Refine
Document Type
- Doctoral thesis (2)
Has Fulltext
- yes (2)
Is part of the Bibliography
- no (2)
Language
- English (2)
Keywords
- BICMOS (2) (remove)
Chronic Obstructive Pulmonary Disease (COPD) is an inflammatory lung disease, causing breathing difficulties in patients due to obstructed airflow in lungs. COPD is one of the main leading causes of death worldwide with an annual mortality rate of three million people. Despite the absence of an effective treatment for COPD, an early-stage diagnosis plays a crucial role for the effective management of the disease. However, majority of patients with objective COPD go undiagnosed until late stages of their disease due to the lack of a reliable technology for the recognition and monitoring of COPD in Point-of-Care (PoC).
Alternative diagnostic approaches such as the accurate examination of respiratory tract fluids like saliva can address this issue using a portable biosensor in a home-care environment. Nonetheless, the accurate diagnosis of COPD based on this approach is only possible by concurrent consideration of patients demographic--medical parameters. Therefore, Machine Learning (ML) tools are necessary for the comprehensive recognition of COPD in a PoC setting. On the other hand, drawbacks of cloud-based ML techniques for medical applications such as data safety, immerse energy consumption, and enormous computation requirements need to be addressed for this application. Therefore, the objective of this thesis was to develop a ML-equipped system for the management of COPD in a PoC setup. A portable permittivity biosensor was developed in this work and its in-vitro performance was evaluated throughout clinical experiments. ML techniques were applied on biosensor results, demonstrating the significant role of these algorithms for the recognition of COPD. Moreover, developed ML models were deployed on a neuromorphic platform for addressing the shortcomings of cloud-based approaches.
This work addressed the challenges of accurate mm-wave characterization of devices fabricated in advanced semiconductor technologies. It developed the in-situ calibration solution that is easy to be implemented for silicon technologies. The new technique was verified up to 110 GHz on three difference processes: high performance SiGe:C BiCMOS from IHP Microelectronics (Germany), BiCMOS9MMW from STMicroelectronics (France), and RF CMOS 8SF from IBM Microelectronics (USA). The measurement frequency was solely limited by the capability of the test equipment. Practical results demonstrated that proposed in-situ calibration significantly outperforms the convention method independently on the process specifics and complexity. Some important aspects of the on-wafer S-parameter measurement assurance were presented as well. The discussion included the analysis of the calibration residual errors caused by the improper boundary conditions of coplanar calibration standards and the impact of the RF probe tip design. In conclusion, some suggestions for further accuracy improvement of the proposed method are given.