Lilienfeld 1925 Inspect Film Coverage

# Inspect Film Coverage: Prove the Compound Film Spans the Intended Current Path

Step 16 answers one narrow but decisive question: does the visible compound film form an unbroken geometrical route from terminal 11 to terminal 12, across the intervening glass and over the exposed edge of control foil 13? It is an inspection step, not yet a materials-analysis or electrical-test step. A film can look continuous and still have the wrong composition, thickness, or resistance; those belong to Step 17. It can also conceal an unintended conductive bridge to the control electrode; that belongs to Step 18.

Lilienfeld's patent is unusually explicit about the required topology. Film 15 extends over both terminal coatings, the intermediate glass surface, and the edge of the aluminum foil. The film is described as minute in thickness and electrically conductive, with copper–sulfur compound given as the suitable example. That language defines the regions that must be covered, but it gives no numerical inspection limit and no evidence that the proposed device was fabricated successfully. A reconstruction therefore needs a declared inspection plan rather than a retrospective judgment from one attractive micrograph.

## 1. Coverage is a connected-path problem

Treat the intended current corridor as a region of interest, $\Omega_{path}$, running from a verified overlap on terminal 11 to a verified overlap on terminal 12. Let a calibrated image or fused image stack yield a coverage mask

The essential acceptance condition is not merely a high covered-area percentage. There must exist at least one connected film component $K$ that intersects both terminal-overlap regions and crosses every required transverse cut through $\Omega_{path}$:

Here $w_K(s)$ is the local usable width of that connected component normal to the path coordinate $s$. The declared $w_{min}$ must include imaging uncertainty and the smallest defect the chosen optics can resolve. A 99% area fraction can fail if its missing 1% is a hairline crack spanning the entire path; a lower area fraction outside the current corridor may be harmless.

## 2. Inspect five required regions, not one generic “active area”

Segment the Step 15 coordinate map into five coverage regions: left-terminal overlap, left glass gap, foil-edge crossing, right glass gap, and right-terminal overlap. Each has a different failure mechanism and optical signature.

1. Terminal 11 overlap. Confirm that film 15 overlaps sound terminal coating rather than merely approaching its edge. Edge beading can look dark while leaving a narrow exposed strip.
2. Left glass span. Search for pinholes, reticulated drying cracks, scratches, edge retreat, and discontinuities associated with the terminal step.
3. Foil-edge crossing. Use registered bright-field, dark-field, and oblique views because metal contrast can mask whether the compound film actually crosses the foil edge. A bright aluminum line is not proof of coating.
4. Right glass span. Apply the same full-width inspection as on the left; do not assume symmetry.
5. Terminal 12 overlap. Confirm continuous contact area and record any lift or blister that began during sulfurization or lead attachment.

Inspect the entire designed current-carrying width, including both lateral margins. A centerline scan can miss a crack entering from an edge. Conversely, an isolated pinhole far outside the declared corridor should be recorded without automatically failing the path. This is why the Step 15 coordinate system must be retained: film findings need physical locations that later measurements can revisit.

## 3. Acquire complementary contrast before classifying pixels

No single optical mode reliably separates copper–sulfur film, copper terminal, aluminum foil, glass, stain, and topography. Build a registered image stack under reproducible illumination:

• calibrated low-magnification mosaic for complete-device context;
• bright-field images for color and reflectance changes;
• dark-field or oblique illumination for cracks, lifted edges, and scattering debris;
• crossed-polarizer or spectral channels if they demonstrably improve film/substrate separation;
• focus stack or surface-height data where terminal steps and the fracture line produce shadows;
• an uncoated witness and representative coated reference acquired with identical settings.

Lock exposure, white balance, illumination angle, objective, and processing parameters. Preserve raw images. Any thresholded mask is a derived measurement and must be traceable to its source images and algorithm version. Hand retouching a mask to “look right” destroys the evidence unless every edit is separately logged.

For each candidate defect, report the local contrast-to-noise ratio

where $\mu_f,\sigma_f$ describe a verified film reference and $\mu_b,\sigma_b$ the relevant background. Low CNR means “unresolved,” not “covered.” Resolve ambiguous locations with another contrast mechanism or place the sample on hold.

## 4. Convert images into a defensible coverage decision

Register every image to Step 15 fiducials, correct lens distortion, and map pixels to physical coordinates. Then classify coverage with a frozen rule trained or tuned on independent references. Apply morphology only when its physical meaning is declared: closing a one-pixel gap for camera noise is different from erasing a real crack.

The useful local metric is not raw area fraction but the width of the connected path after accounting for positional uncertainty. If the measured usable width is $\hat w_K(s)$, the conservative lower-bound width is

Accept the geometrical path only if

and both terminal-overlap lengths meet their declared lower bounds. The factor of two allows uncertainty on opposing edges. When a crack width is below optical resolution, report it as an upper bound tied to the point-spread function; do not silently label it absent.

## 5. Distinguish defect classes by what they threaten

Use a small, stable defect taxonomy so results can be compared across samples:

Do not infer electrical continuity from visual connectivity. The segmentation resolution may be much coarser than the transport-limiting crack, and a visually continuous region may be chemically altered or highly resistive. Conversely, do not use a two-probe resistance measurement to overwrite the coverage map: Step 17 will characterize resistance, while Step 16 preserves where the film is and is not visibly present.

## 6. The foil-edge region deserves its own proof

The patent places film 15 over the exposed edge of foil 13. This is a special interface: aluminum reflectivity, a height discontinuity, fracture debris, oxidation, and local film chemistry can all change contrast. First locate the foil edge from the Step 15 geometry images, then inspect the same coordinates after film formation. Show the before/after registration and a transverse intensity or spectral profile. Require visible film signal on both sides of the edge and across the declared control-interaction width.

A dark line at the foil is ambiguous: it may be compound film, a shadow, the fracture seam, or a focus artifact. Rotate illumination, refocus, and use at least one independent contrast channel. If the film cannot be distinguished from the underlying metal, classify coverage as unresolved—not automatically present because the process recipe intended it to be there.

## 7. Sampling must match the failure geometry

An inspection plan needs a stated pixel size, optical resolution, field overlap, and probability of detecting a defect of specified size. If the entire active corridor fits within a calibrated mosaic, inspect it at full resolution. If sampling is unavoidable, use a risk-based plan that always includes terminal steps, both glass spans, the full foil-edge crossing, lateral film margins, and any Step 15 anomaly coordinates.

For repeated samples, summarize both device yield and defect spatial distribution. A recurring neck at one terminal edge implicates step coverage or shadowing; cracks aligned with the fracture may implicate topography or mechanical motion; random islands may implicate contamination or nucleation. These are hypotheses for process learning, not reasons to relax the coverage criterion after observing failures.

## 8. Minimum Step 16 record

For each device retain:

• device identifier and its Step 15 geometry-map reference;
• the declared current corridor, five required regions, and coordinate datum;
• raw registered images for every illumination or modality;
• calibration, pixel size, estimated resolution, exposure, objective, and processing version;
• film/background references and local CNR at ambiguous sites;
• unedited coverage mask $C(x,y)$ and connected-component labels;
• terminal-overlap lengths, connected-path width profile, $w_{min}$, and uncertainty model;
• every OPN, NEK, NOL, FEX, ISL, LFT, and UNR annotation with coordinates;
• pass, hold, rework, or reject disposition and reason code;
• links forward to Step 17 sampling locations and Step 18 short-check locations.

The primary historical source is [J. E. Lilienfeld, US Patent 1,745,175](https://patents.google.com/patent/US1745175A/en). It supports the required film topology—over both terminal coatings, the intermediate surface, and the aluminum edge—and the copper–sulfur example. The quantitative imaging, segmentation, uncertainty, and release rules above are modern reconstruction controls, not claims that Lilienfeld specified or performed them.

## Inspect Film Coverage’s Place in the Process Lineage

Step 15 released a mechanically intact, dimensionally mapped assembly. Step 16 overlays the compound-film map on those same coordinates and asks whether a single visible film component spans the intended terminal-to-terminal route, including the foil-edge control region, with uncertainty-aware width and overlap margins. Only a geometrically continuous and fully documented specimen proceeds to Step 17, which determines composition, thickness, uniformity, and resistance. Electrical shorts, baseline current, control leakage, modulation, and amplification remain deliberately unclaimed until Steps 18–27.