TalTech invites applications for a fully funded PhD position focused on the security, reliability, and efficient deployment of AI Transformer systems. The project examines how quantization, pruning, sparsity, KV-cache compression, and approximation affect vulnerability to side-channel attacks, fault attacks, hardware Trojans, and soft errors, and develops lightweight protections for secure and dependable Transformer inference on modern AI accelerators.
The project will address the following key areas:
Transformer-based models, including Large Language Models (LLMs), Vision Transformers (ViTs), and Vision-Language Models (VLMs), are now central to artificial intelligence, but their growing computational and memory demands make efficient deployment essential on resource- and energy-constrained platforms. The project will study how optimization methods such as reduced-precision quantization, pruning and sparsity, KV-cache compression, and approximate computing change the security and reliability properties of Transformer-based AI systems. These methods can alter numerical error propagation, reduce redundancy, increase sensitivity to corrupted values, or introduce auxiliary structures such as scale factors, zero-points, sparsity indices, cache metadata, and control information. The research will characterize the effects of these changes across representative Transformer architectures and AI accelerators.
The candidate will investigate threats including side-channel attacks, fault attacks, hardware Trojans, and reliability events such as soft errors. The work will identify vulnerable components and analyze how errors or malicious modifications propagate through model weights, activations, attention mechanisms, memory hierarchies, KV caches, and accelerator control paths. The project will then develop and evaluate lightweight protection, detection, and mitigation mechanisms that preserve the efficiency benefits of optimized inference. Evaluation will consider security, reliability, model quality, performance, memory footprint, energy consumption, and hardware overhead. The expected outcome is a set of methodologies and practical countermeasures for dependable, secure, and efficient Transformer inference, supported by reproducible experimental results and scientific publications.
Responsibilities and (foreseen) tasks:
Define research questions, experimental methodology, and evaluation criteria for secure and reliable efficient Transformer systems.
Implement and benchmark representative Transformer models, optimization techniques, and AI accelerator configurations.
Analyze side-channel, fault-injection, hardware Trojan, and soft-error threats across model, memory, cache, and accelerator components.
Develop lightweight protection, detection, and mitigation mechanisms that balance security, reliability, accuracy, performance, energy use, and hardware overhead.
Design reproducible experiments, document results, and communicate findings through research reports, peer-reviewed publications, and conference presentations.
Collaborate with the supervisor and TalTech researchers, participate in scientific discussions, and contribute to project-related activities.
The research result is expected to be published in at least 6 high-quality journals (Q1-Q2) and pioneer conferences (A1-A2-B1).
The
PhD candidate is expecting to assisting in teaching and help in laboratory on the related topic.
Applicants should fulfil the following requirements:
A Master's degree in computer engineering, computer science, artificial intelligence, or a closely related field.
A strong background in machine learning and deep learning.
A good understanding of Transformer architectures and modern AI models.
Good programming skills in Python, including practical experience with PyTorch.
A good understanding of computer architecture, digital systems, and/or AI accelerators.
Strong written and spoken English.
Ability to conduct independent research, analyze technical literature, and work collaboratively.
Motivation to pursue doctoral research at the intersection of AI, hardware security, and system reliability.
The following experience is beneficial:
Previous research or project experience with LLMs, Vision Transformers, or vision-language models.
Familiarity with quantization, pruning, sparsity, KV-cache compression, or approximate computing.
Knowledge of hardware security, fault tolerance, dependable computing, or hardware reliability.
Experience with GPU, FPGA, embedded, or specialized AI accelerator platforms.
Experience with scientific programming, benchmarking, or reproducible experimentation.
Research publications, open-source contributions, or participation in collaborative research projects.
We offer:
A fully funded PhD position in an internationally relevant research area.
Dedicated supervision by supervisors and regular interaction with researchers at TalTech.
The opportunity to work at the intersection of efficient AI, hardware security, reliability, and accelerator design.
An international and collaborative research environment with opportunities for scientific networking.
Support for doctoral training, research dissemination, peer-reviewed publications, and conference participation.
A research position in Tallinn, Estonia, with access to TalTech infrastructure and a strong digital-technology ecosystem.
Supervisors:
Main supervisor: Assistant Professor Tara Ghasempouri, School of Information Technologies: Department of Computer Systems: Centre for Dependable Computing Systems
Co-Supervisor: Researcher Mohammad Hasan Ahmadilivani, School of Information Technologies: Department of Computer Systems: Centre for Dependable Computing Systems
Tallinn University of Technology (TalTech) is an international scientific community with approximately 9,000 students and 2,000 employees; it is one of the largest universities in Estonia, the leading EU country in digitalisation. The university's strengths are broad multidisciplinary study/research interests, a modern research environment, and strong collaboration with international educational and research institutions. TalTech is aiming to be an organisation leading the way to a sustainable digital future.
The project will be hosted by the Department of Computer Systems at Tallinn University of Technology, in a research environment connected with dependable computing systems. The department brings together expertise in computer architecture, digital systems, artificial intelligence, hardware design, cybersecurity, and system reliability. Its research addresses computing platforms that execute demanding workloads efficiently, securely, and reliably. The PhD candidate will work at the intersection of Transformer models, AI accelerators, hardware security, and fault-tolerant design, benefiting from collaboration with faculty members and researchers across TalTech. The department supports independent doctoral research, scientific discussion, access to research infrastructure, and dissemination through peer-reviewed publications and international conferences. This international and collaborative environment suits candidates who want to connect AI algorithms with real-world hardware.
For information about the admission process, please visit the PhD Admission homepage
Applications can be submitted from 16.09.2026 to 15.10.2026
For further info, please contact:
Prof. Tara Ghasempouri, Tallinn University of Technology (TalTech), at tara.ghasempouri@taltech.ee. The successful candidate will work under her supervision and collaborate with faculty members and researchers at TalTech. For questions regarding the application process, please contact docstudy@taltech.ee
