A wafer reader combines imaging, decoding and communication. Qualification must also establish that the returned identity belongs to the wafer expected at that station. Build the test around independently known identities and representative process stages, then verify how the host handles every outcome.
Establish the expected identity before tuning
Record the mark type, applicable format specification and reader support. Preserve leading zeros, character case and separators according to the approved identity rules. Keep wafer identity distinct from carrier, lot and slot; define how those records are related at the station.
The Cognex In-Sight 1750 overview describes wafer-mark decoding with different optical/lighting configurations and serial or Ethernet communication. That model-specific description helps frame an equipment brief; the required mark format and interface still need confirmation for the chosen configuration.
Create a test manifest whose expected identities were established independently of the reader under test. Include confusing character combinations and representative mark conditions. Reserve some examples for validation after tuning so the final evaluation includes unfamiliar samples.
Keep the optical and process context
Include process stages that change the appearance of the mark, along with expected wafer positioning and orientation variation. Save the source image and configuration for difficult cases. Compare focus, working distance, illumination and mark location before changing decoding settings.
A confidence score describes an algorithm's output under its configuration. Set its acceptance rule using labeled validation evidence. When checksum or redundant-recognition features are available, document exactly which mechanism is used: the Cognex CVL guide distinguishes actual checksums from its virtual-checksum approach. Neither replaces matching the result to the expected manufacturing context.
Report three different outcomes
On a narrow screen, scroll the table horizontally. Keyboard users can focus it and use the arrow keys.
| Outcome | Definition | Review |
|---|---|---|
| Correct identity | Decoded identity matches independent ground truth | Retain process-stage and configuration coverage |
| Wrong identity | A returned identity differs from ground truth | Investigate the image, algorithm and host acceptance |
| No-read | The configured read produces no accepted identity | Use a controlled retry or hold workflow |
Fictional validation: 1,000 attempts produce 990 correct identities, two wrong identities and eight no-reads. Decode yield is (990 + 2) รท 1,000 = 99.2%, while correct identification is 99.0%.
The wrong-identity rate is 0.2% and the no-read rate 0.8% in this sample. Report counts and denominators beside the percentages. Repeated images of one wafer provide different coverage from independent wafers spanning the process.
Set sample sizes and acceptance limits with the responsible quality team based on the consequences of an error. Preserve results by condition; an overall average can hide one problematic stage.
Validate the host transaction and recovery
Associate each trigger, image and result with one transaction. Test a late result arriving after the next wafer is presented, duplicate messages, a disconnect and a reader restart. Require explicit result validity and matching context before the host releases the next operation.
Define a bounded retry policy. If manual identity recovery is permitted, retain the failed result and image, the operator's independently checked evidence, the authorization and the final disposition. A free-text correction without provenance breaks the validation chain.
Re-run affected validation when the mark process, optics, firmware, decoding recipe or host matching changes. Link image evidence to the resulting production decision using the broader machine-vision validation approach. Keep identity checks visible when data proceeds to electrical parametric testing.
Use the working record
Download the editable wafer identification validation. Save a copy, complete the actual requirements and evidence fields, and review open issues with the responsible owner. The plain-text record includes instructions, units where needed, the fictional example, and source references.