Skip to content

03 tps54202_buck

Reference: TI TPS54202 datasheet SLVSD26C, Figure 7-1 "5V, 2A Reference Design" and section 7.4.1 layout guidelines. Read from the PDF this time: C1 10 uF, C2 0.1 uF, C3 0.1 uF boot, L1 15 uH, C4/C5 22 uF, R1 49.9, R2 100k, R3 13.3k, C6 75 pF, R4 511k, R5 105k. Vref 0.596 V gives 5.08 V. Chosen here, not from TI: footprints, the inductor series (Bourns SRN6045TA as a candidate; check its saturation current against 2 A plus ripple before ordering), terminal blocks, the power LED, mounting holes.

Stresses: a layout where the autorouter's answer is electrically legal and wrong. Wide current paths, a hot switching loop, a feedback node to keep quiet.

What broke 1. Power nets at 0.6 mm: Freerouting left SW and VOUT unrouted round the SOT-23-6. 2. The 3D render does not show copper pours. I could not see whether the power polygons existed. 3. Stitching vias I placed came back twice (import-ses re-adds protected vias): "drilled holes co-located". 4. Mounting holes: parity warning, the footprint is exclude-from-BOM and the symbol was not. 5. DRC 0, board wrong. The router ran the sense trace diagonally through the VIN polygon and a GND trace through the SW polygon. Each polygon filled as two islands joined by a 0.4 mm trace. KiCad accepts that. 6. Exporting the polygons to the router as planes did not help: Freerouting stopped routing those nets and still crossed them with others (2 unconnected). 7. With protected hand-routed tracks present, Freerouting's own summary ("7 unrouted") is wrong; KiCad reported 0 unconnected.

Tool changes - boardwright/copper.py: ZONES (local polygons, solid pad connection, priority over the GND pour), VIAS, TRACKS (hand routes, fixed before the autorouter), KEEPOUTS, TEXT (silkscreen), all in board-relative mm. - POURS = [(net, layers, True)] for solid (no thermal relief) pours. - boardwright build <board> pads prints every pad's position and net, which is what ZONES and TRACKS are written from. - boardwright pins Lib:Symbol [Lib:Footprint] prints pins and pads. - Copper plots (out/F_Cu.png, out/B_Cu.png) in the gate. - boardwright/review.py, part of verify: every ZONES polygon must fill as one piece; any foreign net routed through a polygon is named with a location; REVIEW in board.py declares pad-to-pad distance limits, net length limits and pair skew. - gen prints Freerouting's unrouted count; verify fails on a drc.json older than the board. - c.add(..., bom=False) for mechanical parts.

Method that worked: join every pad of the power loop by hand (TRACKS), lay the polygons over them (ZONES), stitch GND next to each return (VIAS), and leave only the small signals to the autorouter. The sense line is hand-routed round the quiet side on the top layer so the bottom GND plane stays whole under the power stage.

Left open - No current-capacity or thermal calculation; polygon widths are by judgement. - Input loop area is not measured, only pad distances. - The SW node runs under the input capacitor body to keep the loop tight. TI's own layout example does not do that; worth a second opinion. - Not built. Regulation, ripple and thermal behaviour are unverified.