EN
Back to the archive

The encyclopedia · Engineering & Operations · Technical decision · 2021–2024

LG Innotek taught AI to predict camera-module defects before they happen

LG Innotek's AI process recipe predicts defects and tunes machine settings, cutting key inspection failures up to 90%.

LG Innotek

the move

LG Innotek, the world's largest camera-module maker, began developing AI that forecasts process defects in 2021, because high-specification modules often suffer low yields at the start of mass production, and rework costs rise with every later step.

In 2023 it became the first in the industry to apply an 'AI process recipe' to its high-spec camera-module lines: the AI reviews the whole initial process, predicts where defects are likely, and simulates machine intensity, conveyor speed and temperature to set the best conditions.

Applied to the flagship product, the recipe — trained on tens of millions of data points — reduced the defect rate in key inspection items by up to 90%, stabilized initial yield, and strengthened cost competitiveness. LG Innotek said it planned to extend the approach to semiconductor processes.

why it works

  • The AI sees patterns across tens of millions of data points no engineer could hold in mind.
  • It acts before defects exist, so inspection is no longer the only safety net.
  • Simulating variables makes the fix concrete: settings, not hunches.
  • Stable ramp-up yields are worth more than a theoretically perfect process.
the payoffPredict the defect, set the recipe before you runclever

what transfers

When a process is too subtle for human tuning, let a model trained on the plant's own history set the parameters — fix the recipe before the part is made, not after inspection fails.

what came after

After the 2023 rollout, LG Innotek reported the up-to-90% defect cut and said it was extending AI process recipes to semiconductor and other processes; the approach became part of its pitch as it expanded into automotive camera modules for robotaxis and other physical-AI sensing.

references

spotted an error? The archive wants to know.

same kind of clever