loopy as possible. I'm generous with top-to-bottom ground vias, I go over all auto-routed traces and make them 'islands' in the ground copper and without pointy edges. Previous projects I've done using high-frequency AC lines have never shown many problems.
My question is... "Why not use a two layer board?"
If I ignore all the warnings and paranoia, what might I expect? Calculations show that my expected impedance on the pair is about 88 ohms Z(odd), 176 ohms Z(differential). Assuming that the electricity on this low-voltage PC board is going only about half the speed of light, and that 80mm trace length is within the 100 picosecond requirement for USB 2.0, that this board is run inside a Faraday cage and a negligible amount of other EMF radiation is present... what? Will it work at all? The higher impedance melts the signal to zero? Diminished USB performance? Cold fusion detonation? International communism? It seems to me that the signal impedance match and ground impedance mismatch will result in a mildly distorted waveform but will still work... especially at a mere Full Speed (USB v1.1) connection.
Any thoughts? I know this is a reeally sloppy question, but I thought maybe I could get a little more insight from the forum. I'm guessing the best answer to this is, "Give it a shot!"
This won't be the final design, I fully expect to go to 4-layer for final production... *if necessary*. It might not be needed. Developing RF speed PCBs requires some trial and error, and I'm prepared for that, but I'm also keen on saving a bit of coin whenever possible. Thank you kindly for your consideration.
PS> extra special bonus points for Michigan State alumni, as the University of Michigan electrical engineers I've asked about this were clueless.