Jason Cheng

Electrical Engineering @ UCSC

Santa Cruz, California, United States

About

Hello! I'm Jason Cheng, a third-year Electrical Engineering student at UC Santa Cruz with a passion for circuit design and semiconductor technology. I’m especially interested in how low-level digital and analog circuits drive the performance of modern computing systems, from embedded devices to large-scale AI accelerators. This summer, I’m conducting research at National Taiwan University, where I’m working on quantization-aware training (QAT) techniques for transformer models like iBERT. My focus is on using efficient integer-based arithmetic and lookup tables to reduce computational cost, bridging the gap between machine learning models and hardware-level efficiency. My core interests include digital system design, signal flow through transistor-level logic, and how physical hardware architectures enable the software we rely on every day. I'm excited by the future of semiconductors and how circuit-level innovation will power the next generation of intelligent systems.

Experience

  • Baskin Engineering at UCSC (Santa Cruz, CA)
    • Undergraduate Researcher, Tactile Manipulation Lab
      May 2026 - Present · 2 mos

    • Undergraduate Researcher, Energy Sustainability Policy and Natural resources (ESPN)
      May 2026 - Present · 2 mos

  • Undergraduate Researcher, Radiological Instrumentation Laboratory (RIL) at Baskin Engineering at UCSC
    Jul 2025 - Dec 2025 · 6 mos

    Radiological Instrumentation Laboratory (RIL): Contributed to PET system modeling using charge transport signal induction (CTSI). Analyzed detector simulation code and system architecture to understand the modeling of charge drift and signal induction in amorphous selenium-based detectors. Organized project files, documented code behavior, and helped develop clearer workflows for future researchers.

  • Undergraduate Research Intern, College of Electrical Engineering at National Taiwan University
    Jun 2025 - Aug 2025 · 3 mos

    Integer-Based Transformer Quantization (iBERT): Worked on quantization-aware training (QAT) and integer-only inference for BERT-style models. Implemented efficient lookup table (LUT) approximations for nonlinear operations like softmax and layer normalization. Debugged training pipelines, analyzed accuracy-performance trade-offs, and prepared the optimized model for deployment on low-power embedded hardware for real-time inference.

  • Swim Instructor at Milpitas Star Aquatics & Fitness
    Oct 2022 - Sep 2024 · 2 yrs

    Taught group and private swim lessons to children and teens at various skill levels. Followed STAR curriculum to develop safe and effective progression in swimming ability. Assisted with swim assessments and collaborated with staff to maintain a fun, supportive learning environment.