MEMS wafers, microfluidic wafers etc. need to be checked and inspected to detect imperfections (defect inspection) as well as measured in 2D and 3D (wafer inspection), in order to improve the control of production processes. The integration of defect detection into WAFERinspect, the high-resolution measurement system from Confovis, creates a process control instrument which can detect the defects and measure all structures in 2D and 3D.
Defect inspection with Confovis: Capturing smaller structures reliably
Defect inspection proves to be a challenge and is often cost-intensive on wafers with very different structures and a required resolution of less than 1µ. Thanks to using microscope optics, the optical measurement systems from Confovis are in a position to analyze defects on the entire wafer surface not only as a large image with shading correction, but also „die-by-die“. Cracks in deeper layers, as an example, can be detected in multi-layered structures by searching the 3D-image stack for defects in z-direction.
Rule-based fault detection followed by a blob analysis is as feasible as the use of artificial intelligence for analyzing the defects. NeuroCheck software applied by Confovis for the defect detection uses classifiers that work on the basis of artificial intelligence (AI). The defects are not only assessed and classified via threshold values, but thru evaluations based on neural networks. The classifiers will make it possible to train and categorize newly emerging defects. Open-source tools, such as Tensoflow, Keras or Caffe, are additionally made available to the customer for the training of neural networks. The automated defect detection benefits greatly from the high transmission rate of the hardware control with real-time capability and from the calculation processes that are based on FPGA and graphic cards by way of Cuda.
The requirements concerning the die placement in the fan-out process can be satisfied with the Confovis WAFERinspect measurement system. The same applies to the calculation of coordinates in various layers. Parts of the structure (e.g. bumps) can also be fully measured in 3D during the setup process, or on demand, within two seconds, so as to establish the reason of variations via the wafer or the panel.
Hairline cracks in the multi-layered processes are mainly created by stress in the surfaces. These hairline cracks are low-contrast and mostly so narrow, so that light optical methods reach their limits. The measurement capability of the confocal Confovis sensor with its nanometer precision can detect cracks on the basis of the changing surface topography, especially when they have been created by compressive stress. Processes can be optimized by preventive measurements of topographies at critical places, before stress cracks occur.
Probe marks make special demands on the defect inspection
The combination of the 2D defect inspection and the high-precision 3D-measurements facilitates a quantitative analysis of changes in the needle prints. This will allow an automated combination of defect inspection and the 3D measurement of the critical probe marks, in order to readjust the probing process as early as possible.
A process control instrument for defect inspection & metrology
The WAFERinspect measurement system from Confovis provides the customer with a universal and cost-efficient metrology platform that integrated defect detection and 2D/3D measurements in one system. The data will be transmitted to the host via SECS/GEM. As regards the wafer handling in the case of its automated solutions, Confovis relies on wafer loaders or handling systems from Mechatronic Systemtechnik that have proven their worth all over the world, when it comes to handling complex wafers, such as MEMS wafers, glass wafers and wafers with a high warpage. The measurements have an accuracy of 3 nm and repeat accuracies of 4 nm @ 3 Sigma (example of a 50 nm step certified by the Physikalisch-Technische Bundesanstalt [Physical-Technical Federal Institute]). The precise measurement of angles, distances, radiuses etc. is also possible.