PCIe Encoding Evolution: Why 20% Overhead Was Too High to Handle

PCIe didn’t scale to Gen 5 and Gen 6 speeds by simply increasing frequency.

A major breakthrough — often called the “Gen 3 Unlock” — came from fundamentally rethinking how data is encoded.

By moving from 8b/10b to 128b/130b encoding, PCIe reduced encoding overhead from 20% to ~1.5%.

This efficiency gain, however, introduced new design challenges.

🧱 The 8b/10b “Comfort Zone” (Gen 1 & Gen 2)

In early PCIe generations, 8b/10b encoding acted as a safety net by guaranteeing:

1. DC Balance
2. Achieved using Running Disparity to prevent voltage drift
3. Transition Density
4. Ensured frequent 0→1 and 1→0 transitions so the Clock Data Recovery (CDR) remained locked
5. K-Codes
6. Special control symbols (COM, SKP, FTS, etc.) that were easy to detect in the data stream

This simplicity made PHY design easier but came at a significant bandwidth cost.

🔧 Gen 3+ Trade-off: Efficiency vs. Complexity

With 128b/130b encoding introduced at 8.0 GT/s, many of the built-in guarantees of 8b/10b disappeared. Modern PCIe PHYs compensate through the following mechanisms:

1. Scrambling Takes Center Stage
2. A robust LFSR (Linear Feedback Shift Register) randomizes data
3. Prevents long runs of zeros or ones
4. Maintains sufficient transition density for CDR operation
5. Sync Headers Replace K-Codes
6. A 2-bit Sync Header identifies block type:
7. 10b → Data
8. 01b → Ordered Sets
9. This is the only unscrambled portion of the stream
10. Managing DC Wander
11. Loss of inherent DC balance is handled via tighter AC-coupling capacitor specifications
12. Typical values moved toward the 176–265 nF range

📦 The Two-Level Framing Model in Modern PCIe

PCIe Gen 3/4/5 uses a two-layer framing architecture:

1️⃣ Block Level

1. Fixed 130-bit blocks
2. Sync Header tells the receiver exactly what type of block it is processing

2️⃣ Stream Level

1. Uses Framing Tokens instead of single K-characters:
2. STP – Start of TLP
3. SDP – Start of DLLP
4. END – End of packet
5. Defined using specific bit patterns within the stream

💡 Key Takeaway

128b/130b encoding is a classic engineering trade-off:

1. ❌ Simpler hardware logic was sacrificed
2. ✅ Massive bandwidth efficiency was gained

Without this transition, scaling to:

1. 32 GT/s (PCIe Gen 5)
2. 64 GT/s (PCIe Gen 6)

would have been physically impossible due to power consumption and signal-integrity limits.

Designers:

Do you miss the simplicity of 8b/10b K-codes, or has block-based encoding made your life easier in the long run?

By Krishnan B

Technology Specialist

Embedded