As optical lithography continues to extend into low-k1 regime, resolution of mask patterns continues to
diminish. The adoption of RET techniques like aggressive OPC, sub-resolution assist features combined with
the requirements to detect even smaller defects on masks due to increasing MEEF, poses considerable
challenges for mask inspection operators and engineers. Therefore a comprehensive approach is required in
handling defects post-inspections by correctly identifying and classifying the real killer defects impacting the
printability on wafer, and ignoring nuisance defect and false defects caused by inspection systems. This paper
focuses on the results from the evaluation of Automatic Defect Classification (ADC) product at the SMIC mask
shop for the 40nm technology node.
Traditionally, each defect is manually examined and classified by the inspection operator based on a set of predefined
rules and human judgment. At SMIC mask shop due to the significant total number of detected defects,
manual classification is not cost-effective due to increased inspection cycle time, resulting in constrained mask
inspection capacity, since the review has to be performed while the mask stays on the inspection system.
Luminescent Technologies Automated Defect Classification (ADC) product offers a complete and systematic
approach for defect disposition and classification offline, resulting in improved utilization of the current mask
inspection capability. Based on results from implementation of ADC in SMIC mask production flow, there
was around 20% improvement in the inspection capacity compared to the traditional flow. This approach of
computationally reviewing defects post mask-inspection ensures no yield loss by qualifying reticles without the
errors associated with operator mis-classification or human error.
The ADC engine retrieves the high resolution inspection images and uses a decision-tree flow to classify a given
defect. Some identification mechanisms adopted by ADC to characterize defects include defect color in
transmitted and reflected images, as well as background pattern criticality based on pattern topology. The final
classification uses a matrix decision approach for achieving the final defect disposition. As a first step for
qualifying ADC for high volume production, the defect classification results obtained with ADC are compared
to the operator classification. Matching rates of greater than 90% were achieved when compared to operator
defect classifications. Moreover, no critical defect has been missed. ADC performance was proven to be
qualified for deployment in full volume mask manufacturing production flow.