My Research Journey
Research Experience
LLM-Aided Test Case Generation for SoC Design
OngoingBangladesh University of Engineering & Technology (BUET)
Supervisor: Prof. Dr. Anindya Iqbal, CSE, BUET
Developing an intelligent offline verification assistant for System-on-Chip design using Large Language Models. The project focuses on automating comprehensive test case generation while maintaining complete data privacy for chip design companies through on-device deployment of quantized models.
Key Contributions
- Built comprehensive RTL code database for model training
- Trained 7B-parameter quantized model using knowledge distillation
- Developed offline verification assistant ensuring data privacy
Stabiliser Encoder Design & Steane Code Analysis
CompletedUndergraduate Thesis, Bangladesh University of Engineering and Technology (BUET)
Supervisor: Prof. Dr. Sajid Muhaimin Choudhury, EEE, BUET
Extended an existing one-way quantum repeater architecture by developing a general stabiliser encoder framework capable of producing logical |0⟩L and |1⟩L states for any stabiliser code. Applied the framework to implement and evaluate the Steane [[7,1,3]] code as an alternative outer code, analyzing its transmission efficiency and fault tolerance for long-distance quantum communication.
Key Contributions
- Encoder Framework: Designed and implemented a universal stabiliser encoder pipeline — converting stabilisers to binary matrices, applying RREF to obtain standard form, and constructing projector-based logical states verified by eigenstate conditions (Algorithms 4–8).
- Automated Correction Tables: Developed Python algorithms to automatically generate Steane [[7,1,3]] correction lookup files: 384 flagged entries (6×64) and 448 erasure cases (7×64), removing manual dependence for syndrome enumeration.
- Repeater Integration & Analysis: Integrated encoded Steane logical qubits into an existing concatenated one-way repeater model to compute SKR, ηe, Σptrans, and cost metrics. Found that Steane performs slightly worse at εr=0.01% but drastically better for εr>0.01% (SKR ≈10³ at 10,000 km compared to ≈10² for the [[5,1,3]] baseline).
- Simulation Pipeline: Built a reproducible workflow producing binary output for postprocessing in Mathematica; the complete Steane simulation and data analysis ran for ~35 hours after optimization.
Impact
- Provides a general encoder automation toolkit for benchmarking stabiliser codes in concatenated repeater networks.
- Demonstrates that the Steane [[7,1,3]] code offers higher robustness at εr > 0.01%, enabling more reliable long-distance quantum links even with moderate gate errors.
- Outlines extensions for GPU-based simulation and exploration of larger codes ([[8,3,3]], [[11,1,5]]) to scale future repeater studies.
