Photolithography Oxide Masking 1957 Classify Accept Release

# Classify, Accept, Downgrade, or Reject: Where a Wafer's Worth of Process Becomes a Batch of Graded Parts

## 1. Why the Product Line Is Discovered, Not Designed

This step packages the diced die that passed Step 16's probe, gives each one a final test, and sorts the results into performance grades — and nothing about any individual device changes here; this step only decides what each one is permitted to be sold as. What makes this step specifically a creature of 1957, rather than a procedure this project could have written for any earlier series, is that a single wafer now yields a population of hundreds of nominally identical devices rather than a handful. The output of this step is a distribution, not a sequence of individual verdicts, and grading becomes an act of partitioning that distribution rather than judging devices one at a time:

$$f_{\text{grade }k} = \int_{p_k}^{p_{k+1}} \rho(p)\, dp$$

where $\rho(p)$ is the measured probability density of some device parameter across the whole batch, and $p_k$ the boundary separating one grade from the next. A manufacturer's product line, read this way, is *discovered* inside the distribution rather than designed in advance — the grades offered for sale are chosen to match where the devices actually landed, a mode of operation that only becomes possible once a single process run produces enough devices to have a distribution at all.

A Product Line Read Off a Histogram grade boundaries drawn where the batch actually landed PARAMETER DISTRIBUTION ACROSS THE BATCH reject grade B grade A bimodal: two mechanisms, not one fgrade k = ∫ ρ(p) dp, over pk to pk+1 — the grades are read off the shape, not assumed in advance a second hump sends the investigation back to the wafer maps from step 16

## 2. Real Diagram: Seven Series, One Discontinuity

This is the closing article of the 1957 series, so its second diagram steps back from any single step to show the whole lineage this project has documented — from the 1947 point-contact transistor through this series' arrival at batch-processed, photolithographically patterned devices.

Seven Series, One Point Where the Line Bends 1947 through 1957, one device at a time until this series arrived 1947 point-contact 1951–53 grown, alloy, surface-barrier 1954 diffused-base 1956 silicon mesa, drift transistor 1957 photolithography, batch processing arrives one device at a time → hundreds at once andrus & bond, lathrop & nall — fairchild founded — philco's madt bridges alloy and diffusion the line bends once, here, and every later decade of this industry runs along the new slope

## 3. Why Grading a Population Is Not Grading a Device

The 1956 process also ended with a sorting step — Step 20 of that series classified each drift transistor as accepted, downgraded, or rejected. The question that step asked was whether a given device had successfully received the drift field the process intended for it: a judgment about one device at a time. This step asks a different kind of question entirely — what does this *population* look like, and how should it be divided — because for the first time in this project's history, "the population" is large enough to have a shape worth asking about. That shift, from judging individual devices to characterizing distributions, is the final and most lasting consequence of batch processing, and it is where the semiconductor industry's modern character fully arrives: grading as statistics, not as inspection.

Step 18 does not finish a device; it closes the series by turning what seventeen steps built, one wafer's worth of silicon at a time, into a population whose shape — not any single device within it — is now the thing this process is actually in the business of producing.

Take photolithography oxide masking 1957 classify accept release further

Ask the copilot about this term, or have our engineers assess it against your process.

Ask CFSGPT about this → Request a technical assessment → Open the CFS app