Doctoral School

Thesis for defense agenda

Reading date: 13/10/2026

  • CASAPINO ESPINOZA, CARLOS ALBERTO: Estudio de eficiencia térmica de bloques de tierra (adobe), y propuesta de desarrollo de prototipo para clima frío. Caso de estudio altoandino Perú.
    Author: CASAPINO ESPINOZA, CARLOS ALBERTO
    Programme: DOCTORAL DEGREE IN ARCHITECTURAL, BUILDING CONSTRUCTION AND URBANISM TECHNOLOGY
    Department: Department of Architectural Technology (TA)
    Mode: Article-based thesis
    Deposit date: 28/07/2026
    Reading date: 13/10/2026
    Reading time: 16:00
    Reading place: ETSAB (Esc.Téc.Sup.Arquit.Bcn)-Pl. Baja-Sala GradosAv. Diagonal, 649Enlace a videoconferencia: https://meet.google.com/xki-ujnw-hqvConnexión 15:30h
    Thesis director: GOMEZ SOBERON, JOSE MANUEL VICENTE | GOMEZ SOBERON, MARIA CONSOLACION
    Thesis abstract: Buildings that use earth as a component of their construction systems represent between 30 and 50% worldwide, with most of these structures being residential and located in rural areas or developing countries. Earth is an economical and sustainable solution with excellent hygrothermal regulation. However, it presents limitations such as lower mechanical resistance and reduced durability in humid conditions. To mitigate these limitations, chemical stabilization (using lime or cement) or compression is commonly employed, which increases production costs, embodied energy, and CO2 emissions.At the same time, the generation of different types of waste, such as construction and demolition waste, mineral waste, and other materials like used tires, represents a serious environmental problem. Incorporating this waste as second-generation raw materials in unfired earth blocks emerges as a circular economy alternative to improve their properties without additives or stabilization methods. Therefore, the objective is to evaluate the technical feasibility and compatibility of incorporating different types of recycled aggregates, such as crushed ceramics (CCB), powdered soapstone (PSR), and end-of-life tires (CTW), by analyzing their effects on the various properties of earth blocks manufactured using traditional methods, without stabilizers or firing.The experimental methodology was structured in three stages: The first stage involved characterizing the soil and the different aggregates. The second stage focused on the mix design and manufacturing process, defining the replacement and curing percentages. The third stage consisted of the experimental campaign, evaluating, under standardized tests, density, porosity, hygroscopic expansion, compressive and flexural strength, elastic modulus, thermogravimetric analysis, thermal tests, and durability tests; these tests are addressed in the articles that comprise this thesis.The results obtained from the recycled aggregates studied demonstrate the possibility of their use through incorporation into soil matrices under regulatory standards, representing alternatives for reducing energy demand, reducing production costs, improving the insulating capacity for CCB and CTW cases, and in the case of PSR it was presented as an alternative mineral stabilizer with the potential to increase mechanical resistance and optimize thermal behavior for energy absorption, thus presenting themselves as construction alternatives for cold climates, also representing alternatives for the management of these wastes.
  • HALDANKAR, RAJASHREE: Strain-Induced Buckling in Suspended Graphene–hBN and Ferroelectric Moiré Domains in Twisted hBN
    Author: HALDANKAR, RAJASHREE
    Programme: DOCTORAL DEGREE IN PHOTONICS
    Department: Institute of Photonic Sciences (ICFO)
    Mode: Normal
    Deposit date: 03/09/2026
    Reading date: 13/10/2026
    Reading time: 10:00
    Reading place: ICFO Auditorium
    Thesis director: BACHTOLD, ADRIAN
    Thesis abstract: This thesis investigates the mechanical and electrostatic behaviour of van der Waals heterostructures based on graphene and hexagonal boron nitride (hBN). The work is centred on two related questions: how fabrication-induced strain affects suspended graphene–hBN devices, and how scanning-probe measurements can be used to study moiré domains in twisted hBN. The first part of the thesis focuses on suspended graphene–hBN heterostructures. These devices were fabricated by dry transfer onto pre-patterned trenches, followed by suspension release using supercritical CO2 drying. After release, the suspended stacks did not remain flat, but instead developed smooth out of-plane buckled profiles. Atomic force microscopy measurements show that this buckling is consistent with built-in compressive strain introduced during fabrication. Thermal-expansion mismatch, transfer-induced stress, and clamping at the contacts are all likely to contribute to the final mechanical state. Electrical measurements under gate bias further show that electrostatic loading softens the upward-buckled configuration and can drive a snap-through transition into a downward-buckled state. In the present devices, this transition is observed from the up state to the down state, while controlled switching back to the up state is not demonstrated. The second part of the thesis studies sliding ferroelectricity in marginally twisted hBN. In these devices, the small relative twist between the hBN layers produces a reconstructed moiré pattern formed by alternating stacking domains. Kelvin probe force microscopy was used to measure the local electrostatic response of these domains. Bias-dependent measurements show that the two domain types have distinct Kelvin- null positions, with an untwisted hBN reference region lying approximately between them. The comparison between first- and second-harmonic responses supports the interpretation that the observed domain contrast is mainly governed by local contact-potential differences rather than by purely capacitive variations. The KPFM domain image is also analysed as a real-space map of the reconstructed moiré network. Representative domain centres are extracted from the image and used to quantify the local moiré geometry. This analysis provides a moiré-scale description of local wavelength, effective twist variation, and network disorder. These quantities are interpreted as geometrical descriptors of the reconstructed domain pattern, not as direct measurements of atomic-scale strain. Overall, this thesis shows that strain, electrostatics, and interfacial polarisation are central to the behaviour of graphene–hBN and twisted-hBN heterostructures. The suspended graphene–hBN devices demonstrate how residual strain controls mechanical stability and electrostatic actuation, while the twisted-hBN measurements show how KPFM can be used to study moiré domains. Together, these results provide a basis for future studies of strain-controlled nanomechanics and electrostatic domain mapping in two-dimensional materials.

Reading date: 14/10/2026

  • RUÍZ GONZÁLEZ, JOSÉ JAVIER: Development of computational tools and pipelines for enhanced interpretation of Raman spectroscopy in biomedical applications
    Author: RUÍZ GONZÁLEZ, JOSÉ JAVIER
    Programme: DOCTORAL DEGREE IN PHOTONICS
    Department: Institute of Photonic Sciences (ICFO)
    Mode: Normal
    Deposit date: 07/07/2026
    Reading date: 14/10/2026
    Reading time: 10:00
    Reading place: ICFO Auditorium
    Thesis director: LOZA ALVAREZ, PABLO
    Thesis abstract: Raman spectroscopy has gained increasing attention over the last few decades due to its ability to non-destructively extract chemical information from samples. However, despite advances in instrumentation, it has not yet become standard practice in the biomedical field. One of the main reasons is the difficulty in interpreting and extracting the chemical information encoded in Raman signal, so as to validate the technique using complementary, well established techniques.Considering this, the aim of this Thesis is to develop tools and analytical pipelines that enhance the biochemical interpretation of Raman spectra, and to demonstrate its usefulness in two relevant biomedical research lines: breast cancer resistance to neoadjuvant treatments and the diagnosis of choroidal melanoma.Firstly, we developed and validated RamanBiolib, an open-source Raman spectral library accompanied by search algorithms, aimed at providing rapid and objective identification of biomolecules. Secondly, we used RamanBiolib after spectral unmixing to identify and quantify cytochrome c molecule in breast cancer cells resistant to neoadjuvant treatments. This revealed a shift towards its reduced state to avoid apoptosis and acquire drug resistance. Then we also characterized lipid droplets in triple-negative breast cancer cells, highlighting the relevance of variable importance analysis in classification models. This approach allows for the biochemical interpretation of their classification performance and assessing possible background contributions to their accuracy. By doing so, significant changes in lipid unsaturation were detected. Furthermore, we developed RamanTF, a transformer-based algorithm that automatically quantifies key biomolecules from raw spectra.Finally, to study melanin-rich samples, we defined a pre-processing and analysis workflow to correct instrumental interferences that were observed in previous studies, but not corrected yet. With this established workflow, key melanin properties were identified, such as melanin structural disorder or defects abundance, which can be used as biomarkers to diagnose choroidal melanomas. Overall, the tools, discussions, and results presented in this Thesis advance Raman spectroscopy towards clinical application and provide a basis for its widespread adoption as a routine technique in biomedical and diagnostic research.
  • SIRACUSA, MARCO: Hardware-Software Co-Design for Accelerating Sparse Tensor Algebra
    Author: SIRACUSA, MARCO
    Programme: DOCTORAL DEGREE IN COMPUTER ARCHITECTURE
    Department: Department of Computer Architecture (DAC)
    Mode: Normal
    Deposit date: 23/07/2026
    Reading date: 14/10/2026
    Reading time: 17:00
    Reading place: C6-E101
    Thesis director: MORETÓ PLANAS, MIQUEL | ARMEJACH SANOSA, ADRIÀ
    Thesis abstract: Many modern scientific and machine learning workloads exhibit inherently sparse and irregular structure. However, traditional supercomputers and datacenters are optimized for dense, regular computation. As a result, sparse workloads use only a fraction of available system resources, leaving significant performance untapped. Although prior solutions have been proposed, they typically target narrow classes of sparse tensor algebra operations, do not fully leverage modern vector compute cores, or leave much of the programmability burden to the user, limiting widespread adoption. To address this gap, this thesis presents a hardware–software co-design for the efficient execution of general sparse tensor algebra operations. We begin by analyzing the architectural implications of sparsity across a broad range of scientific and machine learning applications. Our analysis reveals that traditional compute cores are fundamentally ill-suited to the irregular tensor traversal and coordinate merging required to operate on sparse tensor formats. Building on these insights, we design a Decoupled Access–Execute (DAE) architecture that offloads these operations to a specialized unit, the Tensor Marshaling Unit (TMU), while leaving computation on the core. We demonstrate that the resulting DAE architecture outperforms traditional CPUs and GPUs by up to an order of magnitude. To deliver these gains without exposing the complexity of DAE programming to end users, we also present Ember, a compiler that lowers sparse machine learning operations to DAE code for architectures such as TMU–CPU systems. Ember is implemented in MLIR, enabling natural integration with PyTorch and TensorFlow. By automating this mapping, Ember unlocks the full performance potential of the TMU without imposing an additional programmability burden.

Reading date: 15/10/2026

  • PAREDES AHUMADA, JUAN ANTONIO: Improving Data-Driven Estimations of Physical Fields in IoT Sensor Networks
    Author: PAREDES AHUMADA, JUAN ANTONIO
    Programme: DOCTORAL DEGREE IN COMPUTER ARCHITECTURE
    Department: Department of Computer Architecture (DAC)
    Mode: Normal
    Deposit date: 29/07/2026
    Reading date: 15/10/2026
    Reading time: 15:00
    Reading place: Sala C6-E101
    Thesis director: GARCÍA VIDAL, JORGE | BARCELÓ ORDINAS, JOSE MARIA
    Thesis abstract: This thesis explores data-driven strategies for monitoring physical fields using (IoT} sensor networks. In doing so, we focus on improving the estimation of physical quantities at two complementary levels. First, at the local level, measurements collected at individual nodes are used to estimate a new physical quantity not previously monitored at that location, through Machine Learning (ML)-based virtual sensors. Second, at the network-wide level, spatially distributed measurements are leveraged to determine the optimal arrangement of sensors for reconstructing the whole physical field over a large area. This is achieved by formulating and solving the sensor placement problem using low-dimensional representations of the measured signal.Thus, the first part of this thesis focuses on developing virtual sensors based on proxy models that estimate non-regulated and rarely monitored pollutants, like black carbon (BC) mass concentration, from measurements of other physical variables. The relevant environmental variables are identified and the impact of different meteorological conditions, such as seasonality, on model's performance are studied as well. The evaluation of the proxy model shows how this data-driven technique may help to reduce the infrastructure cost to improve the exposure assessment in urban Mediterranean environments. Furthermore, the black carbon proxy model is coupled with low-cost sensor (LCS) nodes, addressing data quality challenges that may impact on obtaining a reliable estimate in real-world sensing scenarios, such as data noise, missing values, and varying sampling intervals. A robust machine learning proxy (RMLP) to estimate BC mass concentrations is developed. The robust proxy model incorporates key pre-processing tasks and constitutes an improvement over basic proxy models that ignore these phenomena. The second part of this thesis addresses the optimal sensor placement for signal recovery. Using spatially distributed measurements, the values of a scalar physical field are estimated at multiple locations. To do so, the measured signal is represented in terms of a low-rank structure which is exploited to significantly reduce the number of locations necessary to infer the values of the field of interest. In particular, the sensor placement problem is formulated for multiclass monitoring networks, consisting of different classes of devices with varying accuracy. The results show how the proposed algorithm is suitable for monitoring networks that present a great disparity between device noises or scenarios with a limited number of devices. Additionally, the sizing and placement problem is re-framed to achieve a more rigorous control over the reconstruction error. In consequence, the Sizing and Placement with Coordinate Reconstruction ErrorConstraint (SP-CREC) algorithm is proposed, which bounds the per-coordinate error variance of the reconstructed signal. This allows assigning specific thresholds to certain regions of interest within the network, where the error variance can be more or less restricted, according to monitoring needs. The results show that SC-CREC successfully complies with heterogeneous thresholds while deploying a low number of devices. As a whole, the proposed methods improve the monitoring capability of IoT sensor networks by enabling accurate reconstruction of physical quantities from limited sensor data under different deployment constraints.

More thesis authorized for defense