Die sorting and inspection as a service does two things: it separates good die from bad, and it produces a structured record of why each die was called what it was called. The first is table stakes — any inspection step sorts. The second, the per-die record that travels with the parts, is what makes it a data service rather than a bin of die, and it is what a program building traceable hardware is actually buying.
What gets checked
Each die is inspected against a defect and dimensional spec agreed up front, using the tools that match the failure modes that matter for that die:
- Optical inspection and AOI — surface defects, scratches, particles and contamination, edge and corner chips from dicing, cracks, and metallization or passivation anomalies.
- Metrology — critical dimensions, die size, and feature measurements against the drawing.
- SEM — where a defect needs resolution beyond optical, for metallization and fine-feature review.
- X-ray — for internal and subsurface features where a part calls for it.
The check list is defined per program, because "inspected" means nothing without the criteria it was inspected against. What counts as a reject on a power die is different from a MEMS die, and the spec is agreed before the first die is looked at.
What gets logged
Every die inspected becomes a record, not just a pass or a fail. A per-die record ties together:
- The die's location — wafer map coordinate or tray position — so a result maps back to a physical part.
- A die identifier, read by OCR where the die carries a mark, so the record follows the die downstream.
- The classification: pass, fail, or a defect bin, against the agreed spec.
- The evidence — the inspection images and measurements the call was made from.
Assembled across a wafer or a lot, those records are a wafer map and a traceability trail: not just how many die passed, but which die, where, and why. That is the difference between a sort and a data set you can act on. More on why structured inspection data is worth more than a pass/fail count is in what structured inspection actually tells you.
How the sorting is decided — engineer-in-the-loop
Classification is data-driven and AI-assisted, with a machine-learning model as the engine that proposes a call for each die from its inspection data. It does not grade die on its own. An engineer is in the loop: the model flags and sorts, the engineer sets the criteria, reviews the edge cases, and owns the final classification rules. The software augments engineering judgment on repetitive, high-volume sorting; it does not replace it, and it does not make process calls on its own. We built this inspection automation in-house, and the reasoning behind that is in why we built our inspection software in-house.
Where to start
Three facts scope a die inspection and sorting job:
- The defect and dimensional spec — what counts as a reject, which is what "inspected" is measured against.
- The die format — on wafer, in tray, or as singulated die, and whether it carries a die-ID mark.
- The record you need back — a sorted lot, a wafer map, a full per-die traceability data set, or all three.
We run die inspection, sorting and traceability as a standalone data service in Halethorpe, Maryland, on US soil — engineer-in-the-loop, with the inspection record delivered alongside the sorted die. More on our die inspection and traceability page.
Answered.
What is die sorting and inspection as a service?
A service that inspects each die against an agreed defect and dimensional spec, separates good die from bad, and delivers a structured per-die record alongside the sorted parts. The sort is the obvious output; the record — which die, where, and why each was classified as it was — is what makes it a traceability data service.
What gets checked during die inspection?
Surface defects, particles and contamination, dicing edge and corner chips, cracks, and metallization anomalies by optical inspection and AOI; critical dimensions by metrology; finer defects by SEM; and internal features by X-ray where a part calls for it. The exact criteria are set per program, because an inspection result only means something against the spec it was checked against.
What is logged in a per-die record?
The die's location as a wafer-map coordinate or tray position, a die identifier read by OCR where the die is marked, the classification against the agreed spec, and the inspection images and measurements the call was made from. Assembled across a lot, those records form a wafer map and a traceability trail.
Is the die classification automated?
It is data-driven and AI-assisted, with a machine-learning model as the classification engine, but it is engineer-in-the-loop, not a hands-off grader. The engineer sets the criteria, reviews edge cases and owns the final rules; the software augments judgment on repetitive high-volume sorting rather than replacing it or making calls on its own.
Do you offer die sorting and inspection as a service in the US?
Yes — die inspection, sorting and traceability as a standalone data service in Halethorpe, Maryland, on US soil, with an engineer in the loop and the inspection record delivered alongside the sorted die.