Electronic – How to improve this PCB schema for L293D H-brigde

1) Crystals should not be routed this way. Traces should be shorter and as symmetrical as possible. You should connect capacitors to GND in a single point, so that you are not picking any noise from the ground plate. This is especially important for RTC crystal. With current routing you might get problems with generation start/failure if you are unlucky.

2) Checkout my single-layer board for ARM : http://hackaday.com/2011/08/03/an-arm-dev-board-you-can-make-at-home/ - even this nightmare works (only 1 decoupling cap). Defenitely what you have here will work. You may add some extra caps (like some 25uF electrolytic + 2.2uF ceramic) on the backside of the board, you have plenty of space there, and both VCC & GND together. The only thing I don't like is thin traces to your caps. They should be as wide as possible. In my design, the only capacitor was connected by like 2mm-wide traces.

Also, look at C5: You can move it to the right a little, move via closer to the cap and connect it with short wide track. When you via is under the chip, you cannot have wide tracks. Same for C6 and C7.

Also, if you are going to manufacture this at home,you'll have problems making vias under QFP chips.

3) Ground plate is more than enough. There is no much need to have solid ground plane except a square under chip where all decoupling caps are connected, it won't help with ground noise much. Ground plate is needed for controlled impedance, which is not important in your case. But your GND connection to contacts should be as wide as possible. This is general rule: VCC & GND nets should have wide tracks.

4) Yes, this is perfectly ok for low-speed ARMs.

In my case I even had no back side, and it was still working ;-) The only thing to improve if you are manufacturing on a factory is to have a small VCC square on the bottom layer in the middle of the chip, and connect it to the top using some 4-9 vias instead of 1. For VCC & GND planes you always need to have as low as possible resistances and inductance so that caps can easier filter noise => you need wider and shorter tracks and more parallel vias. But in this specific design it is not a requirement.

So, it will work even now without modifications. After mentioned changes it will be perfect.

There is almost no reason to design a one-layer board these days. The entry level service for any board manufacturer will be two-sided boards with plated through holes. The only reason I can conceive of for doing a one-layer board is if you're going to etch it yourself and want to avoid connecting traces on the front to traces on the back.

I haven't used Fritzing's (what appears to be) magic for converting a wiring diagram into a PCB, but at OSHPark pricing it's worth a shot I'd say. It will either work or not depending on how sensitive your interfaces are to things like noise and trace lengths and such. But if you're wiring it up willy-nilly on a breadboard and it works, the chances of it being dumb luck are pretty low. It will probably work with whatever auto-routing Fritzing is doing behind the scenes just as well.

For the record, though, this is not the usual way PCBs are designed. Usually you'll do Schematic Entry, then you'll do Place and Route (that is placement of components on your board, and routing of traces between them). Many experienced designers will tell you that you should do the routing manually and maybe use the auto-router as an indicator of whether your component placement is OK and iterate. Once you're satisfied with your layout, you would generate Gerber files, which are vector artwork format for manufacturing. You get one file per layer, and the layers are typically:

1. Top copper
2. Top silkscreen
3. Top soldermask
4. Bottom copper
5. Bottom silkscreen (optional)
6. Bottom soldermask
7. Board outline
8. Drill file

You might also have a Milling layer, and if you're doing surface mount, you'll generate a Paste layer for the top and bottom as necessary to have a stencil created.