06 stm32f103_4l¶
Reference: as 04; same circuit. Stackup signal / GND / GND / signal.
Stresses: four layers, inner planes, the return-path review, placement by intent, the fast place-and-preview loop.
What happened
1. The 2-layer board (04) fails the new return-path check by a wide margin:
its bottom-layer USB run has no ground on the other layer for about 19 mm.
Two layers cannot fix that; this board is the fix.
2. pad_at put every decoupler's +3V3 pad exactly where asked on the first
run. With toward= it chose a rotation that pointed the capacitor into the
pad row; explicit rot= was the better instruction there.
3. Freerouting honours (active off) for a layer: nothing was routed on
In1/In2 and both planes fill as one piece at 83 %.
4. The return-path check first reported 2.7 and 4.7 mm uncovered. All of it
was the plane's clearance hole round the signal's own vias; those are now
excluded. What is left is 0.8 mm beside a neighbouring via.
5. Six ground vias were added by hand, one beside each USB via.
Tool changes
- LAYERS, PLANES; plane layers closed to the router; inner-layer Gerbers.
- boardwright/model.py: one reader for any board; review.py rebuilt on it with
match, maxpath, ref, via_gnd.
- placement.py (pad_at), preview.py, the place step.
- After routing, open nets are found from the copper, not from Freerouting's
log: the log listed five nets as unrouted on a board with none open.
Second pass¶
via_nearfound two decouplers 2.4 and 2.7 mm from the nearest ground via. Every decoupler now has a ground via placed by hand beside it.- Regenerating with one more via made Freerouting leave PC14 and PA9 unrouted. A completion pass with the pad-to-pad router could not finish them either: both pins were boxed in by tracks the autorouter had already laid. The build now retries with a varied bottom-layer cost; the second attempt completed with nothing open.
Left open
- USB is full speed. No pair was routed with si.pair on a real board yet,
and no impedance target was applied to this stackup.
- Not built.