Text by MVTEC Software
Global semiconductor manufacturing capacity is expanding. However, the production process in this industry is complex, with more than 1000 different steps, so the development of production capacity is equally complex and sophisticated. As a result, the semiconductor industry needs technologies that can be implemented and adapted quickly while improving productivity. Machine vision is one of the key technologies.
The main advantage of machine vision technology is that it can automate, highly accurately, perform many of the necessary inspection and calibration processes in semiconductor manufacturing. MVTEC Software offers machine vision software products HALCON and MERLIC from Munich, Germany, which increase the profitability of all process steps in semiconductor manufacturing.
Figure 1: Machine vision supports semiconductor production with the highest precision.
Powerful machine vision technology for semiconductor manufacturing
In almost all semiconductor production scenarios, there is at least one step that requires a product to be inspected for functional or cosmetic defects. In a highly automated production environment, there are many advantages to using machine vision inspections over manual inspections: machine vision inspections are faster, the results are objective and repeatable, and the quality of the inspection is not compromised by fatigue or tedious tasks. Deep learning techniques such as outlier detection are also suitable for this purpose. For example, it enables automatic surface inspection, detecting and segmenting defects.
Quality checks must check for dimensional accuracy and defects at the same time. Machine vision can measure edges with sub-pixel accuracy in milliseconds along line segments or arcs. In combination with 2D measurement technology, objects of specific geometries can also be inspected. In addition, there are some 3D measurement methods, including reconstructing 3D surface coordinates using parallax images, distance images, or through various methods.
It's also important to find a partner. Most importantly, shape-based matching with sub-pixel precision. The technology enables accurate and robust object discovery in real time. The feature works even if they are rotated, scaled, perspective distorted, partially distorted, partially covered, or located outside the image.
Machine vision technology in different process steps
As mentioned above, there are many process steps in semiconductor production. Here are some examples of processes and how machine vision can provide support.
Dicing is a type of precision machining that uses mechanical dicing or a laser to split a wafer into separate chips. Machine vision enables operators to set up wafers in the machine. This must be done again for each wafer type, as the size of the wafer must be known in order to be set. For example, you can use a fast Fourier transform to determine dimensions. This significantly simplifies the setup and reduces the risk of operational errors. Subsequently, when the mold is cut, it may occur that the edges are cracked. Here, industrial image processing can support anomaly detection using deep learning methods. With just a few "good images", powerful machine vision software can be trained to reliably detect any damage to the mold, i.e. anomalies.
Packaging: The final step in manufacturing the finished chip
Figure 2: During the packaging process, the chip is placed in a plastic housing and the contact points are set using an adhesive process.
In addition to quality checks, another application where machine vision can effectively support is finding and aligning wafers and chips. In this application, shape-based sub-pixel matching comes into play. This technology enables accurate and reliable real-time detection of objects. Detection works even when the object is rotated, scaled, distorted in perspective, locally distorted, partially occluded, or out of the image.
An example of the added value of this technology is the packaging production step, where the chip is encapsulated in a plastic housing. First, a single die is inserted into the housing. They are then connected to the relevant contacts of the respective package in a few steps. In order for the subsequent processing to proceed smoothly, the precise positioning of the chip within the package is a critical step. The shape-based matching method is a good fit for this. In practice, mold models are trained by means of CAD data, etc. By matching the training data with the original image, the mold can be positioned and its position in the plastic housing can be calculated to facilitate subsequent processing.
The above production steps illustrate the added value of machine vision. In addition, as the "eye of semiconductor production", machine vision technology runs through the entire process chain. Especially in the areas of quality assurance and packaging, high-performance machine vision software can make an important contribution to efficient production.
Figure 3: High-precision measurement of contacts (pads) on a chip using subpixel edge detection.
end