# Design and verification journal

## 15 November 2026

| Problem or decision | Investigation and solution | Status |
|---|---|---|
| The project directory was empty | New project. The circuit blocks, schematic, PCB and manufacturing lists use consistent component and net identifiers. | Archived and reopened in Fusion |
| First connection to MCP Fusion failed | I launched the installed copy of Autodesk Fusion and retried the document read. Fusion 2706.1.17 responded. | Solved |
| Nordic’s RF reference design places ground close to the RF trace | The Ebyte module retains the QFN silicon and RF circuit on its own substrate. The antenna area has been verified on both photoplots. | Geometry and stock verified. RF in enclosure requires measurement |
| The panel's RAM is lost upon interruption of the main domain. | The firmware reconstructs the previous image and the new image on Waveshare 4.2 V2/SKU 13186. When the waveform is interrupted, it requests a complete refresh. | Transfers and recovery have passed software tests. Optical test D02 remains to be performed. |
| The RT6160 package has a 0.4 mm pitch | The fan-out, lands, and local step of 0.08 mm of the stencil have been verified for Standard PCBA. | Checked in DRC and on photoplots |
| Module with antenna on a two-layer PCB | The Ebyte manual for E73-2G4M08S1F specifies nRF54L15-QFAA, two crystals, and a PCB antenna. JLC catalog reported 104 pieces in stock, 101 orderable. Free space is allocated for the antenna and radio measurement on the assembled board is planned. | Chosen solution |
| The first RTC proposed had zero inventory | Switched to PCF85063AT/AY and 32.768 kHz crystal. Firmware uses address 0x51. | Corrected |
| The Fusion import hid NAME and VALUE in the first export of the display interface | Re-exporting from Fusion showed that instances had been converted with smashed=yes but without attribute positions. The generator now writes both attribute positions explicitly for each instance. The groups were laid out more compactly, including the test points. | The seven final sheets have been inspected in Fusion and in PDF |
| The EAGLE script export from Fusion does not accept an absolute macOS path as the first argument. | The native Electron.runScript command accepts a relative path to the current Fusion directory. A test PDF has been generated, two pages inspected. | Solved |
| Aligning the 4 internal ground contacts of the module | The manufacturer's annotated drawing and the JLC assembly model enabled a correction of 0.30 mm to the central group from the initial CAD model. | Corrected before final routing |
| The generic Waveshare schematic assigns different names to pins compared to the B/W panel V2 list. | The mechanical drawing and the V2 pinout table match. Pins 1 and 4 remain isolated, pin 5 is VSH2. The connector and the two native netlists are automatically verified after physical pad placement. | Corrected in the library and schematic |
| The Waveshare partial-refresh example relies on panel RAM retained from the previous session | RRAM retains the confirmed scene. After restart, the firmware reconstructs the entire old image plane at command 26h and the new image plane at 24h. We checked the driver sequence, mode register and interruptions during journal writes. | Firmware tests performed. Optical lot verification requires D02 |
| The first lands of C303 and the bodies of L301/D303 were too close together | C303 and R302 have been repositioned. D303 has been moved by rerouting the SW trace, and access to TP301 has been relocated. I moved Q301’s pin 1 silkscreen dot after reviewing the Gerber plot. | Geometry, silkscreen and paste checked on the final revision |
| At final DFA inspection, L301.2 was only 0.025 mm from the nominal body of Q301. | L301 was moved 0.35 mm to the left and both 0.35 mm traces were reconnected. Measurements in the native design show 0.375 mm clearance to the nominal body, 0.305 mm to the maximum onsemi body, and at least 0.220 mm between the Q301 pin-one dot and adjacent mask openings. | Verified. Connectivity and the 24 FPC apertures have been rechecked on f7558226 and on the final board 0da00a40 |
| At each end of the tolerance range, the flex width exceeds the FH12 recommendation by 0.03 mm. | On receipt of the prototype, measure the flex width and check it against 12.43…12.57 mm. The Hirose drawing also recommends a local 0.15 mm stencil thickness at the connector. | Acceptance and process requirements documented |
| The effective capacitance of ceramic capacitors decreases at the operating voltage. | I have replaced the ±20 V capacitors with 100 V / 1206 and the transfer capacitor with 50 V / 1210. I have interpolated the Samsung curves and simulated using ONSEMI models. Inverted D302 produces a positive VGL, detected as a defect during testing. | Nominal, low-voltage, convergence and fault cases completed |
| The additional simulation with trace R/L and a synthetic piecewise-linear control signal stopped numerically at 34.218 ms. | The limiting approximation was changed to a smooth characteristic and explicit ESR values of 30–40 mΩ were added. Ngspice completed the 40 ms run; the maximum terminal-to-reservoir difference in the selected sensitivity case was 1.127 V. Both logs are in verification/display/boost. | Simulation rerun and completed. The waveform on real PCB traces still needs an oscilloscope check. |

### Electrical and mechanical integration

- The first ERC on the complete schematic flagged ITERM at U102 unconnected and two output paths for the same physical stage of U102/U103/U104. R123 explicitly programs the termination current. The symbols use a single functional pin for duplicate outputs and require connection of all physical land pads. The pin names and types for power supplies are retained from the technical datasheets.
- The first Fusion DRC run on routed copper found 123 traces below the general minimum width and 42 below the power class limit. The export from Fusion confirmed the widths specified by autorouter, including 0.090 mm. Routing, mask, marking, and separation via measurement bypass resistors R124/R125 corrections are detailed in the layout journal.
- Independent verification identified an initial SPV1050 resistor value of 8.2 MΩ, outside the recommended 10–20 MΩ range specified by ST. The 11 MΩ network was rechecked for tolerances. The CE stage uses two separate diodes to isolate the comparator outputs. Power supply checks
- The first query function of the 3D model does not have a modeling viewport in the ECAD editor. I used native Fusion Electronics captures for 2D views and the Design API in the 3D PCB document. Package models are inserted into the associated PCB components. Generic height profiles of 0.1 mm created implicitly at import do not describe the height of the parts.
- The first comparison of the updated schematic to an older PCB copy failed when the regulator nets changed. The checker compares each land against the schematic, PCB and electrical definition, including both contacts of the grouped outputs. The Fusion project’s working revisions are in the Engineering checks folder.
- After a complete power cut to the display, the firmware rewrites the 15,000-byte image planes and records the optical transition in the journal. Tests injected power cuts during RRAM writes and errors on SPI. Firmware log

- The SimSurfing curves indicate that the initial 4.7/10 µF components at BAT/OUT could drop below 4.7 µF at operating voltage. Components C109, C110, and C113 have been upgraded to 22 µF/16 V in the same footprint. The recalculated range for BAT/OUT is 5.49…5.89 µF under combined tolerance and temperature conditions. The reel version of U105 had four pieces in stock against a feeder minimum of five, whereas the identical MCP6542-I/MS version had 49. Power calculation and verification

- Fusion DRC reported a trace termination displaced from the center of pin U101.20. The measurement resistor R126 separates PV_IN from PV_IC_IN, retaining capacitor C103 and the R105 divider near the SPV1050 input. A 0.4 mm geometry enables local fan-out as specified. In ERC, the change of identifier for U101.IN_HV was explicitly tracked. ERC review

### Rechecking and archiving revision A

- The initial photoplots contained 30 separate ground regions. After the connections to the plane and the return vias were reworked, DRC found an off-center trace endpoint at C417.2. The trace was extended by 0.267 mm. On the delivered revision, both DRC engines reported zero findings. Reconstruction from the copper Gerbers and drill data recovered all 92 nets without opens or shorts. PCB iterations · Fabricated copper and holes
- A timing test revealed that programming the RTC to 15 minutes after each refresh would accumulate the time spent in active state. The firmware now programs quarter-hour intervals of the clock. None of the 96 tested variable durations shifts the grid. I separately verified all 86,400 possible rearm moments. Firmware tests
- The CAM working file included the line width when measuring the outline. The delivered export gives the nominal 112.00 × 84.00 mm outline on the milling tool centreline. The actual outline geometry, all 214 holes and all traces were compared against the native board. Gerber Job file provenance · CAM acceptance
- A F3Z test archive contained the schematic and PCB, while the 3D document was available separately in Fusion. After saving a new version of the PCB with a provenance attribute, I verified in the cloud the PCB → 3D document link prior to export. I then inspected the nested archives and uploaded the delivered copy to another folder. Fusion reopened the project, the populated board assembly, and the display reference bodies. Verified archive · Reopening in Fusion
- Manufacturing revision passed 36 DFM/DFA/DFT checks. The JLCPCB list had stock available for two units with the required reserves as of November 15, 2026, 12:52 UTC. BME280 had 151 pieces in the SMT inventory and zero at pre-purchase. Bench tests, including display refresh tests on the actual panel lot, reflow height measurement and RF in enclosure testing, still require populated prototype boards. Final verification · [Bench sheet](verification/independent/bench-record.en.md)

## 15 November 2026 · enclosure and desk stand

| Observed problem | Modification and test |
|---|---|
| The 12.50 mm narrow end of the flex did not account for its 36.70 mm shoulder. The clearance around the active glass area was asymmetric. | I redrew the outlines according to the Waveshare layout and the Hirose cross-section. The window frame follows the active area of 84.80 × 63.60 mm, with a 1.50 mm margin on each side. I tested the transition contour and the entire window at normal and oblique incidence. |
| The upper-right screw post intersected the glass with a volume of 2.06 mm³. Some of the frame ears, rails, and posts overlapped other 3D-printed parts. | I extended the body by 2 mm, moved the screw toward the corner, cut back the frame and moved the rail attachment. The six screw bosses meet at the parting plane. The M2.5 × 12 screws engage the inserts by 5.25 mm. |
| With the board already fastened to the front enclosure, the glass cannot move the 4.40 mm needed to insert the flex into J301. | The glass and the board are connected before being lowered together into the enclosure. I have increased the spacing of the rigid stoppers and have provided six soft stoppers that fit the fully connected glass. The simplified electronic model of J301 fills the internal cavity; I measured the actual geometry from the Hirose cross-section and recorded the volume from the model. |
| The solar wire tie eye intruded into the corner of the 13 × 7 mm USB-C plug envelope. The battery wires intersected components on the board and stops on the cover. | I moved the tie eye and both insulated wires, added the solder paths and rerouted the three battery wires at constant length with a 2.60 mm bend radius. For servicing, lift the lid parallel to the board by no more than 12 mm before disconnecting the plug. |
| The initial STL files sat at slightly different heights on the print bed. Each initial button export contained a facet with two identical vertices and an isolated island at the start of the flexible tip. | I calculated the print-bed placement from the exact geometry and adjusted the button tip by 0.03 mm. I separately checked surface topology and the overlap of each deposited line with the previous layer in the G-code. Both shell halves and both rails use supports built from the print bed. |
| The solar product photo showed a single collector band. Initial views had strong screen reflections and overly tight framing. | I modelled the visible solar-panel features from the ABRA photograph and checked the display against the actual firmware raster. I adjusted the lighting and cameras to show the entire enclosure, buttons and plug insertion path. |

The enclosure folder contains the documents for the delivered revision. Printing, force measurements and reassembly, sensor test in enclosure, and radio measurements on the assembled station have worksheets for the first unit.

## 15 November 2026 · USB-C recess revision

| Observed problem | Correction and test |
|---|---|
| In an oblique Fusion view, the PCB and components were visible through the side opening. The axis estimated from the whole connector body was 0.05 mm below the centreline of its metal opening. | I sectioned the GCT connector model at the lip and rebuilt the recess on the centreline Y = 67.000, Z = 1.725 mm. The printed front forms the walls and the step for the plug body. A black dielectric mask covers the clearance between the port and the plastic. The original native view is retained for comparison. |
| Even with the front mask fitted, the solder joints were visible at certain angles through openings in the J101 metal shield. | I checked which surface each ray encountered. I formed thin shields outside the plug envelope and lowered the recess ceiling locally. Tests with the solder joints, wires and mask fitted or removed use the geometry of the populated board. |
| The plug overmould, the bend in the solar leads and the attachment eye occupied the same space. | I have separated the traces, smoothed both outputs from the panel terminals, and inserted a step for the 13 × 7 mm cable. The nominal face-to-step clearance is 1.08 mm. The guide requires comparing the track against the same plug placed in J101 prior to PCB assembly. |
| Print support remained in a hard-to-reach space below the recess. At a grazing angle, the edge of the solar panel was visible through the cover seam. | A footprint connects the front printed board. A wider rib and a bridge close the gap's base. I inspected the print toolpaths and checked support removal, lid clearance and visibility along both diagonals of the seam in the model. |
| In an intermediate version, 41 rays at grazing angles reached the coating on the PCB alongside the connector leads. The two exposed strips were invisible in the usual renderings. | I have symmetrically reduced the lateral sides of the mount right after the entry. At 1.25 mm from the exterior, the free width is 17.12 mm. I have retained the external opening for cable guidance and included the 41 sightlines in the regression tests. |
| In the curved areas of the recess, three 0.45 mm perimeters did not adequately cover the 1.35 mm layer facing the cable. | I sliced the front enclosure separately using four 0.41 mm perimeters and checked deposited material at 84 points along the two walls. The specific slicer profile for the front is included. |
| A slightly oversized and misaligned mask could pass the centering and bonded-area checks while still leaving an oblique line of sight past a corner of the steel shell. | The acceptance check projects the measured mask outline onto the metal lip over the full range of viewing angles admitted by the recess. At least 0.005 mm of coverage must remain after subtracting measurement uncertainty. The checker uses measured outlines and the PET-to-steel distance. Recenter or recut a rejected mask and measure it again. |

Port comparison and section checks · Assembly verification on delivered revision
