Datasets:
Formats:
imagefolder
Languages:
English
Size:
1K - 10K
ArXiv:
Tags:
scientific-images
materials-science
scientific-figures
atomic-layer-deposition
atomic-layer-etching
experimental-usecase
License:
File size: 15,742 Bytes
4b6a819 | 1 2 3 4 | {"file_name":"images/test_atomic-layer-etching_simulation-usecase_11_f15f209f2d385c0b325b181192c48ae716e73fcf75a8c14d4bd8538703ffdfce.jpg","caption":"(a) Data set for Step A generated from the microscopic model.","id":"test/atomic-layer-etching/simulation-usecase/11/f15f209f2d385c0b325b181192c48ae716e73fcf75a8c14d4bd8538703ffdfce","sample_id":"atomic-layer-etching/simulation-usecase/11/f15f209f2d385c0b325b181192c48ae716e73fcf75a8c14d4bd8538703ffdfce","subset":"3d-scatter-plot","split":"test","classification":[{"panel_id":"a","label":"3d scatter plot"}],"summarization":"[{\"panel_id\":\"a\",\"text\":\"3D scatter plot of kMC simulation results for Step A (HF fluorination) showing half-cycle time as a function of temperature (533-598 K) and pressure (10-100 Pa). The data shows strong temperature dependence but near-complete pressure independence, with half-cycle times ranging from approximately 1 s at high temperatures to 12 s at low temperatures. The horizontal bands at each temperature confirm that HF adsorption kinetics are not significantly affected by pressure.\"}]","data_extraction":"[{\"panel_id\":\"a\",\"text\":\"| Temperature (K) | Pressure (Pa) | Time (s) |\\n|----------------|---------------|----------|\\n| 520 | 20 | 19 |\\n| 540 | 20 | 18 |\\n| 560 | 20 | 17 |\\n| 580 | 20 | 16 |\\n| 600 | 20 | 15 |\\n| 520 | 40 | 14 |\\n| 540 | 40 | 13 |\\n| 560 | 40 | 12 |\\n| 580 | 40 | 11 |\\n| 600 | 40 | 10 |\\n| 520 | 60 | 9 |\\n| 540 | 60 | 8 |\\n| 560 | 60 | 7 |\\n| 580 | 60 | 6 |\\n| 600 | 60 | 5 |\\n| 520 | 80 | 4 |\\n| 540 | 80 | 3 |\\n| 560 | 80 | 2 |\\n| 580 | 80 | 1 |\\n| 600 | 80 | 0 |\"}]","vqa":"[{\"panel_id\":\"a\",\"items\":[{\"question_type\":\"Comparative/Trend\",\"questions\":\"\",\"answer_type\":\"Paragraph\",\"answer\":\"Step A shows virtually no pressure dependence, as evidenced by the horizontal bands of data points at each temperature. In contrast, Step B exhibits clear pressure dependence with half-cycle times decreasing at higher pressures. This difference arises from the reaction mechanisms: HF dissociatively adsorbs readily onto the Al2O3 surface regardless of pressure, while TMA adsorption in Step B is more kinetically limited and benefits from higher partial pressures to drive secondary adsorption events.\"},{\"question_type\":\"Structure-Property\",\"questions\":\"\",\"answer_type\":\"Factoid\",\"answer\":\"Half-cycle times range from approximately 1 s at high temperatures (~598 K) to about 12 s at low temperatures (~533 K).\"},{\"question_type\":\"Process-Oriented\",\"questions\":\"\",\"answer_type\":\"Factoid\",\"answer\":\"Temperature governs the rate constants through the Arrhenius equation, so higher temperatures accelerate surface reactions like hydrogen diffusion and water desorption that determine the fluorination half-cycle completion time.\"},{\"question_type\":\"Application/Performance\",\"questions\":\"\",\"answer_type\":\"Factoid\",\"answer\":\"Process engineers have flexibility in selecting HF pressure without significantly affecting Step A half-cycle time, allowing pressure optimization based on other factors such as precursor utilization efficiency or reactor design constraints.\"}]}]","bbox":[{"panel_id":"a","x":5,"y":8,"width":499,"height":433}],"source":"icdar2026-competition-data/test/gold_standard_test_set/atomic-layer-etching/simulation-usecase/11/images/f15f209f2d385c0b325b181192c48ae716e73fcf75a8c14d4bd8538703ffdfce.jpg","provenance":{"source_annotation":"icdar2026-competition-data/test/gold_standard_test_set/atomic-layer-etching/simulation-usecase/11/images/f15f209f2d385c0b325b181192c48ae716e73fcf75a8c14d4bd8538703ffdfce.json","source_content":"icdar2026-competition-data/test/gold_standard_test_set/atomic-layer-etching/simulation-usecase/11/content.json","source_pdf":["icdar2026-competition-data/test/gold_standard_test_set/atomic-layer-etching/simulation-usecase/11/Microscopic and data-driven modeling andoperation.pdf"],"main_category":"atomic-layer-etching","sub_category":"simulation-usecase","paper_id":"11","first_classification_panel_id":"a","first_classification_label":"3d scatter plot","caption_source":"content.json:img_caption"},"width":505,"height":442,"image_format":"jpeg","image_sha256":"48c2269aed17d81ab39416e1f6060ff4cf04d71a93225707e3a9e4a550352fd1","metadata_license":"CC BY 4.0","image_license":"source_publisher_rights_reserved","image_reuse_status":"non_commercial_research_use_only","schema_version":"1.0.0"}
{"file_name":"images/test_atomic-layer-etching_simulation-usecase_11_fig_6.jpg","caption":"Fig. 6 - Mass change over time during 3 cycles with purge time of $30s$ . (b) Data set for Step B generated from the microscopic model. Fig. 7 - Data points collected for Step A and Step B at various operating conditions.","id":"test/atomic-layer-etching/simulation-usecase/11/fig_6","sample_id":"atomic-layer-etching/simulation-usecase/11/fig_6","subset":"3d-scatter-plot","split":"test","classification":[{"panel_id":"a","label":"3d scatter plot"}],"summarization":"[{\"panel_id\":\"a\",\"text\":\"3D scatter plot of kMC simulation results for Step B (TMA etching) showing half-cycle time as a function of temperature (533-598 K) and pressure (10-100 Pa). The data reveals strong dependencies on both parameters: half-cycle times range from approximately 2 s at high temperature/pressure to over 20 s at low temperature. This dataset was used to train the FNN model for Step B.\"}]","data_extraction":"[{\"panel_id\":\"a\",\"text\":\"| Temperature (K) | Pressure (Pa) | Time (s) |\\n|----------------|---------------|----------|\\n| 520 | 20 | 38 |\\n| 540 | 20 | 36 |\\n| 560 | 20 | 34 |\\n| 580 | 20 | 32 |\\n| 600 | 20 | 30 |\\n| 520 | 40 | 28 |\\n| 540 | 40 | 26 |\\n| 560 | 40 | 24 |\\n| 580 | 40 | 22 |\\n| 600 | 40 | 20 |\\n| 520 | 60 | 18 |\\n| 540 | 60 | 16 |\\n| 560 | 60 | 14 |\\n| 580 | 60 | 12 |\\n| 600 | 60 | 10 |\\n| 520 | 80 | 8 |\\n| 540 | 80 | 6 |\\n| 560 | 80 | 4 |\\n| 580 | 80 | 2 |\\n| 600 | 80 | 0 |\"}]","vqa":"[{\"panel_id\":\"a\",\"items\":[{\"question_type\":\"Comparative/Trend\",\"questions\":\"\",\"answer_type\":\"Paragraph\",\"answer\":\"Both parameters significantly influence Step B half-cycle time. Temperature has the dominant effect: at low temperatures (~533 K), times exceed 20 s, while at high temperatures (~598 K), times drop to around 2-3 s, reflecting Arrhenius kinetics. Pressure also matters for Step B because TMA adsorption kinetics depend on partial pressure. Higher pressures accelerate the secondary TMA adsorption step, reducing half-cycle time. This dual dependence contrasts with Step A, which is largely pressure-independent.\"},{\"question_type\":\"Process-Oriented\",\"questions\":\"\",\"answer_type\":\"Factoid\",\"answer\":\"Data points were generated by running kMC simulations across temperature (533-598 K in 5 K intervals) and pressure (10-100 Pa in 5 Pa intervals), averaging 10 simulations per condition to reduce stochastic variation.\"},{\"question_type\":\"Application/Performance\",\"questions\":\"\",\"answer_type\":\"Factoid\",\"answer\":\"Each kMC simulation takes approximately 20 minutes on average, and multiple simulations are averaged per operating condition to account for the inherent randomness of the Monte Carlo method.\"},{\"question_type\":\"Structure-Property\",\"questions\":\"\",\"answer_type\":\"List\",\"answer\":\"Temperature in Kelvin (x-axis), Pressure in Pascals (y-axis), and half-cycle time in seconds (z-axis).\"}]}]","bbox":[{"panel_id":"a","x":3,"y":5,"width":502,"height":435}],"source":"icdar2026-competition-data/test/gold_standard_test_set/atomic-layer-etching/simulation-usecase/11/images/fig_6.jpg","provenance":{"source_annotation":"icdar2026-competition-data/test/gold_standard_test_set/atomic-layer-etching/simulation-usecase/11/images/fig_6.json","source_content":"icdar2026-competition-data/test/gold_standard_test_set/atomic-layer-etching/simulation-usecase/11/content.json","source_pdf":["icdar2026-competition-data/test/gold_standard_test_set/atomic-layer-etching/simulation-usecase/11/Microscopic and data-driven modeling andoperation.pdf"],"main_category":"atomic-layer-etching","sub_category":"simulation-usecase","paper_id":"11","first_classification_panel_id":"a","first_classification_label":"3d scatter plot","caption_source":"content.json:img_caption"},"width":505,"height":442,"image_format":"jpeg","image_sha256":"603a875cd5ef217aaada71e57748f1221ca1a573bfc898020fb0c95fdec8441f","metadata_license":"CC BY 4.0","image_license":"source_publisher_rights_reserved","image_reuse_status":"non_commercial_research_use_only","schema_version":"1.0.0"}
{"file_name":"images/test_atomic-layer-etching_simulation-usecase_11_fig_8.jpg","caption":"Fig. 8 - Comparison of the kMC simulation data and of the predicted data calculated by the PIN for Steps A and B. Scatter (a) and line (b) plots of Step A represent a mean squared error of $0.0856\\%$ , which were calculated from $20\\%$ of the data points of the full kMC data set for Step A. Scatter (c) and line (d) plots of Step B represent a mean squared error of $0.175\\%$ , which were calculated from $20\\%$ of the data points of the full kMC data set for Step B.","id":"test/atomic-layer-etching/simulation-usecase/11/fig_8","sample_id":"atomic-layer-etching/simulation-usecase/11/fig_8","subset":"3d-scatter-plot","split":"test","classification":[{"panel_id":"a","label":"3d scatter plot"},{"panel_id":"b","label":"scatter plot"},{"panel_id":"c","label":"3d scatter plot"},{"panel_id":"d","label":"scatter plot"}],"summarization":"[{\"panel_id\":\"a\",\"text\":\"Scatter plot comparing kMC simulation data (red) with FNN-predicted data (blue) for Step A (HF fluorination) across different operating conditions. The close overlap of red and blue points demonstrates excellent model agreement.\"},{\"panel_id\":\"b\",\"text\":\"Parity plot for Step A showing FNN-predicted half-cycle time versus kMC simulation time. Points cluster tightly along the diagonal y=x line, indicating a mean squared error of only 0.0856% on the 20% test dataset.\"},{\"panel_id\":\"c\",\"text\":\"Scatter plot comparing kMC simulation data (red) with FNN-predicted data (blue) for Step B (TMA etching). The agreement is similarly strong, though with slightly more scatter at higher normalized times.\"},{\"panel_id\":\"d\",\"text\":\"Parity plot for Step B showing FNN-predicted versus simulated half-cycle times. The data closely follows the diagonal reference line with a mean squared error of 0.175%, slightly higher than Step A but still excellent.\"}]","data_extraction":"[{\"panel_id\":\"a\",\"text\":\"| Iteration Index | Top Band (Normalized Time, s) | Bottom Band (Normalized Time, s) |\\n|---|---|---|\\n| 0 | 2.40 | 0.60 |\\n| 25 | 2.40 | -0.80 |\\n| 50 | 2.42 | -0.90 |\\n| 75 | 2.45 | -0.90 |\\n| 100 | 2.50 | -0.90 |\\n| 150 | 2.60 | -0.90 |\\n| 200 | 2.70 | -0.90 |\\n| 250 | 2.80 | -0.90 |\"},{\"panel_id\":\"b\",\"text\":\"| Simulation Time (s) | Predicted Time (s) |\\n|---|---|\\n| -0.9 | -0.9 |\\n| 0.0 | 0.0 |\\n| 1.0 | 1.0 |\\n| 2.0 | 2.0 |\\n| 2.7 | 2.7 |\"},{\"panel_id\":\"c\",\"text\":\"| Iteration Index | Top Band (Normalized Time, s) | Bottom Band (Normalized Time, s) |\\n|---|---|---|\\n| 0 | 3.10 | 0.80 |\\n| 25 | 3.00 | -0.30 |\\n| 50 | 2.90 | -0.70 |\\n| 75 | 2.80 | -0.85 |\\n| 100 | 2.65 | -0.95 |\\n| 150 | 2.40 | -1.00 |\\n| 200 | 2.10 | -1.00 |\\n| 250 | 1.80 | -1.00 |\"},{\"panel_id\":\"d\",\"text\":\"| Simulation Time (s) | Predicted Time (s) |\\n|---|---|\\n| -1.0 | -1.0 |\\n| 0.0 | 0.0 |\\n| 1.0 | 1.0 |\\n| 2.0 | 2.0 |\\n| 3.0 | 3.0 |\"}]","vqa":"[{\"panel_id\":\"a\",\"items\":[{\"question_type\":\"Application/Performance\",\"questions\":\"\",\"answer_type\":\"Paragraph\",\"answer\":\"The FNN model dramatically reduces computation time while maintaining high accuracy. While kMC simulations require approximately 20 minutes per data point on average, the trained FNN predicts half-cycle times in less than one second. This speed enables real-time operational parameter calculations, which is essential for semiconductor manufacturing where batch half-cycles typically complete in under 5 seconds. The FNN also allows rapid screening of operating conditions without costly experiments.\"},{\"question_type\":\"Process-Oriented\",\"questions\":\"\",\"answer_type\":\"Factoid\",\"answer\":\"20% of the kMC data points were used for testing and validation, while the remaining 80% were used for training the FNN models.\"}]},{\"panel_id\":\"c\",\"items\":[{\"question_type\":\"Comparative/Trend\",\"questions\":\"\",\"answer_type\":\"Paragraph\",\"answer\":\"The FNN model achieves slightly better accuracy for Step A than Step B. Step A (HF fluorination) shows a mean squared error of 0.0856%, while Step B (TMA etching) has a MSE of 0.175%. This difference likely reflects the greater complexity of Step B's reaction mechanism, which involves more reaction pathways including ligand exchange and DMAF desorption. Nevertheless, both models demonstrate excellent predictive capability with sub-1% errors.\"}]},{\"panel_id\":\"d\",\"items\":[{\"question_type\":\"Structure-Property\",\"questions\":\"\",\"answer_type\":\"Factoid\",\"answer\":\"A point on the diagonal indicates perfect agreement between the FNN-predicted half-cycle time and the kMC-simulated time.\"}]}]","bbox":[{"panel_id":"a","x":12,"y":0,"width":458,"height":384},{"panel_id":"b","x":548,"y":0,"width":468,"height":387},{"panel_id":"c","x":12,"y":426,"width":460,"height":380},{"panel_id":"d","x":536,"y":433,"width":480,"height":373}],"source":"icdar2026-competition-data/test/gold_standard_test_set/atomic-layer-etching/simulation-usecase/11/images/fig_8.jpg","provenance":{"source_annotation":"icdar2026-competition-data/test/gold_standard_test_set/atomic-layer-etching/simulation-usecase/11/images/fig_8.json","source_content":"icdar2026-competition-data/test/gold_standard_test_set/atomic-layer-etching/simulation-usecase/11/content.json","source_pdf":["icdar2026-competition-data/test/gold_standard_test_set/atomic-layer-etching/simulation-usecase/11/Microscopic and data-driven modeling andoperation.pdf"],"main_category":"atomic-layer-etching","sub_category":"simulation-usecase","paper_id":"11","first_classification_panel_id":"a","first_classification_label":"3d scatter plot","caption_source":"content.json:img_caption"},"width":1017,"height":809,"image_format":"jpeg","image_sha256":"81a19a6274ca23cb24bdcdaae12cd8251b223c75aa6718f116ecc8a243dc968b","metadata_license":"CC BY 4.0","image_license":"source_publisher_rights_reserved","image_reuse_status":"non_commercial_research_use_only","schema_version":"1.0.0"}
|