This use case applies AI to improve the design, fabrication, and performance of RF transceiver components for avionics, SATCOM, and radar applications. Machine learning is used to accelerate design optimization, tune device characteristics across RF bands, and detect fabrication imperfections early, such as roughness or process deviations. The use case relies on quality-control data from certified fabrication and characterization ecosystems and may use supervised learning, regression, decision trees, neural networks, and potentially unsupervised learning. The expected benefits include reduced redesign time, shorter fabrication time, optimized cleanroom and infrastructure use, reduced personnel effort, and lower power consumption, consumables, and chemical waste.
| # |
Technology Name |
Description |
| 1 |
Components RF design acceleration |
RF component design, such as RF MEMS design, is a multiparametric and time-consuming task involving parameters such as thicknesses, materials, and their interactions. The use case aims to accelerate redesign by using supervised learning models trained on 3D electromagnetic simulation data, enabling fast predictions when components need to be redesigned to meet new specifications. |
| 2 |
RF components fabrication optimization |
ML models are used to identify fabrication imperfections after each process step and correlate them with characterization outcomes. The goal is to assess whether an imperfection will affect later fabrication steps or become critical for final device performance, supported by ISO-based quality controls and characterization capabilities. |
| 3 |
New manufacturing paradigm for RF components |
Early identification of imperfections enables new manufacturing decisions, such as stopping fabrication early, applying corrective actions, or reusing a device for another application if possible. This supports better use of infrastructure and personnel effort. |
| 4 |
Greener/more sustainable manufacturing for RF components |
The use case supports more sustainable RF component production by reducing utilized materials, including noble metals, and reducing waste while maintaining high fabrication standards. This is especially relevant for nanofabrication, where simple material replacement is often not feasible. |
| 5 |
Development of RF components capable to support diverse applications |
In parallel with the AI-related activities, hardware components are continuously developed to support new specifications. These include GaN-based semiconductor devices, RF MEMS, MEMS-based sensors, and surface-acoustic-wave-based sensors. |
| KPI # |
Related Technology |
KPI Target |
| 1 |
Components RF design acceleration |
Reduce redesign time by 20%, for example when meeting new specifications such as insertion loss in a specific frequency range or when adapting fabrication details such as interface roughness in RF MEMS. |
| 2 |
RF components fabrication optimization |
Reduce fabrication time by 15% by identifying undesirable imperfections at early fabrication stages, such as photoresist thickness deviations for the MEMS airgap. |
| 3 |
New manufacturing paradigm for RF components |
Optimize infrastructure use time and personnel effort by 15%. |
| 4 |
Greener/more sustainable manufacturing for RF components |
Reduce consumables and chemical waste by 10%. |
| 5 |
Development of RF components capable to support diverse applications |
Develop hardware and devices such as RF MEMS, sensors, and HEMTs suitable for various specifications in RF performance, RF power, and reliability. This is not a proposal KPI, but supports the overall goals of the use case. |
| KPI # |
Baseline |
| 1 |
Standard case-by-case design methodology using 3D electromagnetic FEM-based simulation. |
| 2 |
ISO quality control is applied after each fabrication step, but only to evaluate each process step, not to support prediction of final component performance. |
| 3 |
Optimization actions were not possible when imperfections were identified. Such cases were treated as fabrication failures, and only preventive actions for future runs were applied. |
| 4 |
Only standard environmentally friendly procedures, such as suitable disposal, were applied. |
| 5 |
The two teams have developed a technology based on monolithic integration of GaN-based electronics and RF MEMS for a specific high-power TRX module. |
| KPI # |
Validation Method |
| 1 |
RF MEMS and high-frequency electronic components are designed using specialized 3D electromagnetic simulation software. The resulting data are used to train ML models developed within the use case. The trained algorithms are evaluated on unseen test cases, with acceptable error depending on the application. Redesign acceleration is validated by directly comparing the time needed to redesign a device using the conventional simulation-based approach and the trained-model-based prediction approach. |
| 2 |
Fabrication-time reduction is assessed by estimating the time gained when imperfections are identified at different fabrication stages, based on a defined baseline fabrication sequence and the workplan used before the project. |
| 3 |
Infrastructure use and personnel effort are evaluated similarly to KPI 2, with emphasis on the benefit of corrective actions, early stopping, or possible reuse of imperfect devices for other applications. |
| 4 |
Sustainability improvement is assessed through a sustainability monitoring protocol introduced during the project. Material, chemical, and consumable quantities are monitored in detail, and the achieved reduction is calculated from these records. |
| 5 |
Hardware development is validated against application-specific specifications using suitable characterization techniques, such as S-parameters and pull-in voltage measurements. |
- Reduced re-configuration time of the AI-based system by approximately 30–50%.
- Reduced manufacturing costs in the long term.
- Providing documented methods for dynamic application of tools to improve sustainability.
- Improved performance and faster time-to-market.
- Decreased energy consumption for PoC above 10%.
The use case contributes to faster and more efficient RF component development by using ML models to accelerate redesign and reduce dependence on repeated time-consuming 3D electromagnetic FEM simulations. Fabrication optimization contributes to lower manufacturing cost and faster development by identifying critical imperfections earlier in the process, allowing earlier decisions on whether to stop, correct, or redirect fabrication. The new manufacturing paradigm supports more efficient use of cleanroom infrastructure and personnel effort, while the sustainability-oriented activities directly target reductions in consumables and chemical waste. Together, these results support faster time-to-market, lower production effort, more sustainable RF component manufacturing, and improved capability to develop RF devices for diverse SATCOM, radar, avionics, and telecommunications applications.
| AI Toolbox Tool |
Use within the Use Case |
| Tabular Prediction Tool |
Used to train supervised learning, regression, and classification models on RF design, fabrication, and characterization data. It supported fast prediction of RF component behavior and helped reduce redesign effort compared to repeated case-by-case FEM simulations. |
| Tabular Foundation Models |
Used to support rapid modelling on structured RF component datasets, especially when experimenting with design parameters, material properties, fabrication measurements, and characterization outcomes. |
| Device-Level Quality Detection Tool |
Used to detect device-level fabrication imperfections and link process-step measurements to final RF component quality. It supported the identification of defects such as roughness, photoresist thickness deviations, and other process-related issues that may affect device performance. |
| Explainability Tool |
Used to interpret ML models trained on RF design and fabrication data. It supported understanding which design parameters, material properties, or fabrication indicators influenced predicted RF performance and quality outcomes. |
| Time-Series Anomaly Detection Tool |
Used to monitor fabrication and characterization signals over time and detect abnormal process behavior or deviations that could indicate emerging fabrication problems. |