Mined from Scientific Literature: Process Schemas for Atomic Layer Deposition and Etching in Materials Science

1. Introduction

• ALD and ALE studies report different information across experiments and simulations, including process conditions and measured properties versus computational methods and predicted quantities.This heterogeneity complicates comparison and reuse across publications and research workflows.Source: §1 · pp. 1–2
• The schemas model individual literature-reported processes, separating deposition from etching and experimental from simulation information requirements.Their nested structures organize materials, conditions, process details, and measured or predicted results.Source: §1 · p. 2
• Schema-miner and schema-miner pro support expert-reviewed schema curation and QUDT grounding of quantitative properties in explicit quantity kinds and permitted units.The schemas distinguish required and optional information and constrain datatypes and controlled values.Source: §1 · p. 2
• The paper presents four coordinated schemas for experimental and simulation processes in ALD and ALE, with structured records published through ORKG templates.The schemas are openly available and domain-expert reviewed; the paper also demonstrates schema-guided extraction of ALD process information.Source: §1 · p. 2

2. Related Work

• Existing semantic resources support broad material, process, and data descriptions, while the proposed schemas target machine-validatable records for individual ALD and ALE processes.The reviewed resources include foundational and mid-level ontologies, research-resource descriptions, and computational-materials infrastructures.Source: §2 · p. 2, §2.1 · pp. 2–3

2.1. Semantic resources for materials data and simulation

• The reviewed resources do not prescribe individual ALD/ALE process records or organize process-specific details such as doses, purge durations, supercycles, and growth or etch per cycle.ASMO represents computational methods and their inputs, parameters, and outputs, while NOMAD organizes computational materials data through extensible metadata.Source: §2.1 · p. 3

2.2. Domain-specific schemas and structured ALD/ALE resources

• Plasma-MDS and LAMAS 4 IC provide domain-specific JSON Schema precedents, but neither covers ALD or ALE while jointly representing experimental and simulation processes.LAMAS 4 IC combines controlled terminology, laboratory validation, and mandatory and optional fields.Source: §2.2 · pp. 3–4
• AtomicLimits databases organize ALD/ALE literature records, but the reviewed resources do not combine process-level validation, required and optional fields, and ontology-grounded quantities and units.The proposed four schemas are literature-derived, domain-expert-reviewed, and QUDT-grounded.Source: §2.2 · p. 4

3. ALD/E Experimental/Simulation Schema Curation Methodology

• Separate schemas were curated for experimental and simulation studies of ALD and ALE because those studies report different process conditions, methods, and results.Experiments report measured conditions and characterized materials, whereas simulations report computational parameters and predicted or modeled quantities.Source: §3 · p. 4

3.1. schema-miner

• Schema-miner generates an initial schema, refines it against representative publications with expert feedback, then applies it to a larger corpus until experts judge it sufficiently complete.Refinement identifies missing properties, underspecified concepts, unsuitable datatypes, and structural inconsistencies.Source: §3.1 · p. 4
• The refinement corpora contained 7 ALD experimental, 7 ALD simulation, 9 ALE experimental, and 10 ALE simulation publications; finalization corpora expanded to 59, 51, 38, and 22, respectively.Domain experts selected studies for scientific quality and coverage of varied materials, conditions, methods, and applications.Source: §3.1 · p. 4

3.2. Ontology Grounding with schema-miner𝑝𝑟𝑜

• Schema-miner pro grounds quantitative properties in QUDT quantity kinds and permitted units, representing each as a value, quantity-kind IRI, and one or more unit IRIs.The schemas were also translated into ORKG templates for structured scholarly knowledge representation.Source: §3.2 · p. 5

4.1. Overview and Scope

• Each record captures one process rather than an entire publication, and the schemas differ in structure and required fields to reflect study-specific reporting requirements.A publication may report multiple processes, such as one chemistry at different temperatures or with different precursors.Source: §4.1 · p. 5

4.2. Common Modeling Principles

• Shared conventions organize properties by process stage, distinguish required from optional fields, and encode quantities as value-and-unit objects grounded in QUDT.The QUDT grounding associates quantity kinds and units with their ontology identifiers.Source: §4.2 · pp. 5–6

4.3. ALD Experimental Schema

• ALD experimental is the largest and deepest schema, using compound-centric reactants, per-element composition, supercycle design, and device-performance properties.It has 275 properties across six top-level groups and reaches eight nesting levels; all six groups are required, but most quantitative fields are optional.Source: §4.3 · pp. 6–7, Figure 1 · p. 8

4.4. ALD Simulation Schema

• ALD simulation spans 212 properties in eight groups, capturing computational settings and predicted growth, surface, film, nucleation, and selective-growth behavior.Only simulationParameters and materials are mandatory, and only the methods field is required within simulationParameters; 24 quantitative properties are grounded in QUDT.Source: §4.4 · p. 7, Figure 2 · p. 9

4.5. ALE Experimental Schema

• ALE experimental requires all eleven top-level properties while leaving most quantitative fields optional, grounding 20 properties across 14 quantity kinds and 20 units.Source: §4.5 · p. 10

4.6. ALE Simulation Schema

• ALE simulation has 275 properties in eight required groups, making it the most prescriptive schema; its results include qualitative mechanisms as well as predicted quantities.It grounds 32 quantitative properties across 14 quantity kinds and 23 units, constraining 29 to physically meaningful numeric ranges.Source: §4.6 · p. 10

4.7. Cross-Schema Comparison

• The schemas share six QUDT quantity kinds and concepts such as process chemistry, materials, conditions, and outcomes, while the simulation schemas overlap in computational and surface-related concepts.The six shared quantity kinds are Energy, Length, Pressure, Temperature, Thickness, and Time.Source: §4.7 · pp. 10–11, Figure 3 · p. 11

5. Application & Use Case

• The schema guided extraction of ZnO and IGZO ALD records from AtomicLimits publications, with conformity checks and chemical-entity normalization to PubChem identifiers.Source: §5 · p. 12
• Typed properties and QUDT representations support automatic checks of data types, required structures, quantity kinds, permitted units, and numerical constraints.A reported growth-per-cycle value of 115 Å was traced and marked incorrect as a film-thickness increment.Source: §5 · p. 12
• Chemistry properties were populated relatively consistently, while growth per cycle, dosing duration, and purge duration were reported less uniformly.Source: §5 · p. 12

6. Conclusion

• The four schemas capture a shared process-level foundation while retaining concepts specific to process and experimental or simulation perspective.Source: §6 · p. 12
• QUDT represents quantitative values through value, unit, and quantity kind, supporting unit-aware validation and cross-publication comparison.Source: §6 · p. 12
• Current ontology alignment covers quantitative properties; future work includes non-quantitative concepts and evaluation on broader literature corpora.The proposed extensions include materials, processing methods, structures, properties, and characterization or simulation activities.Source: §6 · p. 12

Declaration on Generative AI

• AI services assisted with manuscript-text refinement and grammar and spelling checks.Source: Declaration on Generative AI · p. 13
• The authors reviewed and edited the content after using the AI services.Source: Declaration on Generative AI · p. 13
• The authors state that they take full responsibility for the publication’s content.Source: Declaration on Generative AI · p. 13

A. Representative QUDT-Grounded Property Structure

• All quantitative properties across the four schemas use the same quantityValue/quantityKind structure.Source: A. Representative QUDT-Grounded… · p. 15, Figure 4 · p. 15
• Figure 4 uses plasmaPower to show the QUDT-grounded representation of a quantitative property.Source: Figure 4 · p. 15, A. Representative QUDT-Grounded… · p. 15

• The appendix provides the complete hierarchical structures of both ALE schemas, complementing the ALD schema diagrams in Section 1.Source: B. ALE Experimental and Simulation… · p. 15
• An asterisk marks required properties, a dagger marks QUDT-grounded quantitative properties, and a double dagger marks groups whose nested structure is omitted for simplicity.The quantityValue/quantityKind structure of QUDT-grounded properties is not expanded.Source: B. ALE Experimental and Simulation… · p. 15
• Full definitions for every property are available in the published schemas in the Zenodo repository.Source: B. ALE Experimental and Simulation… · p. 15

B.1. ALE Experimental Schema

• The ALE experimental schema distinguishes etchedMaterial, processType, and directionality from properties organizing reactants, conditions, etch-control metrics, operating windows, material properties, and selectivity.Source: Figure 5 · p. 16

B.2. ALE Simulation Schema

• The ALE simulation schema organizes the modeled substrate and surface, process parameters, predicted results, and computational methodology.Source: Figure 6 · p. 17

• The comparison excludes recurring QUDT helper properties and manually maps equivalent properties with schema-specific names to common normalized labels, while keeping conceptually distinct properties separate.Source: C. Domain-Property Overlap Across… · p. 18
• Five properties form a shared core across the four ALD/ALE schemas, while other properties are schema-specific or shared by subsets.Source: Figure 7 · p. 18