CV

Grader, EECS 370: Introduction to Computer Organization

University of Michigan

Teaching Assistant

ECE 2150: Introduction to Electric Circuits

Research Assistant (part-time)

Shanghai Jiao Tong University, Prof. Yanming Wang's group

• Built a 3D electrochemical-thermal-mechanical model of lithium-ion cells in COMSOL to predict battery aging and assess performance and longevity.
• Co-developed PyBatSim V1.0, a registered multi-physics simulation software for Li-ion cells.

Ph.D.

Purdue University

Electrical and Computer Engineering

• Advisor: Prof. Jinghan Sun

B.S.

Shanghai Jiao Tong University

Mechanical Engineering

• Core GPA: 3.65/4.0

B.S.E.

University of Michigan

Computer Engineering

• Summa Cum Laude, GPA: 4.0/4.0
• Advisor: Prof. Nathaniel Bleier

Exchange Student

University of Auckland

Statistics

• GPA: 4.0/4.0

Precision As You Need: Stochastic Computing for Mixed-Precision Inference

Advisor: Prof. Nathaniel Bleier, University of Michigan. NeurIPS 2026 (co-first author) and HPCA 2027.

• Reframed stochastic computing (SC) as a dense adaptive quantizer whose precision is set by bit-stream length, reducing each multiplication to a single AND/XNOR gate.
• Built a GPU library emulating SC matrix multiplication at scale and a retraining-free per-row mixed-precision policy competitive with fixed-format INT quantization.
• Co-developed PaYN, an SC mixture-of-precision accelerator with 128 selectable stream lengths per datapath; in TSMC 22 nm it cuts stream compute by 60–74% within 1.10× FP16 perplexity and reduces EDP by 2.0× over an INT8 array.

Chisel-Verilog Aligner (Undergraduate Thesis)

Advisor: Prof. Nathaniel Bleier, University of Michigan.

• Developed a formally verified aligner using Yosys and FIRRTL to map optimized Verilog signals back to Chisel source with signal-level accuracy, outperforming ML-driven methods and enabling automated HDL debugging.

TurnRL: RL-Driven Turn-Restricted Adaptive Routing for Chiplet-Based Systems

Advisors: Prof. Nathaniel Bleier and Prof. Ronald Dreslinski, University of Michigan. ICCAD 2026 (first author).

• Developed fast turn-addition algorithms and applied GNN-based Deep Q-Learning to learn optimal turn sets for arbitrary topologies and traffic patterns.
• Improved saturation injection rate by 35.9% and 33.3% on HexaMesh and MI300A-like topologies in gem5 Garnet, and reduced queuing and end-to-end latency by 20.1% and 8.8% on congested NAS-NPB workloads.

Double-P: Hierarchical Top-P Sparse Attention for Long-Context LLMs

Advisors: Prof. Jishen Zhao and Prof. Zhijian Liu.

• Designed a hierarchical top-p sparse attention framework with cluster-level top-p estimation via size-weighted centroids and adaptive second-stage refinement.
• Achieved up to 1.8× less attention computation and 1.3× end-to-end decoding speedup over state-of-the-art methods with near-zero accuracy loss.

Long-Context LLM Inference Acceleration with ANN Search

Advisors: Prof. Jishen Zhao and Prof. Zhijian Liu.

• Proposed a CPU-GPU heterogeneous ANN pipeline for KV-cache retrieval with cross-layer cluster reuse and an ILP-based scheduler.
• Achieved up to 3× end-to-end decoding speedup over PQCache at 128K context length while maintaining comparable accuracy.

Course project: EECS 470 Out-of-Order RISC-V Superscalar Processor

University of Michigan.

• Designed the RTL of an out-of-order RV32IM processor based on the MIPS R10K microarchitecture, with N-way superscalar execution, early branch resolution, prefetching, non-blocking caches, an LSQ with forwarding, and TAGE/perceptron/CNN-based branch predictors.
• Achieved an average CPI of 1.68 at a 10 ns clock period; ranked top 2 in the course.

Course project: EECS 570 Directory-Based MSI/MESI Cache Coherence Protocol

University of Michigan.

• Designed latency-optimized three-hop MSI and MESI protocols with an Exclusive state for silent upgrades, verified with Murphi model checking.

Course project: EECS 471 GPU Optimization of CNN Forward Pass

University of Michigan.

• Optimized a CUDA CNN forward-pass layer with shared memory tiling, register-level GEMM, and loop unrolling, reducing runtime from 13.5 s to 0.11 s.

Student Travel Grant, ICCAD 2026

ICCAD

Summer Undergraduate Research in Engineering (SURE) Stipend ($6,000)

University of Michigan

Dean's Honor List (Fall 2024, Winter 2025)

College of Engineering, University of Michigan