Skip to main content
Samsung Semiconductor

Intern, LLM Inference Acceleration Architecture (Distributed Memory-Centric Accelerators)

Samsung Semiconductor San Jose, California, United States Internship 1 day ago
Digital Design & SoC Architecture

Originally posted November 18, 2026 by the employer.

About the role

As an LLM Inference Acceleration Architecture Intern in the AGI team at Samsung Semiconductor, you will own a research-driven architecture study exploring memory-centric system architectures for AI inference. The work focuses on design-space exploration of merge and reduction architectures across modern attention workloads, including GQA, MLA, and sparse top-k attention.

What you'll do

  • Extend the team's in-house performance model and simulator to capture attention workloads—GQA, MLA, and sparse top-k attention—during LLM decode.
  • Analyze how attention variants alter the balance between local computation and cross-device merge/reduction traffic to determine communication bottlenecks.
  • Explore the merge/reduction design space, covering reduction location (centralized, in-network, or hierarchical), interconnect topology, link bandwidth, latency, and partial-result precision.
  • Characterize token-selection patterns of sparse attention models and evaluate distributed top-k selection and data-placement strategies.
  • Estimate first-order area and power costs of proposed reduction hardware against performance gains, and validate numerical accuracy against full-precision reference implementations.
  • Present findings and architecture proposals through team meetings, a technical report, and a final presentation.

What you'll need

  • Currently enrolled in a degree program with at least 1 academic quarter/semester remaining.
  • Background in computer architecture, including memory systems, interconnects/networks-on-chip, and performance analysis.
  • Understanding of Transformer inference, including attention variants (MHA/GQA/MLA), KV cache, and the decode phase.
  • Proficiency in Python and/or C++, with hands-on experience using or building architecture simulators or analytical performance models.

Nice to have

  • Pursuing a Master’s or PhD in Computer Engineering, Electrical Engineering, Computer Science, or a related field.
  • Familiarity with sparse attention (e.g., DeepSeek Sparse Attention, NSA, Quest), online-softmax merging (FlashAttention/FlashDecoding), or in-network reduction.
  • Publications in venues such as ISCA, MICRO, HPCA, ASPLOS, or MLSys.

Skills: LLM inference, memory-centric, attention, accelerator architecture, interconnect

Market context
  • Employer posting date: 18 Nov 2026
  • First seen in our index: 18 Nov 2026 (our observation date)
  • Last confirmed on the employer's career site: 19 Nov 2026
  • Specialty: SoC/Chip Architecture (SOC Architecture)
  • Comparable active jobs: 529 at 78 employers, as of 18 Nov 2026
  • Related specialties (Digital Design & Verification family): 3704 active jobs at 129 employers

View all Samsung Semiconductor jobs · SoC design jobs · San Jose jobs

How we counted: postings read from employer career sites; dates are our observations unless labelled as the employer's. How we count

Samsung Semiconductor

Samsung Semiconductor

Apply now
San Jose, California, United States
On-site
Internship
1 day ago

Share this job