We use cookies to improve your experience. By your continued use of this site you accept such use. To change your settings please see our Privacy Policy.

Non-destructive Analysis (NDA)

Circuit Edit (CKT)

Materials Analysis (MA)

Application Forms


MA-tek FTP

CSR Report



In IC fabrication, the performance of metrology equipment directly influences the ability of process tuning and promotion of productivity. Foundries and vendors have to make sure that all measurements are within tolerances and ISO and quality system identification. This has become more and more challenging with the decreasing of device dimension and tolerance.


In order to meet this rigorous metrological requirement, the semiconductor industry is continuously seeking for any possible solutions. Until now, many metrology tools have been developed to measure important parameters, such as critical dimensions (CD), thin film thickness, morphology, dopant concentration, and defect analysis.


For devices in nanometer dimension, it will easily cause mistakes if the measurement and analysis are done in manual or semi-automatic way. It will be much effective to use automatic measurement. Also, the data will be accurate and have meaningful statistics information.




Thickness measurement for multi-layers

The lighting performance and epitaxy strain of LED strongly falls on the thin film thickness and spacing of the epitaxial MQW and SL. Therefore, it is highly demanded to have precise and fast auto-metrology tools for LED foundries.


(a) TEM image of LED MQW

(b) Boundaries for each MQW after image processing.

The figure exhibits how the auto-metrology works. First, boundaries are identified for each MQW layer by comparing contrast/brightness of every pixel in TEM images with an image processing software. Then, parameters of MQW can be calculated by statistics, as shown in table 1.


With this method, all data from the whole image can be converted to meaningful statistics numbers, not just a result from a single point. Therefore, data and thickness variation averaged from a bigger area provides more scientific feedback for development of epitaxy growth technology.


An example of calculations is shown in table 1. With the image auto-metrology, the number of layers, average or root mean square thickness for each layer, thickness range for each layer, and standard deviation of thickness variation can be calculated. All engineers can read this data and correctly make the right judgement for the next R&D step.


Statistics data of LED MQW.

(a) Number of layers

(b) Mean thickness for each layer

(c) Thickness range for each layer

(d) RMS for each layer

(e) Thickness standard deviation of each layer.




Nano Particles Size Measurement and Size Distribution

The general definition of nano-particles is that the particle's dimensions are smaller than 100 nm. When the dimension of materials is smaller than 100 nm, their physical and chemical properties will be obviously different than those of bulk materials. Therefore, industries, academies, and governments all over the world have spent enormous resources on new nano-particle materials development. Because the properties of nano-particles are highly related to their shape and dimension, it is needed to have an auto-metrology system to give precise and fast shape and dimension measurements for nano-particles.




(a) TEM image of nano-particles

(b) Nano-particle image after image identification

(c) Data after image processing

(d) Statists distribution of area of nano-particles.

(a) is a TEM image of nano-particles. With the help of image identification system, as shown in Fig. (b), the TEM image can be converted into shape and dimension data of nano-particles

(c)The shape and dimension distribution of nano-particles based on statistics analysis is shown in Fig. (d).




Grain Size Measurement and Size Distribution

For the measurement of grain size of metal materials, in the past, simple definition and method can be found in Metal Handbook. To have precise statistical calculations for enormous number of grains and clear definition and qualitative description of shape, image processing method is needed.


Here, we take ITO grain as an example. Clear boundary contrast can be observed to identify the shape and size of grains. Figure 14 exhibits the results analyzed by software provided by the equipment vendor.


(a) TEM image of ITO grains

(b) Boundaries after image processing

(c) Size distribution of grains.



For some materials, there are twin boundary and stacking faults in a grain, which will influence the calculation results. We take Cu metal grains as an example. Image processing can keep twin boundaries and stacking faults or ignore such boundaries in order to make the results more close to the real situation, as shown in the figure below.


(a) SIM image of Cu grains

(b) Locations of Cu grain boundaries after image processing. Twin boundaries have been removed

(c) Size distribution of Cu grains.



About the grain size measurement for poly-Si, calculations on Al grains can be used for high dopant and long annealing poly-Si grains. For low dopant and non-annealing grains, however, the measurement cannot be done due to no clear boundary definition from the TEM image, as shown in the following figure. New metrology needed to be developed.


(a) TEM iamge of poly-Si

(b) Locations of poly-Si boundaries after image processing.




Auto-profiling of FinFET

When the technology node is smaller than 20 nm, traditional 2D structure of transistors has changed to the 3D FinFET structure, as shown in the figure (a). For such 3D FinFETs, precise measurements are important not only in the thickness of each layer but also in shape control of Fin. The Fin shape is one of important factors that influences the performance of transistors.


Traditionally, measurements are manually carried out based on TEM images. It may cause many mistakes. In order to lower such mistakes, we develop an image auto-metrology technology. This technology includes three steps:


  1. Detecting the values of brightness and contrast in a TEM image;Profiling between layers can be identified by differentiation.
  2.  Converting the FinFET profiling to data, as shown in Figure. (c).
  3. Calculating the data to statistical information, as shown in Figure (d)





Table 1(a) to (e) exhibits the results after calculation. Alternatively, analysis results can be also presented as shown in Fig. (d) based on different purposes and R&D goals. Such data presenting is much better than just calculated numbers.



(a) TEM image of FinFET

(b) Profiling for each layer in FinFET

(c) Concerting profiling to data

(d) Shape statists of FinFET