Kangqi "Amos" Zhang's Homepage

Kangqi "Amos" Zhang

Email: zhkangqi@umich.edu
CV & Google Scholar & GitHub & Linkedin

I am an undergraduate student at the University of Michigan in Computer Engineering and Shanghai Jiao Tong University in Mechanical Engineering, currently working in Crucible Computing Group.

My research lies at the intersection of computer architecture and machine learning systems, with a focus on chiplet-based high-performance computing, AI infrastructures, and ML systems.

Education

B.S.E. in Computer Engineering

College of Engineering

University of Michigan
GPA: 4.0/4.0

2024 - 2026
B.S. in Mechanical Engineering

University of Michigan-Shanghai Jiao Tong University Joint Institute

Shanghai Jiao Tong University, Shanghai

Core GPA: 3.65/4.0

2022 - 2026

Honors and Awards

• Summer Undergraduate Research in Engineering (SURE) Stipend, University of Michigan

2025/05
• Dean's Honor List at College of Engineering, University of Michigan

2024/12, 2025/05

Research Experience

TurnRL: RL-Driven Turn-Restricted Adaptive Routing for Chiplet-Based Systems
- Advisor: Nathaniel Bleier
- Developed fast turn-addition algorithms and baselines for arbitrary topologies.
- Applied Deep Q-Learning to learn optimal turn sets under different traffic patterns.
- Evaluated with gem5 Garnet and PARSEC workloads; demonstrated robust latency reduction and scalability.
2024/09 - Present
Long-Context LLM Inference Acceleration with ANNS Search
- Advisors: Jishen Zhao, Zhijian Liu
- Proposed a CPU-GPU heterogeneous ANN pipeline for efficient KV-cache retrieval.
- Exploited inter-layer query similarity via cross-layer cluster reuse, forming a cross-layer search pipeline.
2025/05 - Present
3D Multi-Physics Modeling of Li-ion Batteries
- Advisor: Yanming Wang
- Built a comprehensive 3D electrochemical-thermal-mechanical model of lithium-ion cells.
- Predicted battery aging behavior, enabling more accurate performance and longevity assessment.
2023/02 - 2024/02

Publications

Mozart: Ecosystem-Accelerator Codesign Framework
- H. Jin, J. Yang, Y. Liu, B. Lyu, K. Zhang, N. Bleier.
- Mozart: An Ecosystem-Accelerator Codesign Framework for Composable Heterogeneous Chiplet Based Neural Network Accelerators.
- ASPLOS 2026, under review.
2026

Intellectual Property

PyBatSim V1.0
Multi-physics Simulation Software for Li-ion Cells
- Developed a 3D coupled electrochemical-thermal-mechanical simulation tool for lithium-ion cells.
- Registered as a software copyright with the China National Copyright Administration.
2024/10

Course Projects

EECS 470
Out-of-Order RISC-V Superscalar Processor
- Advisors: Krisztian Flautner, Ronald G. Dreslinski
- Proposed framework for RTL design on MIPS R10K microarchitecture with RV32-IM ISA.
- Developed advanced features: N-way superscalar execution, early branch resolution, early tag broadcast, prefetching, non-blocking associative caches, and a cycle-accurate simulator.
- Integrated advanced branch prediction subsystems: TAGE, perceptron, and CNN-based predictors, along with a BTB and RAS.
- Achieved an average CPI of 1.5 under 10ns clock; final system ranked top 2 of the course.
2025/02 - 2025/05
EECS 573
Adaptive Feature Selection for Heterogeneous Core Scheduling
- Advisor: Todd Austin
- Extended GEM5/QEMU to extract comprehensive MICA features and modeled realistic migration costs.
- Designed an information-theoretic feature selection framework to identify Pareto-optimal MICA subsets balancing accuracy and overhead.
2025/09 - Present
EECS 471
GPU Optimization of CNN Forward Pass
- Advisor: Reetuparna Das
- Implemented the forward pass of a convolutional neural network (CNN) layer in CUDA with performance-critical kernels.
- Applied shared memory tiling, register-level GEMM transformations, and loop unrolling, reducing runtime from 13.5s to 0.11s.