AOI for Solar Cell Manufacturing — Defect Detection Guide

Automated optical inspection (AOI) is the fastest inspection method on a PV production line — and the only one that works in the visible spectrum with standard industrial cameras, no current injection and no laser excitation required. But its speed comes with a trade-off: AOI sees only what is on the surface. This guide explains where AOI fits in solar cell and module manufacturing, what it catches, what it misses, how AI-based defect classification is changing the game, and how AOI layers with EL and PL to form a complete inspection strategy.

What AOI does — and what it does not

AOI captures high-resolution images of a solar cell or module surface under controlled visible light, then compares each image against a reference — either a CAD dataset, a golden sample, or a trained defect model. Anomalies are flagged, classified, and logged with coordinates for repair or rejection.

What AOI sees	What AOI misses
Grid finger width deviation from specification	Internal microcracks hidden under metallization
Busbar misalignment and print smearing	Electrically inactive regions (current still flows, surface looks normal)
Edge chips, chamfer defects, and corner cracks	Bulk material defects that do not alter surface appearance
Surface contamination and foreign particles	Passivation layer degradation
Color variation across cells within a batch	Recombination-active defects (PL territory)
Paste residue and incomplete edge isolation	Shunts and high-resistance regions (EL territory)

The rule of thumb: if a defect changes how the cell looks, AOI can catch it. If it changes how the cell behaves electrically but leaves the surface intact, you need EL or PL. This is why production lines run all three — not one instead of the others.

Where AOI fits in the production line

AOI is deployed at multiple stages because the type of visible defect changes as the cell moves through processing:

Production stage	What AOI checks	Why it matters here
Incoming wafer	Saw marks, edge quality, surface contamination	A chipped wafer edge will propagate cracks during thermal processing
Post-texturing	Etch uniformity, residual saw damage	Non-uniform texture = non-uniform light trapping = efficiency loss
Post-metallization	Grid finger width, busbar alignment, paste coverage	A finger printed 10 μm too narrow collects less current; this is the highest-value AOI station
Pre-firing	Print registration, paste smearing	Catch misprints before firing locks them in
Cell sorting	Color binning, visual grade	Module buyers expect uniform appearance across a shipment
Post-stringing	Cell-to-cell gap, ribbon alignment	Misaligned ribbons create stress points that crack during lamination
Post-lamination	Edge seal integrity, glass defects, frame alignment	Last visual check before the module ships

The post-metallization station is usually the most heavily instrumented: it catches print defects at the point where rework is still possible. Once the cell is fired, a misprint becomes scrap.

AI-based defect classification: beyond template matching

Traditional AOI uses template matching — compare the image to a golden reference and flag anything that differs by more than a threshold. This works for dimensional checks (finger width, busbar position) but generates false positives on anything that looks unusual but is harmless — a dust speck, a cosmetic mark, a legitimate process variation.

Deep learning models trained on labeled defect libraries are changing this:

Traditional AOI	AI-based AOI
Template matching against a reference image	Classification against a trained defect taxonomy
Flags any deviation above a threshold	Distinguishes a microcrack from a cosmetic mark
High false-positive rate → good cells wasted	Lower false-positive rate → higher yield
Hard thresholds → needs re-tuning when process changes	Model retrains on new defect types
Pass/fail output	Defect type + severity + coordinates → enables repair routing

The economic case: a line producing 10,000 cells per day with a 0.5% false-positive rate throws away 50 good cells daily. At $0.50 per cell that is $25/day, or roughly $9,000/year in unnecessary scrap. AI classification that cuts false positives by half pays for itself within months.

How AOI, EL, and PL work together

No single inspection method catches everything. The three methods are complementary by physics — they probe different properties of the same cell:

Method	What it probes	What it catches	Speed
AOI	Surface appearance (visible light)	Print defects, edge chips, color, contamination	Fastest
EL	Current flow (electroluminescence)	Microcracks, broken fingers, shunts, soldering defects	Moderate (needs electrical contact)
PL	Material quality (photoluminescence)	Bulk defects, carrier lifetime, passivation quality	Moderate to slow (depends on SNR requirement)

For a detailed explanation of how EL and PL work, see our EL vs PL Inspection guide. The key point for production planning: AOI is the throughput gate, EL is the electrical gate, and PL is the material-quality gate. Running only one or two leaves a category of defects uninspected.

A typical 100 MW/year line deploys them in this sequence:

Incoming wafer → PL (material screening)
Post-metallization → AOI (print quality)
Post-firing → EL (electrical defects)
Cell sorting → AOI (color binning) + EL (final grade)
Post-stringing → EL (interconnect integrity)
Post-lamination → AOI (visual) + EL (module-level cracks)

Each station catches what the previous one could not. Removing a station does not save its cost — it shifts that cost downstream, where rework is more expensive or impossible.

Which TDM instruments support AOI and combined inspection

TDM's inspection portfolio covers AOI, EL, and PL at cell and module level. The instruments below are designed for inline production or offline quality control, not laboratory characterization:

SE68 EPL/AOI/Degradation Integration Instrument: EL + PL + AOI + Degradation for cell (310×310 mm). Combined platform: 5K linear-scan PL, area-camera EL/AOI, 1-sun spectral degradation — three inspection modes on one instrument.
SE80 Hi-Resolution AOI & EPL Mapping: Dedicated high-speed optical inspection for grid print quality, edge defects, and color binning — the highest-throughput AOI in the TDM line.
SE60 EL/PL & Defect Mapping: 5K-resolution PL mapping plus area-camera EL for defect mapping on samples up to 310×310 mm.
SE69 Hyperspectral EL/PL: Spectral resolution on top of spatial EL/PL mapping for cell/module work.
SE200x Inline PL Mapping & AOI: Research-grade combined EL/PL characterization with overlay diagnostics for cell inspection.
SE20 Defect Mapping Instrument: High-speed PL imaging for inline material quality sorting.

Important: Inspection throughput, camera resolution, defect classification capability, and automation level (manual/semi/fully automatic) depend on the specific configuration and software package deployed. Confirm with TDM application engineers for your production line speed and defect sensitivity requirements.

View all PV inspection instruments →

Frequently asked questions

Can AOI replace EL inspection?

No. AOI sees surface defects visible under ordinary light. EL detects electrically inactive regions — microcracks, broken fingers, shunts — that may leave no visible trace on the surface. A cell that passes AOI can still fail EL. Production lines use both.

How fast is AOI compared to EL and PL?

AOI is the fastest of the three because it uses standard visible-light cameras and requires no electrical contact or laser excitation. It is the only method that can run at full production-line throughput without becoming the bottleneck.

What types of defects can AOI not detect?

Internal microcracks hidden under metallization, bulk material defects, passivation degradation, shunts, and high-resistance regions. These require EL or PL. AOI also cannot distinguish a cosmetic mark from a structural crack — this is where AI classification adds the most value.

Does AI-based AOI eliminate the need for EL?

No. AI improves classification accuracy within what AOI can see, but it does not extend AOI's physics. A crack under grid metallization that produces no surface feature is invisible to AOI regardless of how smart the algorithm is.

At which production stages is AOI most valuable?

Post-metallization (catch print defects before firing) and cell sorting (color binning for module uniformity). The post-lamination AOI station is also important as the final visual gate before shipment.

Which TDM instrument combines AOI with EL and PL?

SE68 integrates EL, PL, and AOI plus 1-sun spectral degradation on a single platform for 310×310 mm samples. This allows all three inspection modes on the same sample without repositioning.

Related guides and products

Request a measurement configuration