See.AI™ Semiconductor AOI Inspection
Upgrading AOI Inspection for a
Malaysian OSAT Manufacturer
A publicly listed semiconductor manufacturer in Bayan Lepas, Penang was experiencing declining yield rates from limitations of conventional AOI inspection software. PixeVision augmented their existing hardware with See.AI™ — lifting yield from 90% to 98.5% without replacing a single machine and hardware.
LOCATION
Bayan Lepas, Penang
INDUSTRY
Semiconductor / OSAT
SOLUTION
AOI Inspection Software
CLIENT
Semicon Client A (Confidential)
FINAL YIELD RATE
WASTAGE REDUCTION
CAPEX REQUIRED
THE CHALLENGES
Conventional AOI Software Hitting Its Limits & Costly Over-rejections
The manufacturer operated high-volume OSAT production lines relying on conventional AOI machines running traditional rule-based inspection software. As product complexity and throughput demands increased, the fixed-threshold inspection logic could no longer keep pace.
Rule-based AOI software works by comparing components against predefined pixel tolerances. When production variability — subtle die orientation shifts, minor surface variation, lighting inconsistency — fell outside these rigid thresholds, good components were incorrectly flagged as defective. The result was a chronic over-rejection problem driving yield rates down to 90%.
The hardware was functioning correctly. The inspection intelligence driving it was not. Replacing machines would not have solved the problem — the client needed smarter software, not newer equipment.
THE SOLUTION
See.AI™ — AI Inspection Intelligence Over
Existing AOI Hardware
Rather than replacing the client’s AOI machines, PixeVision deployed See.AI™ as an AI software layer running directly on the existing hardware. See.AI™ replaced the conventional rule-based inspection logic with a deep learning model trained on the client’s own production data — learning what a genuine defect looks like, rather than relying on brittle pixel thresholds that generate false rejections.
Existing AOI Camera Feed
See.AI™ receives the live image stream directly from the client's existing AOI cameras — no hardware changes, no downtime required.
Learned Model Evaluation
A deep learning model trained on the client's own production data evaluates each component — replacing rule-based thresholds with intelligent, data-driven decisions.
Accurate Pass / Fail Output
See.AI™ distinguishes genuine defects from acceptable variation — eliminating the over-rejection that was causing good components to be discarded.
Live Production Dashboard
Every inspection decision is logged in real time — providing shifts with immediate, granular visibility into yield rates, defect trends, and line performance.
PROVEN OUTCOMES
Three Outcomes,
One Deployment
Deploying See.AI™ as a software upgrade gave the client inspection accuracy that conventional rule-based AOI logic simply cannot achieve — without touching a single piece of hardware.
98.5% Yield — Revenue Recovered
Yield improved from 90% to 98.5% within 48 hours of deployment — good components that were previously over-rejected are now correctly passed, recovering significant lost revenue per production cycle.
OPEX Savings & Labour Redeployment
Manual secondary inspection — required to re-check components flagged by the legacy software — was significantly reduced. Skilled inspection staff were freed for higher-value quality assurance tasks.
Existing Hardware Fully Retained
No AOI machines were replaced. The problem was the inspection software, not the hardware — and See.AI™ solved it entirely through a software deployment, extending the usable life of existing equipment.
Baseline yield limited by rule-based inspection software generating high false-rejection rates on good components.
Stabilised yield rate — entire production line output quality restored.
Zero Hardware Replaced — Software Only Deployment
Ready to boost your line's yield without changing hardware?
Talk to our AI experts about your specific inspection challenge. We will show you what is possible with your existing infrastructure.