Saturation profile based conformality analysis for atomic layer deposition: aluminum oxide in lateral high-aspect-ratio channels

Atomic layer deposition (ALD) raises global interest through its unparalleled conformality. This work describes new microscopic lateral high-aspect-ratio (LHAR) test structures for conformality analysis of ALD. The LHAR structures are made of silicon and consist of rectangular channels supported by pillars. Extreme aspect ratios even beyond 10 000 : 1 enable investigations where the adsorption front does not penetrate to the end of the channel, thus exposing the saturation profile for detailed analysis. We use the archetypical trimethylaluminum (TMA)-water ALD process to grow alumina as a test vehicle to demonstrate the applicability, repeatability and reproducibility of the saturation profile measurement and to provide a benchmark for future saturation profile studies. Through varying the TMA reaction and purge times, we obtained new information on the surface chemistry characteristics and the chemisorption kinetics of this widely studied ALD process. New saturation profile related classifications and terminology are proposed.


Supporting information to Experimental details
shows the PillarHall-3 chip layout design with rectangular channels on a silicon substrate. A top polysilicon membrane was supported by silicon pillars (Figure S1 b and c). The main test area consists of nine different LHAR channels (and six of VHAR type). The LHAR channels in the main test area are mirrored [ Figure S1a (2)], while the largest LHAR channel with the lateral length L of 5000 µm is one-sided [ Figure S1a  Each LHAR channel has a different opening width W (Table S1) before the channel entry (i) to recognize the individual channel e.g. in cross sectional scanning electron microscopy Electronic Supplementary Material (ESI) for Physical Chemistry Chemical Physics. This journal is © the Owner Societies 2020 and (ii) to use opening widths as a length scale indicator in top-view analysis. The opening is sufficiently wide (W >> H) to not limit the film growth.
The PillarHall-3 conformality test chips contain LHAR channels with four different pillar arrays: layouts v1a, v1b, v2a, and v2b, as presented in Figure S3. Pillars are located in hexagonal symmetry with different pillar inter-distance a, pillar diameter d, and distance between rows of pillars l depending on the layout (Table S2). Layout v1b is the standard design, with a = 49 µm, d = 4 µm, and l = 42.4 µm. The lateral distance l is calculated as √3 2 .The pillars at the channel entry are elongated in all designs. Figure S1 (a) Top view layout of PillarHall-3 having microscopic LHAR channels: (1) chip number, (2) LHAR channels in the main test area, (3) the largest LHAR test feature, (4) pillar layout indicator, (5) channel height indicator (silicon oxide between silicon and polysilicon), (6) additional LHAR channels (located symmetrically on top and bottom of the chip), (7) polysilicon tensile stress test circles, (8) VHAR channels, and (9) cleaving notch. Panels (b)-(c) are schematic side views of rectangular LHAR channels with x and y directions, not in drawn scale. Figure S2. Optical microscopic image of PillarHall-3 conformality test chip with its top membrane on (sample 13, Table 1). A distance indicator scale is marked on top of the membrane for optical analysis.   Table S2.   Figure S4. The as-measured saturation profile of samples 1 to 3 with different pillar layout designs were measured with reflectometry after removing the membrane (Table 1 Series A).  Table 1) measured by SEM-EDS. ALD Al2O3 was coated on the chip using the 1000 cycles with the ALD process sequences of TMA pulse-purge-water pulse-purge of (0.1-4.0-0.1-4.0) s, and the top membrane was removed by adhesive tape before the measurement.  Table 1.   Table 1, Series D within PillarHall-3 LHAR channel with design channel height of 500 nm).

ALD film thickness vs. cycles
The relationship between the Al2O3 ALD film thickness and number of ALD cycles was analyzed. In Figure S9, linear film growth against the number of cycles was observed from the films made in 250, 500, and 1000 cycles over the design channel height of 500 nm. The GPC in the channel decreased slightly with increasing number of ALD cycles ( Figure  S9 b). The higher GPC in the beginning could have been caused by the rough surface of the etched channel. The intercepts of y-axis in the film growth plot (Figure S9 a) indicate the existence of native silicon oxide layer and the surface roughness in the channel (Region IIb). To exclude the effect of native oxide layer and rough surface, thickness values were corrected by subtracting those intercept values (Figure S9 c), and GPC was recalculated using the corrected thickness ( Figure S9 d).

Fabrication issues related to LHAR channels in different heights
The LHAR channel fabrication targeted channels with the heights of 100, 500, and 2000 nm. For each wafer, some fabrication-related parameter was varied, and on a wafer, up to four layouts were experimented with. Our work has concentrated on reporting on the channels with a targeted 500 nm channel height, where the fabrication worked out best. The channels with 2000 nm had fabrication (dry etch) related issues, leading to that some of the features were accidentally rounded and some channels were lost or narrowed. The channels with targeted 100 nm channel height, in turn, suffered from large relative error in the channel height because of membrane hanging between pillars and already small channel. On the basis of the analysis presented, the LHAR channel with Layout v1b will give the most reliable and comparable results.

The effect of pillars on saturation profile
The pillars in PillarHall-3 conformality test chips keep the top membrane from touching the bottom of the structure. They also act as distance indicators in top-view microscopy. Yet, the pillars alter the surface area to be coated and increase diffusion resistance. Here, we mathematically assess the effect of the pillars on diffusion.
F (mol/m 2 s) is the flux of material in direction x, and it is calculated as (4) If the diameter of the pillar is d (m) and the distance between them is a (m), the diffusion resistance of one row of pillars with the number of n is The diffusion resistance of the space between pillar rows is If there would be no pillars the total diffusion resistance would be open = . (7) The ratio of diffusion resistances with pillars to without them is The diffusion resistance ratio is plotted in Figure S11 as a function of the dimensionless pillar distance a/d. In the most dense pillar configuration the effect on the diffusion resistance is slightly above 2% which might be noticeable in the most precise penetration depth measurement. Therefore, the pillar distance should be at least 11 times larger than the pillar diameter so that the effect on diffusion resistance is below 1%. To estimate how the surface area to be coated changes, we assume that the pillars have a cylindrical shape having a diameter d. The surface growth area decreases by the top and bottom parts of pillars 2 ( 2 ) 2 , whereas it increases by the surface area of the side of each pillar 1 , which is 1 = .
(9) The channel height H also influences the surface growth area, since the pillar area A1 depends on the height. The surface area of the channel roof and ceiling in a triangle formed by three pillars is (10) where a is distance between pillars. The total surface growth area with pillars is where n is the number of whole pillars in the triangle. Because the internal angles in an equilateral triangle are 60° each of the three pillars have 1/6 of their volume in the triangle. The total number of whole pillars thus is n= 0.5. The ratio of film growth area with pillars to no-pillars case is With pillars having the diameter of 4 μm in H = 0.5 μm channel the effect on film growth area is below 1% if the pillar inter-distance a is larger than 25 μm (Figure S12 a). Therefore, for the typical pillar layout v1b the effect on film growth area is negligible as its pillar inter-distance is large enough (49 µm). Increase of the diffusion resistance decreases the film penetration depth, while the decrease of the film growth area increases the penetration depth. The apparent diffusion constant with pillars is D = D0/(rRrA) where D0 is the diffusion coefficient without pillars. Thus the effects tend to cancel each other and the net effect on the apparent diffusion coefficient D is small (Figure S12 b). Figure S11. The relative increase of diffusion resistance when the distance between the pillars a decreases. For the default design, the pillar diameter d is 4 µm. Figure S12. (a) The reduction in the growth area when the distance a between the pillars changes. (b) The change in the relative apparent diffusion coefficient D when the distance a between the pillars changes. D0 is the diffusion coefficient without pillars.

The comparison of PillarHall generations
The PillarHall-3 design reported in this work differs from the PillarHall-1 design, used in earlier works. [1][2][3] The most important differences are described in Table S6. In addition, a fundamental difference is that in PillarHall-3, the entrance to the channel was at the same level as the channel itself, whereas in PillarHall-1 contained a recess in front of the channel caused by etching of silicon. With PillarHall-3, one could thus determine the film thickness in front of the channel for reference purposes (Region I), whereas for PillarHall-1, such reference measurement could not be made. We observed a nanostep near the start of the channel of PillarHall-3 ( Figure S15 and Figure 3b). While the origin of the step is not fully understood, it could have been caused by the rough channel surface resulting from plug-up process. 4 This roughness should have been similar in PillarHall-3 and PillarHall-1, as the same plug-up process is used in both fabrication processes.
To qualitatively compare the results obtained from PillarHall-3 with those of PillarHall-1, scaled saturation profiles for Al2O3 films made with the TMA-water process with the same number of cycles at 300 °C in the two test structures are shown in Figure S15. The Al2O3 film penetrated slightly deeper in PillarHall-3, likely because of lower diffusion resistance due to less dense pillars. The saturation profile was smoother with less noise, most likely because the pillar remnants did not disturb the measurement in PillarHall-3 (spot size ~5 micrometer fits in the space between pillars). A small difference in slope at the saturation front is observed (-0.00099 nm and -0.00136 nm for sample 11 and Gao et al., 1 respectively). Figure S15. Scaled saturation profile of Al2O3 ALD thin film obtained from our experiment compared to Gao et al. 1 Our study used PillarHall-3 while the reference used PillarHall-1. Both prototypes had design channel heights of 500 nm, and the films were made in 1000 cycles. Figure S16. Repeated saturation profile measurements for Sample 8 (Table 1) after removing the top membrane: (a) line scans measured by reflectometer with an occasional spike in Region IV, (b) the corresponding root mean square error (RMSE) fitting residual of the reflectometer measurement, and (c) and (d) microscope images related to the measurement spots marked with x and ☐ in panel (a) for channel W100L5 and W80L0.5 (1st scan), respectively. LHAR design channel height 500 nm and pillar layout design v1b. Figure S17. Repeated saturation profile measurements after removing the top membrane: (a) line scans measured by reflectometer with occasional spikes in Regions II to IV, (b) the corresponding RMSE fitting residual of the reflectometer measurement, and (c) microscope images related to the measurement spots marked with x in panel (a), channel W80L0.5 (1st and 3rd scan). For this sample (V0005, not in Table 1), as a sacrificial layer, low-pressure chemical vapor deposition process based on tetraethyl orthosilicate (LPCVD TEOS) was used instead of thermal silicon oxide layer. Al2O3 was coated on LHAR channels with a design channel height of 500 nm and pillar layout design of v1a by ALD sequence of TMA-water-purge-water of (0.1-4.0-0.1-4.0) s at 300 ˚C. 500 cycles were used.