Jinsong Zhang

M.S. in Computer Engineering @ UC Santa Barbara
Email: jinsongzhang [at] ucsb [dot] edu
📢 Looking for Opportunities: I am actively seeking Ph.D. positions for Spring/Fall 2027 in Computer Architecture, AI Accelerators, and ML Systems.

I am currently pursuing my M.S. degree in Computer Engineering at the University of California, Santa Barbara. Previously, I earned my B.Eng. in Electrical and Electronic Engineering from the University of Liverpool.

My research is supervised by Prof. Zheng Zhang, focusing on low-precision tensorized neural network training, hardware-algorithm co-design, and domain-specific hardware accelerators. Previously, I worked as a visiting student in the WINDY Lab at Westlake University under Prof. Shiyu Zhao on autonomous robotic systems and edge perception.

Broadly, I am fascinated by exploring novel computer architectures to break through the memory and compute walls—ranging from spatial/systolic architectures to neuromorphic and brain-inspired computing systems.

Publications

Comprehensive Design Space Exploration for Tensorized Neural Network Hardware Accelerators

Jinsong Zhang*, Minghe Li*, Jiayi Tian, Jinming Lu, Zheng Zhang

Proposed a unified Design Space Exploration (DSE) framework that jointly optimizes tensor contraction paths, hardware PE partitioning, and multi-level dataflow mappings for Tensorized Neural Networks. Deployed and evaluated on a Xilinx Virtex UltraScale+ (VU9P) FPGA, achieving 3.28x–4.00x inference and 3.42x–3.85x training latency speedups over dense baselines.

Education

Technical Skills

Research Projects

Tensorized NN
Comprehensive DSE for Tensorized Neural Network Hardware Accelerators

Jinsong Zhang*, Minghe Li*, Jiayi Tian, Jinming Lu, Zheng Zhang

Graduate Researcher @ UCSB (Jan 2025 – Nov 2025)

Constructed a latency-driven analytical cost model with MAC-guided path pruning to search Pareto-optimal points. Implemented a streaming Tensor Train contraction kernel and reconfigurable systolic GEMM engine in Vitis HLS on Xilinx VU9P, demonstrating up to 4.00x latency speedup with 19.19 GOPS/W energy efficiency across ResNet and ViT.

SNN Architecture
Brain-Inspired Scalable Spiking Neural Networks via Reward Broadcasting

Jinsong Zhang

Graduate Researcher @ UCSB (Jan 2025 – Mar 2025)

Proposed an energy-efficient neuromorphic learning framework combining Reinforcement Learning with Brain-Inspired Reward Broadcasting (BIRB) to counter catastrophic forgetting. Formulated an SNN-Q learning pipeline with temporal-difference updates, achieving 96.88% accuracy on MNIST while cutting training convergence time by 66%.

YOLOv5 Swarm
Distributed Edge Vision & Swarm Perception System

Jinsong Zhang, Jiachen Liang, Zhao Ma

Visiting Research Assistant @ Westlake University (May 2023 – Sep 2023)

Architected an end-to-end distributed swarm vision framework integrating custom YOLOv5s and ROS. Deployed Triton Inference Server on edge robotic platforms, refactoring image preprocessing and ROS topic communications into C++ to slash edge inference latency and maximize control loop throughput.

Selected Projects

RISC-V
Full-Custom 32-bit 5-Stage Pipelined RISC-V Processor (RTL to Transistor)

Jinsong Zhang

UC Santa Barbara (Sep 2024 – Dec 2024)

Architected an RV32I 5-stage pipelined core in Verilog featuring dynamic hazard detection, stall management, and forwarding paths. Designed the entire transistor-level datapath in Cadence Virtuoso using NCSU FreePDK 45nm, achieving timing closure at 50 MHz and verifying cycle-exact behavioral equivalence with RTL via SPICE simulations.

MIPS
FPGA Implementation & Extension of MIPS Processor

Jinsong Zhang

University of Liverpool (Jan 2024 – Mar 2024)

Implemented a multicycle MIPS microprocessor on FPGA with custom instruction extensions in Verilog, optimizing datapath execution cycles.

Real-Time Multi-Sensor Slope Perception Payload for UAVs

Jinsong Zhang

Westlake University (May 2023 – Jul 2023)

Engineered an airborne terrain-monitoring payload combining multi-beam laser rangefinders and an IMU with microcontroller telemetry. Implemented moving-average filtering to suppress motor vibration noise, providing robust slope angle feedback for autonomous terrain-adaptive landing.