Is my memory routing good enough for a first processor board?

Oct 19, 2026 Last reply: 8 hours ago 5 Replies

First reply

hweber

0.8mm is stiffer than it feels in your hand once everything's soldered down, but the weak points are where connectors take side loads. Headphone jacks and card slots get levered every time you plug something in, so if the case doesn't support the board right at those spots you…

Read the full reply

First time putting a real processor on a board and I could use a sanity check before I send this off to fab. It's a little Linux audio player I've been chipping away at for about a year, with a small e-ink style screen, headphone outs, sd card slots, the usual. The part that keeps me up at night is the memory routing. I did serpentine tuning on the data lines to keep them length matched, and I buried them on an inner layer with ground planes on both sides. My stackup is six layers, signals on the outside, power and grounds in the middle. I read that a thinner board lets you hit the right impedance with narrower traces, so I went with 0.8mm, but now I'm second guessing whether that's too flimsy for something that lives in a pocket or a bag. The other thing nagging me is the wireless module placement. It sits near one corner with the antenna area hanging off the edge, but there's a card slot and a headphone jack not too far away. Will all that metal around it hurt the range or mess with the signal? Happy to share the layout files if anyone wants to poke around. No formal training here, so if something looks obviously dumb, just tell me straight.


0.8mm is stiffer than it feels in your hand once everything's soldered down, but the weak points are where connectors take side loads. Headphone jacks and card slots get levered every time you plug something in, so if the case doesn't support the board right at those spots you can crack traces over months of use. Design the enclosure so it clamps near the jacks and slots and you'll be fine. Plenty of pocket devices run thinner than that. On the serpentine work, one trick that saves a lot of headache: route the longest memory trace as short as you possibly can first, then match everything else down to it. If you match to whatever your longest net happens to be, all the meanders stay big. Shorten the boss net and the tuning shrinks with it, which frees up board space too. And when you do need meanders, keep them gentle and well spaced. Tight little serpentine combs couple the trace to itself more than to anything else, which defeats the point. For the wireless corner, the big enemy is metal directly under or in front of the antenna, not metal a few centimeters away. Jacks and slots off to the side will cost you some, but a pour or a grounded shield sitting in the antenna's near field will hurt far more. Hang the module off the edge like you're doing, then check the module maker's layout guide for thier minimum clearance numbers and keep the area around the antenna clean. No traces, no pour, nothing following it out past the board edge. Burying the memory lines between planes was the right move. Just watch the via fields underneath them. Rows of vias punching through the planes slice the return path into strips, and that's where you get crosstalk and weird edges. Stagger the vias so the planes stay mostly solid under the bus, never more than a couple side by side in one gap.

One last thing: get the actual pressed thickness and dielectric numbers from your fab before you lock in trace widths. The stackup you draw and the stackup they build are rarely identical, and your tuning depends on what really shows up.

One thing to check before fab: where the headphone ground return actually flows. On the last player board I did, I had the amp's analog ground sharing a path with the SD card return current and got a faint ticking that followed card reads. Fixed it by giving the codec its own little ground island tied back to the main plane at one point. Cheap to do now, miserable to hunt down after the boards arrive. Also curious how you're handling the e-ink cable. If it's a flex tail, leave some slack in the footprint so the connector isn't taking the strain every time the screen gets pressed in a pocket.

Seconding the ground island thing. On a little DAC board I built years back I had a whine that changed with screen activity, turned out the display supply was sneaking across the audio return path. Cut the ground under the codec free from the rest and tied it back at one spot near the amp and it went silent. Took an evening with a knife blade to slice the pour, way easier to do it in layout like you still can. defintely poke around their files if they share them, that kind of return path problem never shows up in the schematic.

Before you send it, build in a way to fix mistakes. On my first Linux handheld I swapped two address lines and figured the board was junk. A cheap hot air rework station off the usual hobby sites saved it, I pulled the chip, scratched a couple of jumpers, and reflowed it the same night. Now I never order a board like this without that station sitting next to the soldering iron. Also add a row of test pads on the memory bus while it's free, tiny ones tucked between the vias. When it boots half dead you'll want to probe those without balancing a scope tip on a pin.

Update: took the ground island advice and gave the codec its own little patch tied back near the amp, and the noise following card reads is basicly gone. Still stuck on the wireless corner though, since the jack and card slot are right there. How much clearance does the antenna area actually need before nearby metal starts eating the range?

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required