Principal Engineer, Systems Design Engineering

Own the end-to-end PCIe system design for an NVMe SSD product line across client laptops and enterprise servers, from PHY/MAC review through ASIC/SoC integration, PCIe SFR/register analysis, and firmware design guidelines for robust link training, link transitions, low-power behavior. This role sits at the intersection of PCIe spec compliance, NVMe behavior, FW architecture, platform interoperability, and power/performance tuning.

• Own system‑level PCIe Gen5/Gen6 architecture from an NVMe SSD endpoint perspective
• Define and review PCIe + NVMe integration across SSD products
• PHY + MAC IP review, integration requirements and constraints
• SoC/ASIC integration: clocks, resets, power domains, straps, lane mapping, sidebands
• PCIe SFR + FW guidelines: flow control, LTSSM observability, power states, error handling
• Link & low power transitions: DLRM, L1, L1SS, L0p, ASPM, clock-down, APST Coordination
• Bring-up + debug: enumeration, speed negotiation, width detection, stability, AER/error recovery
• Customer requirement tuning: latency/power, performance, reliability and consistency
• Provide deep expertise in PCIe configuration and extended capability registers, including: Link, power management, MSI/MSI‑X, AER, BARs, L1SS
• Lead platform bring‑up and debug: Enumeration, link training, speed negotiation, power states, error handling
• Act as the technical authority for cross‑team and customer escalations
• Understand criteria for PHY/MAC/controller IP: Gen5/Gen6 readiness, equalization capability, margining hooks, lane mapping flexibility
• SRNS/SRIS tolerance, clocking modes, power management support
• Observability: LTSSM state visibility, error counters, replay/NAK stats, equalization telemetry
• Review IP documents: Reset sequences, compliance features, link speed change support
• L1SS behavior, CLKREQ#/REFCLK control expectations
• AER robustness, surprise down handling, hot/warm reset behavior
• Specify platform-facing requirements: Retimer/redriver compatibility assumptions (backplane/adaptor/cables)
• Integrate PCIe subsystem with: Clocking: REFCLK handling, clock request gating, clock-down sequences
• Resets: PERST# behavior, internal resets, warm/hot resets, FLR support as applicable
• Power domains: retention strategies, wake sources, D-state coordination
• Sidebands: WAKE#, CLKREQ#, presence detect patterns (platform dependent)
• Define lane policy: x4 typical NVMe, lane reversal/polarity, width detection & recovery from degraded width
• Define a clean SFR map that FW uses for: LTSSM control/observability (state, substate, timers, retries)
• Link speed/width control and status (negotiated vs target)
• Low-power triggers: ASPM enable/disable, L1SS policy, L0p policy (if implemented)
• Clock request & clock gating behavior (safe entry/exit rules)
• Error logging counters (replay, NAK, ECRC, timeout, malformed TLPs)
• Recovery controls: link disable/enable, retrain, directed speed change, error clear policy
• Provide FW runbooks: “What to do when”: training fails, width reduces, speed fallback, AER floods
• Safe sequencing across power modes and APST transitions
• Own/define the exact sequencing rules for: Enumeration readiness
• Ensure config space stability, BAR mapping correctness, MSI/MSI-X readiness timing
• Speed negotiation / Directed Speed Change
• When to allow Gen5/Gen4 fallback
• policy for stability vs performance
• Width detection & recovery
• Handling degraded width events (x4 → x2) and reporting/telemetry
• Link power management
• ASPM policy and its constraints with NVMe latency targets
• L1 entry/exit triggers and guard timers
• L1 Substates (L1.1/L1.2) enablement conditions, wake sources, and clock requirements
• DLRM handling (as applicable to platform/system) with safe NVMe readiness on resume
• L0p (if supported) and interaction with performance bursts
• Clock down / clock request
• Define clock request gating conditions, and safe “no transactions in flight” criteria
• NVMe APST alignment
• Coordinate NVMe power states (APST) with PCIe L-states so you don’t create: long resume latencies (client)
• link instability under load (enterprise)
• Own differences across laptop and server: Client: aggressive power policies, fast resume, frequent idle entry/exit, D3hot/cold patterns
• Enterprise: stable performance, high queue depth, error containment, hot-plug-ish behaviors on some platforms
• Validate across: Multiple root complexes, BIOS implementations, OS stacks