CMOS: advanced and beyond
Discover why imec is the premier R&D center for advanced logic & memory devices.
Integrated photonics
Take a look at technologies for optical transceivers, sensors and more.
Health technologies
See how imec brings the power of chip technology to the world of healthcare.
Sensing and actuation
Explore imec’s CMOS- and photonics-based sensing and actuation systems.
Connectivity technology
Look into our reliable, high-performance, low-power network technologies.
More expertises
Discover all our expertises.
Research
Be the first to reap the benefits of imec’s research by joining one of our programs or starting an exclusive bilateral collaboration.
Manufacturing
Go from idea to scalable and reliable semiconductor manufacturing with IC-Link.
Venturing
We create and co-create new ventures, and we invest in and accelerate existing ventures.
Education and workforce development
Imec supports formal and on-the-job training for a range of careers in semiconductors.
NanoIC pilot line
Aligned with the EU Chips Act, access to the pilot line will accelerate beyond-2nm innovation.
Hardware-Software Co-Design for Efficient Knowledge-Graph-Based AI Agents
PhD - Gent | Just now
Accelerate neuro-symbolic AI by co-designing knowledge-graph software and hardware for fast, efficient, and scalable reasoning.
Large language model (LLM)-based AI agents benefit from knowledge graphs by combining neural language understanding with explicit, queryable knowledge, improving factual accuracy, grounding, and explainability. However, integrating knowledge graphs into AI agents introduces significant performance challenges, as agents may generate large numbers of complex and heterogeneous graph queries for retrieval and reasoning. This PhD research will investigate how these emerging workloads can be made faster and more energy- and cost-efficient through hardware-software co-design. The project will first characterize how knowledge graphs (including RDF 1.2-based and labeled property graph systems) are used by AI agents and what query and reasoning workloads they generate. It will then explore how these workloads can be optimized across the computing stack, from knowledge-graph generation and representations, indexing, and query processing to heterogeneous execution on CPUs, GPUs, graph processing units, and emerging memory architectures. Particular attention will be given to SPARQL and other graph query languages, adaptive query optimization, and the trade-offs between flexible CPUs and highly parallel accelerators. By jointly considering AI-agent workloads, knowledge-graph systems, and computer architecture, the research aims to develop efficient platforms for the next generation of neuro-symbolic AI agents, while advancing principles for system technology co-optimization (STCO).
Required background: Computer Science, Engineering Science (Computer Engineering) or equivalent, with an interest in data management, query processing and computer architecture
Type of work: 70% software design and implementation, 20% modeling, simulation and benchmarking, 10% literature
Supervisor: Ruben Taelman (UGent)
Daily advisor: Stefan Lefever
The reference code for this position is 2027-101. Mention this reference code in your application.