Jobs / United States / Samsung Semiconductor INC

Samsung Semiconductor INC

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

San Jose, California, United States Posted yesterday

Apply on employer site ↗ Tailor my CV first Source: Company careers site

Optional: we can tailor a CV and cover letter for this job from your real experience first (the first one is free).

Sponsorship verdict

Sponsorship possible

One solid signal, not two. Worth applying, and worth asking about sponsorship early.

  • Employer is on a government sponsor recordThe US Department of Labor certified 123 H-1B/E-3 labor condition applications for this employer between Oct 2025 and Jun 2026 (latest Jun 2026) — the step every H-1B hire needs first.
  • The posting doesn’t mention sponsorshipSilence isn’t a refusal — ask the recruiter before investing much time.
  • No salary bar for this routeH-1B has no fixed salary bar: the employer must pay at least the prevailing wage for the role and area.
  • Confirmed live todayWhen a source last listed this job as open.
Details and sources
Employer is on a government sponsor record
The US Department of Labor certified 123 H-1B/E-3 labor condition applications for this employer between Oct 2025 and Jun 2026 (latest Jun 2026) — the step every H-1B hire needs first. Source: LCA disclosure data (US Department of Labor (OFLC)).
No salary bar for this route
H-1B has no fixed salary bar: the employer must pay at least the prevailing wage for the role and area. Cap-subject employers enter a lottery weighted by wage level. Source: https://www.federalregister.gov/documents/2025/12/29/2025-23853/weighted-selection-process-for-registrants-and-petitioners-seeking-to-file-cap-subject-h-1b, rules effective 2026-02-27.

US H-1B: cap-subject employers enter a lottery weighted by wage level — Level I gets 1 entry, Level IV gets 4 (DHS projected selection odds ≈15% at Level I to ≈61% at Level IV). Universities and non-profit research employers are cap-exempt. The $100,000 fee for new petitions from abroad is currently blocked by a court order (appeal pending).

A verdict summarises public evidence. It is not legal advice and never a guarantee: the employer and the immigration authority decide. Sign in to factor in where you can already work.

About the role

Please Note: To provide the best candidate experience amidst our high application volumes, each candidate is limited to 10 applications across all open jobs within a 6-month period. Advancing the World’s Technology Together Our technology solutions power the tools you use every day--including smartphones, electric vehicles, hyperscale data centers, IoT devices, and so much more. Here, you’ll have an opportunity to be part of a global leader whose innovative designs are pushing the boundaries of what’s possible and powering the future. We believe innovation and growth are driven by an inclusive culture and a diverse workforce. We’re dedicated to empowering people to be their true selves. Together, we’re building a better tomorrow for our employees, customers, partners, and communities. What You’ll Learn • Project Overview: In memory-centric AI accelerators, many memory devices each process data in place and only exchange small partial results, which are then merged. As LLM attention evolves from grouped-query attention (GQA) to multi-head latent attention (MLA) and sparse top-k attention, the work done locally on each device shrinks while cross-device merge and reduction increasingly dominate decode latency. Using the team’s in-house performance model and simulator, you will explore the design space of merge/reduction architectures — where reduction happens, how devices are connected, and at what precision partial results are exchanged — and propose an architecture that performs well across modern attention workloads. • Skills You’ll Learn: • Architecture-level performance modeling and design-space exploration for distributed, memory-centric AI accelerators • How modern attention variants (GQA, MLA, sparse top-k attention) map onto distributed memory, and how they shift the balance between computation and communication • Hardware–software co-design of interconnect and reduction (collective) hardware for LLM inference What You’ll Do The AGI team at Samsung Semiconductor explores memory-centric system architectures for next-generation AI inference, where computation happens close to where data lives. We co-design hardware and software to make large-scale LLM inference faster and more efficient, and our work informs Samsung’s future memory and accelerator products. As an intern, you will own a research-driven architecture study end to end — from workload modeling to design-space exploration and an architecture proposal — working closely with architects and performance-modeling engineers, and you will present your findings to the broader organization at the end of the summer. Location: Daily onsite presence at our San Jose headquarters in alignment with our Flexible Work polic Reports to: Director • Extend the team’s in-house performance model and simulator to capture modern attention workloads — GQA, MLA, and sparse top-k attention — during LLM decode. • Analyze how each attention variant changes the balance between local computation and cross-device merge/reduction traffic, and pinpoint when merge becomes the bottleneck. • Explore the merge/reduction design space, including where reduction happens (centralized, in-network, or hierarchical), interconnect topology, link bandwidth and latency, and partial-result precision. • Characterize token-selection patterns of sparse attention models and design distributed top-k selection and data-placement strategies that minimize communication and load imbalance. • Estimate first-order area and power costs of proposed reduction hardware and weigh them against performance gains. • Validate the numerical accuracy of reduced-precision merging against full-precision reference implementations. • Collaborate weekly with your mentor and architecture engineers, and share your results

Apply on employer site ↗