Automated Macro Inspection Offers Sensitivity, Throughput

Traditionally, manual after-develop inspection has been troubled by endemic human factor problems. With a human in the loop, inspection can be inconsistent, and results can vary significantly from operator to operator. Other challenges lie in the fact it is difficult for the human eye to detect defects below 110 nm; imaging data cannot be stored; and throughput/cost considerations permit only random sampling inspection, which complicates and delays sometimes crucial feedback to the manufacturing floor.

Nikon Instech (Kanagawa, Japan) has introduced an automated macro inspection system, the AMI-3000, that addresses several problems confronting existing systems. One of these is insufficient visibility. Normally, visual macro inspection systems use diffracted, not reflected light, which makes them subject to defects such as uneven color and brightness. But when the pattern pitch is 0.1 µm or smaller, they do not provide the operator with sufficient diffracted light. Other factors have come into play with the introduction of thin-film coatings. Minute variations in resist patterns, which previously did not affect the actual pattern, often show up as defects. Another problem solved by the new platform is that it can point out the limits of trench chart management, which is influenced by pattern miniaturization and high integration.

The Nikon system is a high detection sensitivity macro inspection system that enables the inspection of a wafer’s entire surface and quantifies inspection results. (Source: Nikon)

The Nikon system was developed using three concepts: high detection sensitivity for a macro inspection system; high throughput, enabling inspection of every wafer's entire surface; and the capability to quantify inspection results based on the same reference criteria. Detection sensitivity is supplied through the use of diffracted and scattered light, which provides a 95% defect capture rate for scattered light inspection within its dark-field inspection system, and 80% with the diffraction mode.

The tool captures the entire surface of every wafer in one image, while its ≥150 wph throughput enables 100% inspection without introducing delays into the production line. Depending on process changes, artificial intelligence algorithms can exercise auto-rework judgment, including automatic defect classification. The system offers 5× the defect capture rate of traditional systems using human operators.

The platform's diffracted light detection system provides it with excellent sensitivity for detecting Z-axis pattern variations, which result from factors such as defocus or uneven coating. Even when the inspection is within the tolerances of CD-SEM dimensional measurements, the platform detects the Z-axis pattern variations as macro images. Because the system's proprietary algorithms recognize diffracted light from the top pattern layer, it is possible to identify defects in underlying patterns. Compared with other systems that provide sampling and point inspection, the new platform can deliver an abundance of inspection data across a broad range of criteria. The AMI-3000's inspection results can be fed to other measuring and inspection systems. The platform's learning function quantifies the acceptable value of good wafer image results for each process. This provides sufficient versatility to respond to process changes and deliver stable inspection results, making it possible to quickly detect and repair process equipment problems.

The platform has auto-rework judgment software for each process. This is an important function that enables users to specify their own rework criteria for each process, and adjust the threshold of the defect criteria in the recipe. This, in turn, makes automatic rework judgment in wafer and lot units possible.

The AMI-3000 is in use at Toshiba's Yokkaichi fab, where an 80% pattern defect capture is reported vs. 14% with conventional human inspection.

For additional information on inspection, measurement and test, go to www.semiconductor.net/imt.