Skip to main content
Personal computers, mobile phones, digital TVs and a variety of other electronic products are rapidly evolving, making our lifestyles more convenient and affluent. What makes this evolution possible is "semiconductors," a core component of electronic products, which are always making progress in the microscopic world on a "nanometer" (nm) scale.
Production of semiconductors is supported by semiconductor inspection equipment. iPQ (Inspection & Process Qualifier), a semiconductor inspection application Hitachi has developed jointly with Hitachi High-Technologies Corporation, also continues to evolve progressively along with semiconductors.
HARADAThat is correct. Semiconductors are evolving in a way where more circuits are integrated in those tiny chips. So the production process of semiconductors is increasingly refined in terms of microlithography. Presently, a defect that may cause a fatal failure of a semiconductor is as small as ten or so nanometers (nm). When inspecting semiconductors, we must detect defects of such a size.
Conventionally, optical microscopes have been used for inspecting semiconductors to detect defects, as they can very quickly cover a wide scope. However, optical wavelengths are hundreds of nanometers, while the defects we want to find are ten or so nm. Therefore, although defects can be located by using optical microscopes, we cannot tell what type of defect they are. That is why it has become necessary to conduct inspections with higher sensitivity by using scanning electron microscopes (SEMs).
HARADASemiconductor production processes are ever progressing. To support this progress, we are working on research and development of inspection systems that use SEMs.
In the semiconductor production lines, defects are located using optical microscopes and, based on the location information, images are taken using SEMs to closely examine the defects. Defect Review SEM equipment automatizes the collection of images for examination. The equipment not only automatically collects images of defects but also automatically classifies the collected images in accordance with the type of defect.
iPQ (Inspection & Process Qualifier) we are developing is an inspection application that utilizes the Defect Review SEM. This application automatically takes SEM images of pre-designated locations (fixed points) and quantifies the tendency of defect generation as well as the process quality. Although iPQ is slower and covers a smaller scope in defect detection compared to optical inspection equipment, it can conduct super high-sensitive inspections.
Figure 1: Flow from inspection to process improvement using iPQ
HARADAIn producing semiconductors, circuit patterns are "burnt" onto a silicon substrate (wafer) having a diameter of about 300 millimeters. This wafer is diced into pieces, and you get many chips like those installed in mobile phones. These chips are shipped after undergoing electrical inspection.
However, if problems are found in the inspection before shipping, that is too late in terms of semiconductor production. When burning circuit patterns onto semiconductors, membranes are formed on the wafers, circuit patterns are drawn on the membranes (exposure), then the circuit patterns are etched, and these procedures are repeated many times. Circuits are formed on many layers, so hundreds of processes are required to produce a single wafer. If any problem occurs somewhere in these processes, a huge volume of defective products will result.
Therefore, it is important that wafers are inspected during the production process and discover any problem and solve it as early as possible.
HARADAiPQ stands for Inspection & Process Qualifier. As this name indicates, the application performs inspections (defect inspection) and process qualification (measurements of circuit patterns).
In defect inspection, iPQ checks the number and size of defects, the brightness and other factors of defects, and quantifies them. In process qualification, iPQ checks the area and brightness of circuit patterns, surface roughness (unevenness) and other factors and quantifies them. Users utilize the quantified values to find out where the problem lies, and feed it back to exposure, etching and other processes.
HARADAIn principle, defects are detected by comparing images of wafers being inspected to images of non-defective products. Non-defective images are images of wafers without any defects. iPQ compares the images for inspection with the non-defective images to check how different they are, and detects defects.
Figure 2: Defect detection through comparison with non-defective images
HARADAHowever, here is the problem of how to obtain non-defective images. For example, after taking hundreds of images of wafer surfaces, it is rather difficult to find images without any defects from among them. Or there may not be any non-defective images among them, in the first place.
To solve this problem, we have developed a function by which sections without defects are selected from the images taken and synthesized into a single complete non-defective image. This function has made it possible to depict defects in various scenes, such as in the "process development stage" for which no non-defective images exist.
HARADAIt's not that simple, as there is the issue of "fabrication tolerance." Even if there are differences found when comparing images, they are not deemed to be defects if they are within the scope of fabrication tolerance. This is the difficult part in the detection of defects.
For example, there are cases in which the size of circuit patterns change. Naturally, the difference in size is detected. However, semiconductor manufacturers sometimes tell us that such differences are not product defects and thus should not be subject to detection. Accordingly, we have created a function for extracting defects through a filtering process, so that the equipment only detects the defects desired or that are the focus at that particular time.
iPQ quantifies the appearance of the detected differences. The quantified values are called "feature values" of which there are tens of types. By using filtering function, iPQ detects defects that have consistent features based on the feature value.
However, another difficulty is setting the features of defects as the conditions for filtering. Defect features vary quite a bit according to the process, and the fabrication tolerances also vary. So the feature by which we can distinguish between a defect and a fabrication tolerance is different every time. Thus, we have also created a function in which iPQ itself automatically tunes the filtering conditions based on the several locations of defects designated by users.
Figure 3: Filtering function based on feature value
HARADAProcess Qualifier provides a function to automatically recognize the circuit pattern areas and measure the features of their shapes as well as their superposition accuracy. The measurement of the superposition accuracy is called "overlay measurement."
Circuit patterns of semiconductors are formed in a stack of layers. It is important that the circuit patterns are properly formed in respective layers horizontally, but the superposition of the layers is also important. Any vertical dislocation between layers may impact the device's characteristics. So we need to measure the dislocation and make adjustments to keep it within a certain range.
In conventional overlay measurement, patterns for measurement purposes were formed in advance on a semiconductor wafer in each layer through separate processing. The dislocation of the patterns is then measured by using optical microscopes. However, this method produces measurement errors that are too large for practical application in the increasingly refined semiconductor processing. To overcome this problem, studies are being made on a method in which the patterns for measurement purposes are smaller and are embedded Inside of the chip surface and then measured using SEM.
In contrast, iPQ does not require patterns for measurement purposes. Instead, it achieves an overlay measurement method that is conducted directly on circuit patterns.
Figure 4: Patterns for overlay measurement
HARADAExactly. Moreover, memory device manufacturers need to conduct overlay measurement directly from actually formed circuit patterns. They want to do so because it is difficult to embed the patterns for measurement purposes in some types of semiconductors. Such patterns can be embedded relatively easily with "logic" semiconductors such as CPUs (central processing units) but are harder for "memory" semiconductors, as it is essential that memory chips form increasingly higher-density circuits.
Furthermore, microlithography technologies advance faster for memory chips compared to with other semiconductors and thus require more highly accurate measurement. That is another reason for the need of direct overlay measurement.
HARADAIn terms of mechanism, iPQ compares non-defective images and defective product images, as in the defect detection stage. However, in the case of overlay measurement, the layers shown in a single image are divided into upper layers and lower layers through image processing and are recognized as such, and then their dislocation is measured. What is technologically difficult to do is "to divide into upper and lower layers and recognize them as such." Since how the circuits are formed is not known, a presumption is made that "these must be upper layers" and they are then divided.
As of now, I believe that this particular flow in which "circuit patterns are divided and recognized as such, and overlay measurement is conducted by comparing their locations with non-defective images" is a unique technology of Hitachi.
Figure 5: Mechanism of overlay measurement by iPQ
HARADASemiconductor manufacturers of memory chips are the primary users of iPQ. I believe that, since iPQ provides added value, it contributes to the sales of Defect Review SEMs by Hitachi High-Technologies Corporation.
Originally, iPQ was created to enhance the product competitiveness of our Defect Review SEMs. Semiconductor manufactures want to conduct as many inspections as possible with a single apparatus in order to reduce costs. To meet this need, we have been working to add various inspection applications to our Defect Review SEMs so that other inspections can be conducted in addition to automated detection of defects. iPQ is one of these inspection applications.
At the early stage of development, iPQ performed only the simple function of comparing non-defective images with images for inspection and indicating their differences as numerical data. However, we received various requirements of semiconductor manufacturers, who told us that they "want a function that can accurately count the number of defects" or "want to have quantified data for various, more subdivided items." In response to such, it was decided that Hitachi High-Technologies Corporation and Yokohama Research Laboratory of Hitachi, Ltd. would jointly develop iPQ.
HARADAYokohama Research Laboratory and Hitachi High-Technologies Corporation had been conducting joint R&D on the function for detection of defects for defect review. The reasoning behind the decision to make the development of iPQ a joint development was so that the know-how and expertise of the defect detection technologies could be fully utilized.
I have been involved in defect detection research since I joined Hitachi. When I heard of the development of iPQ, I thought it would be quite interesting, and I became a member of the development team.
In the early stages of the development various needs became evident in a sporadic manner, so we started by analyzing requirements and studying software structures. The initial work included system engineer type tasks of image processing and was quite hard.
Basically, the roles in the joint development are that Yokohama Research Laboratory engages in the R&D of the image processing algorithm and that Hitachi High-Technologies Corporation works on the development of the user interface and communications with other systems.
Figure 6: Sample iPQ operation screens
HARADAFor defect inspection, we must enhance the technology to differentiate defects from fabrication tolerance. Distinguishing defects from fabrication tolerance is becoming extremely difficult as a result of the further refinement of semiconductor processing. We need to make it possible to only detect actual defects. Moreover, simply being able to detect them is not enough. Another challenge is to make the equipment more user-friendly.
For process qualification, overlay measurement must be more precise and faster. There is a great need for overlay measurement to be performed repeatability, which means that the same value is produced every time a measurement is conducted on the same section, and to conduct such measurement on a scale of 1 nm or less. Currently, this is the most difficult part of the technologies related to overlay measurement.
In addition, since the production processes of semiconductors are continuously changing and evolving, it is also important to meet our customers at their manufacturing sites, absorb their needs and respond to them as quickly as possible.
For example, process layers had traditionally been flat, but now they are starting to become three-dimensional. For three-dimensional layers, we must observe the bottom of deep trenches. However, the bottom cannot always be seen with images produced by SEMs, as electrons may not come up from the bottom of the trenches. Therefore, we need to develop technologies that can distinguish defects from fabrication tolerance using limited signals.
HARADAIt is particularly important to respond to a wide range of processing technologies. We cannot observe or look at all semiconductor processes in the world. Therefore, it is important that we create functions that can be applied to as wide a range of processing methods as possible, based on the data we have now.
When aiming to improve the performance of applications, there is a tendency to focus on specific processes. But it is important to think how to improve the overall performance, and this is the most difficult and requires the most brain-power. Unless we relentlessly consider the "overall" rather than just a "specific" part, the application will be useless on the manufacturing floor.
Indeed, it is difficult to create the ultimate application that can inspect anything because the objects of inspection are wide-ranging and continuously evolving. Nevertheless, I desire to strive forward to get as close as possible to this ultimate goal.
(Publication: December 18, 2013)