Submit Resume

Sr. Principal AI Interconnect Architect

  • California, Milpitas

  • 11/04/2026

  • Full Time

  • Active

Job Description:

  • Job Summary
    An AI Interconnect Architect defines and engineers high-speed networking and communication systems for AI Inference infrastructure which include servers, racks, and chips with a focus on bandwidth, power efficiency, scalability, and optimized transport protocols. Design solutions should take into account interconnect technologies such as PCIe, CXL, co-packaged optics, UALink, Ultra Ethernet, and be able to specify innovative scale-out architectures for optimizing power, cost, and performance.

    Key Responsibilities
    1. System Architecture & Design: Develop architectures for chip-to-chip interconnects and switched fabrics tailored for AI/ML scale-out.
    2. Performance Optimization: Analyze trade-offs in bandwidth, latency, power, area, and reliability.
    3. Technology development and standards: Participate in industry standard bodies and contribute/influence/shape the direction of industry specifications.
    4. Cross-functional Collaboration: Work with SoC, package design, and software teams to ensure seamless integration.
    Required Qualifications
    • Master's or Ph.D. in Electrical Engineering, Computer Engineering, or Computer Science.
    • 10 - 15 years experience developing interconnect technologies including transport and link level protocols, switching fabrics, QoS and reliable communication methods, and Software Defined Networking.
    • Familiarity with various fabric topologies such as Fat tree, Leaf-Spine (Clos), Torus, Meshed and their applicability to various workload and system configurations.
    • AI Hardware Architecture: Familiarity with GPU/accelerator clusters and data center infrastructure
    • Deep, working knowledge of various interconnect technologies and protocols such as PCIe, CXL, NVLink, UALink, Ethernet, Ultra-Ethernet, and serial links.
    • Modeling and SimulationL Ability to develop performance models.

.

.

.