SoC & Systems Validation Engineer - Ethernet
發佈於: 2026/11/19
Hsinchu Northern Taiwan
Permanent
半導體
Core Responsibilities
-
SoC & Scale-Up Ethernet Integration: Validate integrated Ethernet controller IP within next-generation AI accelerator SoCs from pre-silicon environments through silicon bring-up. Debug interactions across on-chip interconnects (AXI), DMA, memory maps, interrupts, and firmware alongside RTL and firmware teams.
-
High-Speed Fabric & Congestion Tuning: Configure NICs and switches for demanding AI communication patterns. Analyze switch NOS behavior and tune link-layer flow and congestion mechanisms (buffers, queues, PFC, LLR, CBFC) while evaluating next-generation scale-up switch platforms.
-
End-to-End Root-Cause Isolation: Reproduce data-transfer errors and latency/bandwidth bottlenecks using packet captures, hardware counters, logs, and telemetry—pinpointing issues across silicon, firmware, drivers, host software, or switch fabric.
-
AI Cluster Workload Validation: Build and run multi-node clusters executing collective communications (all-reduce, all-to-all) and LLM training/inference workloads to expose real-world corner cases. Measure throughput, tail latency, and long-term stability across configuration changes.
-
Physical Layer & Interconnect Qualification: Analyze MAC/PCS/FEC/SerDes telemetry and validate link integrity across optical transceivers (DR, FR, etc.), copper/active cables (DAC/AEC/AOC), and switch ports. Use CMIS telemetry to monitor module states, firmware, thermals, and optical power across thermal cycles and resets.
-
Unified Rack & Fabric Management: Integrate node BMCs, switches, power shelves/BBUs, PDUs, and thermal subsystems under a single rack management architecture handling hardware inventory, power sequencing, remote control, and firmware updates. Build standards-based interfaces (Redfish, PLDM, gNMI/YANG) and fabric management controls (endpoint addressing, topology discovery, bring-up sequencing, domain partitioning).
-
Telemetry, Observability & RAS: Consolidate power, thermal, per-lane FEC, optical, and device health metrics into unified time-series observability. Design and fault-inject detection, isolation, recovery, and RAS mechanisms.
-
Full-Rack Integration & Open Rack Transition: Build and stage-validate complete racks combining accelerator compute trays, switch trays, power shelves, management networks, and high-speed cabling, while driving migration toward open rack architectures.
-
Lab Automation & Regression Infrastructure: Automate remote console access, power cycling, test execution, and telemetry collection, building dashboards and documentation for continuous regression and long-duration soak testing.
-
AI-Augmented Lab & Engineering Workflows: Incorporate coding and execution agents into daily bring-up, test execution, log analysis, and documentation workflows, while building lab infrastructure (resource locking, remote-control APIs, job queues, verification harnesses) to keep automated runs deterministic and conflict-free.
-
Industry Standards & Vendor Engagement: Track emerging open scale-up and rack standards (OCP, UEC, SONiC ecosystems) and collaborate closely with switch, power, BMC, ODM, and test-equipment partners.
Qualifications
Required Background
-
Bachelor’s degree in Electrical Engineering, Computer Science, or a related technical field (or equivalent hands-on experience).
-
Strong foundation in data-center Ethernet switching, lossless networking/RDMA concepts, and Linux network configuration and debugging.
-
Hands-on experience developing or debugging low-level system software interacting with hardware registers and memory maps (e.g., SoC firmware, Linux kernel/device drivers, silicon bring-up, or hardware diagnostics).
-
Practical experience bringing up, integrating, or benchmarking multi-node rack or cluster systems combining servers, switches, and AI accelerators.
-
Proficiency in Python or shell scripting for hardware control, test automation, and telemetry/log parsing.
-
Active, disciplined use of AI coding/workflow tools paired with rigorous empirical verification against hardware logs and counters.
-
Strong hypothesis-driven debugging methodology and clear cross-functional documentation habits.
Preferred Experience
-
Link bring-up, performance tuning, or interoperability testing in 100G–800G Ethernet environments.
-
Hands-on familiarity with MAC/PCS/FEC/SerDes, QSFP-DD/OSFP optics, DAC/AEC/AOC cabling, and CMIS management/qualification.
-
Experience tuning lossless Ethernet fabrics and congestion control (PFC, ECN, buffer/queue management).
-
Familiarity with open or commercial switch NOS architectures and merchant-silicon switch platforms (e.g., Broadcom Tomahawk class).
-
Experience diagnosing DMA, remote memory access, and collective communication (all-reduce, all-to-all) bottlenecks on NPU/GPU clusters.
-
Experience debugging Ethernet controller IP + SoC subsystem interactions across C/C++ and RTL with hardware and firmware teams.
-
Background in out-of-band server/rack management (BMC/OpenBMC, Redfish, PLDM/MCTP, OpenRMC) or fleet/rack controller development.
-
Experience building hardware validation labs, CI regression frameworks, time-series telemetry, or long-duration stability harnesses.
-
Experience building multi-agent automation infrastructure (device-control APIs, resource scheduling/locking, automated result verification).
-
Familiarity with open rack power/mechanical standards (e.g., Open Rack, 48V/800VDC power shelves, BBUs) or ODM rack bring-up.
-
Familiarity with emerging scale-up interconnect standards (e.g., ESUN, SUE-T, UEC, UALink) or open networking communities.