PCIe Gen4 x8 with ReDriver to SlimSAS 8i (SFF-8654) Add-in-Card

A practical design and deployment guide with FAQ for engineers and system integrators

---

Introduction

The demand for high-speed interconnects has grown rapidly with PCIe Gen4 becoming mainstream in servers, storage appliances, and high-performance workstations. One increasingly common design is an Add-in-Card (AIC) that takes an x8 PCIe Gen4 interface from the host and routes it to a SlimSAS 8i (SFF-8654) connector.

To ensure signal integrity across the entire path (slot → AIC → SlimSAS cable → backplane/device), the card typically includes a ReDriver, which compensates for signal degradation by equalizing and re-shaping the eye diagram.

This blog will walk you through the architecture, design considerations, signal integrity challenges, validation strategies, and deployment notes for such a card, followed by a Q&A section addressing common questions.

---

System Architecture

1. Host PCIe Slot (Gen4 x8)
   • Provides the initial PCIe lanes (8 total at 16 GT/s each).
   • Supports bifurcation into 2×x4 if the platform allows.
2. ReDriver (Linear Equalizer)
   • Restores degraded signals caused by PCB traces, connectors, and cable losses.
   • Low latency compared to Retimers (nanoseconds vs tens of nanoseconds).
   • Often configurable via I²C or straps for equalization gain, output swing, and de-emphasis.
3. SlimSAS 8i Connector (SFF-8654)
   • Breakout to an 8-lane external cable.
   • Connects to storage backplanes (U.2/U.3, E1.S), PCIe switches, or accelerator shelves.
   • Supports sidebands and refclk depending on implementation.

---

Why a ReDriver?

At 16 GT/s (Gen4), PCIe channels can only tolerate so much insertion loss before failing link training or downshifting to Gen3. A ReDriver:

• Compensates insertion loss (typically 6–12 dB improvement).
• Maintains low latency, crucial for storage and accelerator workloads.
• Reduces cost and complexity compared to Retimers (no clock recovery, no protocol awareness).

---

Key Design Considerations

1. Signal Integrity

• Loss budget: Keep end-to-end ≤ ~28–30 dB.
• Impedance: Maintain 85 Ω differential with tight tolerance.
• Vias: Minimize stub lengths; use back-drilling if possible.
• Skew: Intra-pair < 3 ps, inter-pair < 10 ps.
• Connector breakout: Keep trace lengths short and symmetric.

2. Lane Mapping & Polarity

• Preserve lane ordering (0–7).
• Polarity inversion is acceptable if supported by both the ReDriver and endpoint.
• If bifurcation is used, ensure BIOS/UEFI settings align with cable/backplane topology.

3. Clocking & Sidebands

• SRIS: Devices generate their own refclk.
• Common refclk: Required for some backplanes and switch shelves.
• Sidebands like PERST#, CLKREQ#, and SMBus may or may not be routed depending on the system.

4. Power & Thermal

• Typical ReDriver power draw: 0.5–1.5 W.
• Provide local decoupling and airflow to maintain junction < 100 °C.
• Add heatsinks for dense or hot environments.

---

Validation Strategy

1. Bring-up tests
   • Cap to Gen3 first to confirm basic connectivity.
   • Remove cap and verify Gen4 stability.
   • Use both short and long SlimSAS cables.
2. Signal integrity validation
   • Eye diagram checks before and after ReDriver.
   • Bit Error Rate (BER) testing to 1e-12 or better.
   • EQ sweep for optimal preset combinations.
3. System tests
   • Device enumeration across reboots.
   • NVMe stress tests with fio/IOmeter.
   • Hot-plug cycles if required.

---

Deployment Use Cases

• NVMe backplanes: U.2/U.3 bays via SlimSAS cabling.
• Switch shelves: Extending lanes to PCIe switch enclosures.
• Accelerator offload: Connecting GPUs, FPGAs, or DPUs via external PCIe cabling.
• E1.S carriers: High-density storage nodes needing low-loss cabling.

---

Common Pitfalls

• Assuming all SlimSAS pinouts are the same (they vary by vendor).
• Not exposing EQ tuning options (straps or I²C).
• Forgetting thermal design (ReDrivers can overheat in dense servers).
• Overusing EMI filters that add unnecessary insertion loss.

---

Q&A / FAQ

Q1: How long a SlimSAS cable can I use at Gen4 x8?
A: Typically 0.3–0.7 m depending on AWG and shielding. Beyond that, signal loss often forces Gen3 fallback unless a Retimer is used.

Q2: Why not just route PCIe lanes directly to SlimSAS without a ReDriver?
A: At Gen4 speeds, the combination of slot + AIC + SlimSAS + cable + backplane often exceeds the PCIe spec loss budget. A ReDriver restores enough margin for stable operation.

Q3: Can I split x8 into 2×x4 with this card?
A: Yes, if the host BIOS supports bifurcation and your backplane routes groups correctly.

Q4: What’s the difference between a ReDriver and Retimer here?
A: A ReDriver equalizes and amplifies signals but does not perform clock recovery. A Retimer fully regenerates PCIe signals with protocol awareness but adds cost and latency.

Q5: Do I need to route PERST# and refclk through the SlimSAS?
A: Depends on the target backplane. NVMe drives often use SRIS, while switch shelves may need a common refclk and sidebands. Always check pin maps.

Q6: What should I check first if the link trains only at Gen3?
A: Validate cable quality, check BIOS bifurcation/clocking settings, then adjust ReDriver EQ presets. Shorter cables often reveal whether it’s a loss issue.

Q7: How do I confirm the link width and speed in Linux?
A: Use:

lspci -vv | grep -i "LnkSta"

Look for Speed 16GT/s, Width x8 to confirm Gen4 x8.

---

Conclusion

A PCIe Gen4 x8 to SlimSAS 8i Add-in-Card with a ReDriver is a practical solution for extending high-speed connectivity to external backplanes and devices while maintaining signal integrity and minimizing latency.

The key to success is careful PCB design, proper EQ tuning, and thorough validation across all cables and endpoints. By planning for clocking modes, sideband routing, and thermal management, you can deploy a robust solution that scales with enterprise workloads.

Part # PCI8-1736-SLM