[NanoIC topic] AI-Driven semiconductor test data analytics in Heilbronn

The mission

We areaiming to reduce the "Time to Market" in our semiconductor R&D byrevolutionizing how we analyze test data. Currently, our engineers manuallyparse complex JSON files containing parametric extraction and electricalcurves.

Inspired byrecent literature on AI-Driven Semiconductor Test Data Analytics, wewant to build a system that combines a Data Translator with anInteractive AI Agent. Your goal is to create a python-based framework where anengineer can ask natural language questions (e.g., "Show me the defectpattern on Wafer X" or "Compare the yield of Lot A vs LotB") and receive instant statistical insights and visual diagnostics.

Key responsibilities

• Literature review & strategy: Start by analyzing state-of-the-art papers (such as Wang et al. on IEA-Plot and recent TPOR frameworks). Benchmark the pros and cons of using Knowledge Graphs vs. Vector RAG vs. SQL Agents for our specific data topology.
• Universal data translator: Design a pipeline to ingest strandardized JSON data files (curves, parameters, flags) and standardize them into a unified format (Parquet?) suitable for AI querying.
• Agentic AI development: Build a "Code Interpreter" agent using Python (LangChain/LlamaIndex) and local LLMs (Llama 3, Mistral) on our GPU servers. The agent must be capable of writing code to perform statistical aggregations (PCA, Mean, Sigma) without hallucination.
• Advanced visualization: Go beyond basic charts. Implement an automated plotting module capable of generatingdistribution curves, and correlation plots based on the chat context.
• Prototyping: Wrap this technology in a user-friendly web interface (Streamlit) for immediate feedback from R&D engineers. It's possible that the student considers an integration in Azure or in Copilot studio but a stand alone solution is prefered. Decision will be decided after point 1 (Literature review & strategy)

What you will learn

• Research-to-Production: How to take concepts from academic papers and implement them in a real industrial environment.
• Agentic workflows: Mastering the intersection of LLMs and deterministic code execution.
• Semiconductor domain knowledge: Understanding the Parametric Chip/Device MOSFET testing.
• High-Performance computing: Utilizing local GPU infrastructure for secure, private model inference.

Requirements

• Master's student in CS, Data Science, AI, or EE. Ph.D exchange.
• Understanding of LLM architectures (transformers, LSTM) and AI fundamentals is a must. Familiarity with advanced LLMs architecture is preferred. Experience in already trained LLM models locally is preferred.
• Strong Python skills: Experience in deep learning frameworks (tensorflow, pytorch) and related libraries such as pandas, numpy, is a must.
• Academic mindset: You are comfortable reading IEEE/research papers and extracting the methodology to apply it to code.
• GenAI Knowledge: Understanding of RAG (Retrieval Augmented Generation) and LLM limitations.
• Familiarity with Git and Linux/Unix environments.

Nice to Have

• Knowledge of Knowledge Graphs (Neo4j, NetworkX).
• Experience processing spatial data (heatmaps/wafer maps).

How toApply
Pleasesend your CV. Bonus: In your email, brieflymention one challenge you foresee in applying LLMs to numerical scientific data

Type of internship: Master internship, PhD internship

Duration: 6

Required educational background: Computer Science

University promotor: Siegfried Mercelis (UAntwerpen)

Supervising scientist(s): For further information or for application, please contact Jerome Mitard (Jerome.Mitard@imec.be)

The reference code for this position is 2026-INT-074. Mention this reference code in your application.

Only for self-supporting students.

Applications should include the following information:

• resume
• motivation
• current study

Incomplete applications will not be considered.