Chenhe Yuan
Veritatem quaero, etiam contra me.

Chenhe Yuan

MEng Student in Electrical & Electronic Engineering
📍 London, United Kingdom

Education

University College London

Sept 2022 - Jun 2026
MEng (Integrated Master's) in Electrical and Electronic Engineering
  • GPA: 3.79/4.0
  • Relevant Coursework: Advanced Digital Design, DSP and Systems, Programming and Control Systems.

The University of Texas at Austin

Aug 2024 - May 2025
Undergraduate Exchange, Electrical and Computer Engineering
  • GPA: 3.5/4.0
  • Coursework: Computer Architecture, Algorithms, Compilers, Digital Electronics.

Systems & Engineering Projects

Low-Latency FPGA Market Data Parser

Apr 2026
Personal Project, Xilinx VU13P
  • Designed a deterministic AXI-Stream UDP packet parser in SystemVerilog for market data price extraction, achieving an approximately 2 ns critical path (WNS +7.94 ns at 100 MHz) on Xilinx VU13P; further optimization ongoing.
  • Implemented strictly ordered fail-fast header filtering (EtherType -> Protocol -> Dst Port) with byte-counter-driven parsing, discarding uninteresting packets at the earliest valid beat to eliminate downstream logic activity.
  • Designed a host-reconfigurable runtime format table with masked discriminator lookup, decoupling payload format changes from FPGA re-synthesis and enabling live feed format updates without hardware rebuilds.
  • Engineered a cross-beat price assembly datapath supporting arbitrary 4-byte alignment across AXI-Stream beat boundaries, with per-lane tkeep gating and single-cycle price-valid output upon full field capture.

LiDAR-Camera Fusion Sparse Inference Architecture for FPGA Classification

Oct 2025 - Jun 2026
Team Project, advised by Prof. R. Killey, University College London
  • Proposed a LiDAR-camera fusion approach that pruned distant background pixels before inference, converting dense visual input into a sparser computation problem.
  • Designed a sparse tiled-convolution architecture to exploit spatial continuity in foreground regions, improving tile occupancy and DSP utilization under on-chip resource limits.
  • Implemented the Jetson-based comparison pipeline and used it to evaluate trade-offs against the FPGA-oriented design.

LC-3B ISA Simulators and Assembler in C

Sep 2024 - Dec 2024
ECE 460N Computer Architecture, The University of Texas at Austin
  • Built LC-3B simulators in C at the instruction-level, microarchitectural, virtual-memory, and cycle-accurate levels to validate behavior across abstraction layers.
  • Extended the microarchitecture with virtual memory support, including user/system address-space separation.
  • Implemented an assembler translating LC-3B assembly into executable binaries with correct starting-address handling.

Research Experience

Streaming Sparse 3D CNN Accelerator on FPGA

May 2025 - Dec 2025
Advisor: Prof. Lizy K. John, The University of Texas at Austin
  • Designed a streaming FPGA accelerator for sparse 3D convolution on Xilinx VU9P, using four-stage pipelines and FIFO buffering to overlap adjacent layers without global synchronization.
  • Built a hierarchical BRAM-based occupancy-bitmap pipeline to skip empty voxel regions online, improving performance by up to 4.4x on spatially sparse layers.
  • Implemented DRAM-backed kernel tiling with URAM accumulation buffers and identified off-chip READ bandwidth as the primary bottleneck through DMA/compute throughput analysis.

Compute-in-Memory Architecture for In-Sensor CNN Evaluation on FPGA

May 2024 - Oct 2024
Advisor: Prof. Chao Li, Shanghai Jiao Tong University
  • Designed a digital emulation of an analog crossbar memristor array for CNN evaluation, translating compute-in-memory behavior into a configurable FPGA prototype.
  • Implemented the design for a Xilinx Zynq-7000 target through synthesis, implementation, and bitstream generation across multiple configurations.
  • Evaluated trade-offs between architectural configurations and workload characteristics to guide further exploration.

Embedded Acceleration Module for CNN Convolution

Jun 2023 - Oct 2023
Advisor: Prof. Tinghuan Chen, The Chinese University of Hong Kong (Shenzhen)
  • Contributed to a parameterized Winograd-based convolution IP generator in Chisel/SystemVerilog, reducing multiplication count in CNN inference layers; authored the templated compute logic in Chisel, enabling configurable array sizes across generated IP variants.
  • Proposed a pipelined synchronization scheme for correct handshaking across pipeline stages in the generated hardware.

Skills

Hardware Description SystemVerilog, Chisel (Scala-embedded), Vitis HLS
Simulation & Verification Icarus Verilog, Verilator
FPGA Vendor Tools Vivado, Quartus
Architecture Research Tools gem5, ChampSim, Accel-Sim, Timeloop/Accelergy, MATLAB
Programming Languages C, C++, Java; experienced with typed and functional programming paradigms (Scala/Chisel, Java generics, C++ templates)

Visitors