02 / Research topics
Potential PhD Research Topics in Computer Engineering
Explore suggested topics and the methods, tools and expertise that could support each investigation.
1
Energy Efficient Edge AI Architectures for Real Time Autonomous and Industrial IoT Systems
Although edge AI enables rapid local decisions, limited energy, memory, and compute capacity restrict deployment at scale. Existing methods often optimize accuracy or latency separately and inadequately address changing workloads. This research could create adaptive hardware-software architectures using workload prediction, model compression, approximate computing, and runtime resource management. Objectives include minimizing energy per inference, preserving safety-critical accuracy, and validating performance across autonomous vehicles, robotics, and industrial sensing platforms.
Applications and use cases
2
Trustworthy Confidential Computing for Heterogeneous Cloud Edge and Federated AI Platforms
As AI workloads move across cloud and edge environments, sensitive data remains vulnerable to inference attacks, malicious infrastructure, and compromised devices. Current confidential-computing solutions frequently impose substantial performance costs and provide incomplete protection for accelerators and distributed learning. The proposed research could develop an attestation-driven architecture combining trusted execution environments, secure accelerators, privacy-preserving learning, and adaptive scheduling. Evaluation should quantify security guarantees, overhead, scalability, and resilience under realistic adversarial conditions.
Applications and use cases
3
Chiplet Based Hardware Architectures for Scalable Sustainable High Performance Computing and AI Acceleration
Chiplet integration promises modular scaling, but communication bottlenecks, thermal hotspots, packaging costs, and reliability issues limit practical adoption. Many existing studies optimize individual chiplets without jointly considering topology, workload mapping, memory movement, and lifecycle sustainability. Chiplet research can introduce an energy-aware architecture and design methodology that co-optimizes interconnects, placement, memory hierarchy, and fault tolerance. Objectives should include improved performance per watt, reduced development cost, and validation using multicore and AI workloads.
Applications and use cases
4
Post Quantum Secure Hardware Accelerators for Resource Constrained Embedded Systems
Post-quantum cryptography is essential because future quantum computers threaten widely deployed public-key mechanisms, yet lattice and hash-based algorithms can exceed embedded resource budgets. Current accelerators often target speed while overlooking side-channel leakage, reconfigurability, and energy constraints. This research could design configurable, formally analyzed accelerators for selected NIST algorithms, integrating masking, instruction extensions, and memory-aware datapaths. Objectives include reducing latency and energy while maintaining provable resistance against implementation-level attacks.
Applications and use cases
5
Digital Twin Driven Cyber Physical Systems for Resilient Smart Manufacturing and Infrastructure
Modern cyber-physical systems face increasingly coordinated faults, cyberattacks, and physical disruptions, while conventional monitoring reacts too slowly or lacks contextual understanding. Digital twins can support prediction and recovery, but current implementations often rely on static models and incomplete synchronization between physical and virtual assets. This research could develop adaptive twins that fuse sensor streams, physics-based models, and machine learning for early anomaly detection and resilient control. Evaluation should measure detection accuracy, recovery time, safety, and operational efficiency.
Applications and use cases
6
Neuromorphic Computing Architectures for Continual Learning in Event Driven Intelligent Sensors
Neuromorphic processors offer exceptional efficiency for sparse event-driven workloads, yet continual learning remains difficult because of limited memory, catastrophic forgetting, device variability, and immature programming models. Existing research often demonstrates isolated algorithms without robust end-to-end sensor-to-decision evaluation. This work could develop adaptive spiking architectures with local learning, memory-aware synaptic updates, and hardware-aware plasticity. Objectives include sustained learning under changing environments, low-power operation, and comparison with conventional edge AI using real event-based sensor streams.
Applications and use cases
7
Advanced Memory Centric Architectures for Data Intensive Scientific and Generative AI Workloads
Data movement increasingly dominates the time and energy of scientific computing and generative AI, particularly for large models and irregular simulations. Conventional processor-centric architectures cannot efficiently exploit locality, high bandwidth memory, or near-memory computation. Rapidly changing workloads also make fixed accelerators difficult to sustain. This research could develop programmable memory-centric architectures with intelligent data placement, compression, and near-data processing. Objectives include reducing movement costs, improving throughput, and demonstrating portability across scientific kernels and transformer workloads.
Applications and use cases
8
Autonomous Hardware Software Co Design for Safe Explainable and Verifiable Robotics
Robotic autonomy must operate under uncertainty while satisfying strict safety, timing, and energy requirements, but current AI platforms rarely provide dependable guarantees across perception, planning, and control. Hardware and software are commonly optimized independently, creating latency and verification gaps. Robotics research can create a co-designed platform combining deterministic acceleration, runtime monitoring, interpretable decision models, and formal safety constraints. Objectives include predictable response, explainable behavior, certified operating envelopes, and robust validation in simulation and physical robots.
Applications and use cases
9
Quantum Classical Hybrid Architectures for Practical Optimization and Machine Learning
Quantum processors may eventually improve selected optimization and learning tasks, but present devices suffer from noise, limited connectivity, compilation overhead, and scarce qubits. Classical systems therefore remain essential, yet existing hybrid architectures lack principled scheduling and resource management. Hybrid computing research could develop hardware-software mechanisms that partition workloads dynamically, mitigate errors, and coordinate quantum and classical execution. Objectives include identifying useful application regimes, minimizing communication overhead, and establishing reproducible performance comparisons against strong classical baselines.
Applications and use cases
10
Sustainable 3D Integrated Systems with Intelligent Thermal Management and Reliability Prediction
Three-dimensional integration improves density and interconnect efficiency but creates severe thermal gradients, aging mechanisms, and difficult repair problems. Existing thermal solutions are often reactive, hardware-expensive, and disconnected from workload behavior and reliability prediction. Sustainable 3D systems could combine embedded sensing, machine learning, dynamic task migration, cooling control, and lifetime-aware design optimization. Objectives should include lower energy and carbon impact, improved mean time to failure, reduced cooling overhead, and experimental validation across representative multilayer architectures.
Applications and use cases