¶ UC18 - Semantic Data Model as a Base for AI-Optimized Economic, Ecological, and Societal Supply-Chain Improvements
This use case uses a semantic web-based Digital Reference platform as a foundation for AI-supported semiconductor production and supply-chain optimization. The semantic model enables seamless data exchange while also embedding expert knowledge, allowing AI systems to discover hidden relationships and support tailored economic and ecological improvements. A gaming-based approach is planned to train users when to rely on AI and when human decision-making is needed. The use case targets reductions in variability, improved SPC, yield, OEE, ATP availability, customer service, workforce qualification, and significant CO₂ reduction per transistor function.
| # |
Technology Name |
Description |
| 1 |
Semantic web-based Digital Reference |
A machine-readable and machine-understandable knowledge representation is developed to describe the use-case objectives and semantic relationships. These ontologies allow AI systems to identify hidden knowledge and support tailored production and supply-chain improvements. |
| 2 |
Digital Reference Ecosystem |
The Digital Reference Ecosystem has been extended in collaboration with WP4 through the generation of a semantic model containing 48 classes, 72 edges, and 58 object properties. The model has been uploaded into the current version of the Open Access Platform. |
| 3 |
Open Access Platform |
The Open Access Platform supports visualization of the semantic model and is expected to enable easier ontology generation through semi-automatic ontology generation research. |
| KPI # |
Related Technology |
KPI Target |
| 1 |
Economic and ecological alpha |
Potential for high two-digit percentage improvement in lower CO₂ per transistor function within the next 3–5 years in production. |
| 2 |
SPC potential |
Significant reduction of being off from the middle. The distribution, cp, remains constant, while cpk can improve by a two-digit percentage or more. |
| 3 |
Yield |
As a result of better cpk, 1-yield has the potential to decrease by a mid-percentage range. |
| 4 |
OEE |
As a result of earlier problem detection, 1-OEE has the potential to decrease by a mid-percentage range. |
| 5 |
ATP availability |
As a result of SPC, yield, and OEE improvements, a stabilized overall ATP in a low two-digit percentage range is expected. For missed ATP, a low to mid two-digit percentage improvement is expected. |
| 6 |
Customer service level |
As a result of improved ATP, an absolute increase in customer service level in a low two-digit percentage range is expected. Relative to competition, a mid one-digit percentage improvement is expected. |
| 7 |
Manpower qualification and retention |
Manpower qualification and retention should improve by a low two-digit percentage range. |
| 8 |
CO₂ reduction |
Potential to reduce mid-level two-digit percentage CO₂ emissions, contributing to carbon neutrality efforts. The UC18 target is to maintain approximately 30 times CO₂ savings in semiconductor applications. |
| KPI # |
Baseline |
| 1 |
The economic and ecological alpha concepts had already been researched and published before the project, especially in relation to the ecological operating curve. The baseline at project start was 0 established semantic descriptions, although ecological operating curve research already existed. |
| KPI # |
Validation Method |
| 1 |
First, expert-validated semantic models are generated through conceptual model review, formal consistency checking, rule verification, empirical validation, stakeholder evaluation, and automated semantic validation tools. Then, the semantic model is applied to the economic and ecological alpha concept to evaluate CO₂ improvement in percentage terms. Measurement is based on class definitions and percentage improvement. |
| 2 |
First, an expert-validated semantic model is generated. Then, the model is applied to evaluate the reduction of being off from the middle, where cp remains constant and cpk is expected to improve by a two-digit percentage or more. Measurement is based on class definitions and percentage improvement. |
| 3 |
First, an expert-validated semantic model is generated. Then, the model is applied to evaluate the expected decrease of 1-yield as a result of improved cpk. Measurement is based on class definitions and percentage improvement. |
| 4 |
First, an expert-validated semantic model is generated. Then, the model is applied to evaluate the expected decrease of 1-OEE through earlier problem detection. Measurement is based on class definitions and percentage improvement. |
| 5 |
First, an expert-validated semantic model is generated. Then, the model is applied to evaluate stabilized overall ATP and missed ATP improvements. Measurement is based on class definitions and percentage improvement. |
| 6 |
First, an expert-validated semantic model is generated. Then, the model is applied to evaluate the expected customer service level improvement resulting from improved ATP. |
| 7 |
Validation is based on the cumulative evaluation of models needed for the project. Measurement is based on semantic modelers and their length of stay since project start. |
| 8 |
First, an expert-validated semantic model is generated. Then, the model is applied to evaluate whether the semantic web approach, combined with corresponding datasets, can show potential mid-level two-digit percentage CO₂ emission reductions. Measurement is based on class definitions and percentage improvement. |
- Industrial process phases made more effective and precise through integrated and tailored AI algorithms, methods, and tools.
- Improved overall productivity of the manufacturing factory by approximately 20%.
- Reduced re-configuration time of the AI-based system by approximately 30–50%.
- Providing documented methods for dynamic application of tools to improve sustainability.
- Reduced manufacturing costs in the long term.
- Decreased energy consumption for PoC above 10%.
The use case contributes to Objective 2 by using semantic models as a basis for faster knowledge discovery, data inferencing, and decision support in semiconductor manufacturing and supply-chain contexts. The semantic model enables structured representation of concepts such as the ecological operating curve, ATP, SPC, yield, OEE, customer service level, and CO₂ reduction. In combination with data, these models support faster discovery of relationships that are difficult to identify using traditional methods. The semi-automated ontology generation research in WP4 can reduce the effort required to generate ontology models, while the Digital Reference Model and Open Access Platform support documentation, visualization, and reuse. The use case therefore contributes to improved productivity, sustainable decision-making, and the dynamic application of tools for economic and ecological optimization.
| AI Toolbox Tool |
Use within the Use Case |
| Automatic Ontology Learning Framework |
Used to support the generation and refinement of semantic models for the Digital Reference platform. It helped structure domain knowledge related to supply-chain, production, sustainability, ATP, yield, OEE, and CO₂-related concepts. |
| NL2SPARQL Semantic Error Detection Framework |
Used to support validation of semantic-query interactions with the Digital Reference model. It helped detect potential errors in SPARQL queries used to retrieve and reason over ontology-based production and supply-chain knowledge. |
| Embedding Tool |
Used to represent textual and semantic knowledge from production, supply-chain, and sustainability data in a searchable form, supporting semantic retrieval and similarity-based knowledge discovery. |
| LLM Tool |
Used to support human-facing interaction with the semantic data model, including explanation, knowledge exploration, and AI-assisted reasoning over economic, ecological, and supply-chain improvement opportunities. |
| Prompt Context Injection Tool |
Used to provide LLM-based components with relevant Digital Reference, ontology, supply-chain, and sustainability context, enabling grounded responses and more reliable semantic reasoning. |