中文 English

Accelerating Dry Etch Processes During Feature Dependent Etch

Using visibility etch modeling to compensate for challenging aspects of etch recipe development.

popularity

In dry etching, the trajectory of accelerated ions is non-uniform and non-vertical, due to collisions with gas molecules and other random thermal effects (figure 1). This has an impact on etch results, since the etch rate at any point on the wafer will vary depending on the solid angle visible to the bulk chamber and the ion flux for that angular range. These non-uniform and feature dependent etch rates complicate the development of etch recipes during semiconductor process design. In this article, we will describe how to compensate for this challenging aspect of dry etching by using visibility etch modeling in SEMulator3D.


Fig. 1a: 2D representation of neutral gases randomly traveling inside a chamber. The angular travel direction of gases is equally populated in all directions depicted in the graph (Fig 1b). In Fig 1b, we show positively charged ions and a negatively biased wafer. Ions will be accelerated downwards due to the electric field; however, perfect vertical trajectories will not be achieved due to random thermal effects and collisions with other ions or gas molecules. Angular velocity distribution can be approximated as a Gaussian function (Fig. 1b).

Angular dependent etching

The simplest method of determining a material etch rate (ER) is to measure the material thickness of a blanket wafer before and after actual etching. Using flat wafers during the etch process ensures that all locations within a local region have the same exposed opening angle and ion flux, which should result in a uniform etch rate (figure 2a) that can be measured. Determining this etch rate during feature dependent etching, such as when trenches and hardmasks are etched, is not possible due to the different etch angles and changing ion flux inside these features. SEMulator3D is able to model this type of etching using its “multi etch” function. The software measures the exposed solid angle for any given point and calculates a normalized etch amount that is proportional to the ion flux distribution over the range of the exposed solid angle (Figure 2c). Ion flux distributions over the incident angle are assumed to be Gaussian with a characterized standard deviation.


Fig. 2a: On a flat wafer surface, every location (A,B,C,D) is fully exposed to the chamber (opening angle equal to 180) and receives the full flux of ions in all directions. Fig. 2b: In the bottom of pits and trenches (E, F), the angular range within the line of sight of the chamber is reduced. The etch rate can be scaled as the integral of the distribution within the angular range (shaded area under the curve between the vertical lines).

Profiling an etch recipe

The ion angular spread for a given etch chamber setting (RF power and pressure settings) can be empirically determined using SEM images of a time lapsed etch sample. A “virtual” structure inside a virtual chamber can be subsequently modeled in SEMulator3D. A virtual DOE can be run in this model, by varying the angular spreads until the virtual etch modeling result matches the actual SEM image profiles. In figure 3, a hypothetical time lapse SEM of an etched sample is compared to several thickness lapse simulation models in SEMulator3D, showing etch shapes and depths at various angular spread values. The thickness setting in SEMulator3D describes the maximum theoretical material removal in a region that has full visibility to the bulk chamber. This setting would be proportional to the maximum ion flux on the sample in an actual etch chamber. The simulation settings that best match the actual etch recipe will have a simulation profile (3D simulation image) that matches the SEM image for every proportional thickness and time step increment. Developing a simulation recipe that matches a corresponding actual etch recipe is extremely valuable. It can be used to predict etch time evolution within a sample and make it possible to use the virtual etch model in other applications and structures during process exploration.


Fig. 3: Comparison of a simulation DOE vs time lapse SEM. The simulation DOE used constant etch amounts with varying angular spreads (standard deviation of the Gaussian distribution). The simulation was performed and the incremental material removal step lapse is displayed. The right histogram illustrates the angular distribution correlation with the numerical setting within the software (not to scale). The actual angular spread of an etch process is determined by finding the simulation DOE result that best matches the etch profile.

Using a profiled recipe to optimize an SADP sample

As an example of visibility etch in SEMulator3D, we will use a profiled SiO2 and SiN etch process model to determine the optimal ALD thickness needed to ensure hole CD uniformity for SADP pillars (figure 4). The sample consists of a 50 nm SiN layer and a 100 nm tall carbon mandrel. The mandrels are 20 nm in diameter and have a horizontal pitch of 80 nm. The final goal is to create a 40 nm pitch hole array using SADP. The profiled SiN/SiO2 etch has an angular spread of 0.08 and a selectivity of 0.3 for all foreign materials. The asymmetric shape of holes formed by ALD takes on a diamond shape with a rounded opening, in contrast to the cylindrical hole formed at the mandrel. Since the size of this diamond shaped hole can be adjusted by ALD, we need to determine the critical ALD thickness where the total ions entering this diamond hole region for the duration of the etch equals that of the ions entering the cylinder area. This should result in an equal etch depth and shape.


Fig. 4: Diamond SADP on hole arrays. Mandrels have a diameter of 20 nm and are separated by 80 nm in the horizontal direction. The holes formed by the expanding outward deposition form a diamond shape and have a rounded opening. Using a profiled SiO2 etch, the shape of the etched hole can be explored for various ALD thicknesses.

An ALD thickness DOE is performed in SEMulator3D to determine this optimal thickness. The results of this simulation are shown in figure 5, with the top down shape of the etch and the bottom cross section visible. With increasing ALD thickness, the hole shape at the SiN/substrate interface changes from square to circular and gradually decreases in size. At sufficient ALD thickness, the cusps of the diamond shaped holes have limited visibility which results in low etch rates and the etch retaining a circular shape. An ALD thickness of 23.5 nm exhibits the most uniform hole shape for this particular profiled SiO2 and SiN etch process.


Fig. 5: ALD thickness dependence and layer etch. Using profiled anisotropic etching of the SiO2 (blue) and SiN (green), the resulting hole shape can be determined using varying ALD thicknesses. The best shape is found at a 23.5 nm ALD value, using a SEMulator3D visibility etch model that was previously validated again actual etch results.

Conclusion

The visibility dependent etch feature in SEMulator3D offers a method to model etch rates similar to those in an actual etch chamber. SEMulator3D visibility etch settings, such as angular spread and selectivity, can be compared with time lapse SEM images to validate the process model. The process model can then be used to explore the effects of etch recipe changes on different structures and varied etch times, without the time and expense of actual wafer fabrication and testing. To learn more about how predictive structural modeling can accelerate the development of complex process flows, click here.



Leave a Reply


(Note: This name will be displayed publicly)