Photo of Yosen Lin

Yosen Lin

M.S. ECE @ Carnegie Mellon
GPU Performance & Modeling · Parallel Computing · LLM Inference · Systems Software

I grew up in Taipei, Taiwan, and came to engineering from a data science background at UCSD. I'm now a master's student at Carnegie Mellon, graduating in May 2027 and seeking full-time roles starting soon after. This past summer I interned at Realtek, building boot-time profiling and CI tooling for ARM64 Linux, and earlier this year I did GPU performance research at the NEXUS Research Group.

I love the intersection of ML, hardware and software, and I'm especially interested in ML systems, parallel programming and computer architecture. I enjoy working in C, C++, Python and CUDA. In my free time, I like to run, explore cafes and travel.

Education

Aug 2025 –
May 2027 (expected)

M.S., Electrical & Computer Engineering

Carnegie Mellon University · Pittsburgh, PA
  • Coursework: Computer Systems, Computer Architecture, Parallel Programming, Machine Learning Systems, FPGA
Sep 2021 –
Mar 2025

B.S., Data Science

University of California, San Diego · San Diego, CA
  • Coursework: Deep Learning, Optimization, Reinforcement Learning
  • Courses tutored: DSC 40A (4x), DSC 140A, MATH 18

Work & Research

Jul – Aug 2026

Software Engineer Intern

Realtek Semiconductor Corp. · Hsinchu, Taiwan
  • Built boot-time profiling and UART log analysis tooling for ARM64 Linux, cutting measurement variance by 85% (6s to 0.9s) to enable automated boot-regression detection.
  • Profiled 800+ kernel initcalls and found Networking and UART/Serial to be the top two contributors to the boot critical path.
  • Built an end-to-end Jenkins/Gerrit CI pipeline with LLM-based patch review plus automated build and boot validation on an ARM64 SoC, reducing the patch-to-boot-result workflow to 2 manual actions.
Jan – May 2026

Graduate Research Assistant

NEXUS Research Group, CMU · Pittsburgh, PA
  • Developed cooperative CUDA kernels with grid-wide synchronization for color-parallel Gibbs sampling on a Blackwell GPU, scaling to 1M+ spins.
  • Isolated sync overhead with barrier-injection linear regression: grid-sync reaches ~25% of sweep time at 10K spins and fades toward the bandwidth-bound limit at 1M+.
  • Showed the workload is memory bound via roofline analysis and Nsight Compute/Systems, contributing the GPU bottleneck decomposition to a cross-platform performance study.

Projects

Nov 2026 – Present · In progress

Scalable DLRM Inference across GPUs and CPU Memory

Recommendation models have embedding tables too big for a single GPU. This project spreads inference across multiple GPUs and host CPU memory, so models that would normally run out of memory can still be served at scale.

  • CUDA
  • C++
  • Parallel Programming
Website → · Source →
Oct 2026 – Present · In progress

FPGA DRAM Bandwidth Study on the Ultra96-V2

LLM decoding is limited by memory bandwidth, not compute. This project pushes an Ultra96-V2 FPGA toward its DRAM limit on a large fully connected layer and measures what actually moves the needle: memory access settings, clock speed, port configuration and weight precision.

  • Vitis HLS
  • C++
  • FPGA
Feb 2026 – May 2026 · Completed

DAG Operator Fusion Scheduler

In deep neural networks, shuffling data between fast memory and DRAM often costs more than the math itself. This scheduler decides which operators to fuse and which tensors to keep close, slashing data movement across large DNN graphs.

  • C++
  • Operator Fusion
  • Graph Scheduling
  • DNN Dataflow
Details
  • Scaled schedule search with graph-cut decomposition, reducing solver wall time by 88% (43s to 5s) with no loss in schedule quality.
  • Fused MatMul→Pointwise chains with in-situ output activation reuse, keeping intermediates in fast memory and admitting fusion candidates the baseline wrongly pruned.
  • Designed a retain-aware DP that jointly chooses fusion boundaries and which tensors stay resident, contributing ~25% latency reduction on memory-bound benchmarks within a combined 35.8% end-to-end improvement.
Poster (PDF) → · Source →
Jan 2026 – Apr 2026 · Completed

Power-Efficient RV32I Pipelined Processor

A 7-stage pipelined RISC-V CPU in SystemVerilog, optimized for both performance and performance-per-watt.

  • SystemVerilog
  • RTL Design
  • C
Details
  • Architected the pipeline in RTL with hazard detection, data forwarding and branch resolution.
  • Collaborated on a GShare branch predictor and write-back D-cache, contributing to a 26% geomean IPC improvement.
  • Designed a speculative Return Address Stack, cutting JALR mispredictions on FFT from 99.8% to 35%.
  • Reduced the critical path and tuned the cache for a 45% MIPS/Watt gain and 60% higher clock frequency over baseline.
Oct 2025 – Nov 2025 · Completed

Dynamic Memory Allocator

A from-scratch 64-bit malloc/free in C, tuned for memory utilization and throughput.

  • C
  • GDB
  • Memory Management
Details
  • Built segregated free lists with best-fit search, reaching 74.5% utilization at 6,705 Kops throughput.
  • Designed 16-byte mini-blocks with bit-packed header metadata, improving utilization by 24% over the implicit-list baseline.
  • Implemented boundary-tag coalescing for adjacent free blocks, plus a runtime heap checker validating alignment, free-list consistency and block invariants.

Skills

Languages

  • C
  • C++
  • Python
  • Bash

Systems

  • Linux Kernel
  • ARM64
  • Multithreading
  • Synchronization
  • Lock-Free Programming
  • Memory Management

ML Systems

  • PyTorch
  • TensorFlow
  • Operator Fusion
  • Graph Scheduling
  • DNN Dataflow

Parallel Programming

  • CUDA
  • OpenMP
  • MPI
  • SIMD (ISPC)

Profiling & Performance

  • Nsight Systems
  • Nsight Compute
  • Linux perf
  • GDB
  • Roofline Analysis
  • Workload Characterization
  • UART Log Analysis

Hardware

  • RISC-V
  • SystemVerilog
  • RTL Design
  • Vitis HLS
  • Pipelining
  • Caches
  • Memory Hierarchy
  • Branch Prediction
  • Out-of-Order Execution

Simulation

  • gem5
  • SystemC

Build & CI

  • Jenkins
  • Gerrit
  • Git

Runs Completed

Half marathons
  • Rock 'n' Roll San Diego Half Marathon, 2024
  • Taipei Marathon (half), 2025