Research Areas - CASL: Circuits & Systems Lab

We are actively doing research work on digital hardware design for AI processing at Korea Aerospace University, focusing on low-resource, fault-tolerant, and energy-efficient processors for neural network inference, targeting mission-critical & cost-effective aerospace applications. Our work bridges digital circuit design, machine learning, and aerospace-grade reliability.

🛸Current Key Research Areas

Low-Resource Inference Processors for CNNs, RNNs, and Transformers using binary-weight quantization and pruning.
Fault-Tolerant Neural Processing Units (NPUs) resilient to soft errors and radiation effects.
Efficient Normalization and Softmax Accelerators for large language model (LLM) inference.
HW/SW Co-Design for On-Orbit Federated Learning, enabling distributed AI in space environments.

✨Recent Representative Papers (Mar. 2026)

Δ²-PSUM (FCCM 2026): A low-latency soft-error-resilient binary neural network inference processor.🎞️
Softme^x (TCAD 2026): Lightweight softmax compute engine based only on exponentiation.
MiniBRNN (TCAD 2026): Ultra-compact RNN processor using speculative pruning and interleaved scheduling.🎞️
SR-BIP (TCAD 2025): Soft-error-resilient binary neural network inference processor.🎞️
BiNPU (TCAS-II 2024): Binary CNN processor with every parameter on chip.🎞️
ROSETTA (TETC 2023): Resource-efficient RNN accelerator with dynamic activation pruning.🎞️

🚩See the full publication list.

Our research advances trustworthy and efficient AI computing for edge and aerospace applications, targeting future TRL9-grade inference processors for space missions.