FEC (Forward Error Correction) Complete Guide

1. What is FEC?

What is Forward Error Correction?

FEC (Forward Error Correction) is an error correction technique that adds redundant data (parity) to transmitted information, enabling the receiver to detect and correct errors without retransmission. PCIe 6.0/7.0 Flit Mode uses Reed-Solomon FEC.

Why FEC in PCIe 6.0/7.0?

PAM4 Challenges: PAM4 has ~9.5 dB SNR penalty vs NRZ
Higher BER: Pre-FEC BER can be 10⁻⁴ (vs 10⁻¹² target)
Latency: Correction is faster than retransmission
Throughput: Avoids retry bandwidth overhead

2. Reed-Solomon FEC Overview

Reed-Solomon Characteristics

Type: Block code (operates on fixed-size blocks)
Symbol-based: Works on multi-bit symbols (bytes)
Burst correction: Good for correcting burst errors
Widely used: CDs, DVDs, QR codes, storage

RS Code Notation

Reed-Solomon RS(n, k)


n = total symbols (data + parity)
k = data symbols
n - k = parity symbols
t = (n - k) / 2 = correctable symbol errors

PCIe FEC uses RS(150, 148):
  n = 150 bytes total
  k = 148 bytes data
  n - k = 2 bytes parity
  t = 1 byte correction capability per codeword

3. PCIe FEC Structure

3-Way Interleaved FEC

PCIe uses three interleaved RS codewords per Flit:

    256-Byte Flit Structure with FEC
    
    ┌──────────────────────────────────────────────────────────────┐
    │                        256 Bytes                              │
    ├──────────────────────────────┬─────────────┬─────────────────┤
    │      TLP Data (236 bytes)    │ DLP (6B)    │   CRC (8B)      │
    │                              │             │   FEC ECC (6B)  │
    └──────────────────────────────┴─────────────┴─────────────────┘
    
    FEC Coverage (250 bytes → 3 codewords):
    
    Codeword 0: Bytes 0, 3, 6, 9, ... (every 3rd byte starting at 0)
    Codeword 1: Bytes 1, 4, 7, 10, ... (every 3rd byte starting at 1)
    Codeword 2: Bytes 2, 5, 8, 11, ... (every 3rd byte starting at 2)
    
    Each codeword: ~83 data bytes + 2 parity bytes = 85 bytes
    Total parity: 6 bytes (2 per codeword)

Interleaving Benefit

Spreads burst errors across multiple codewords
3-byte burst error becomes 1 byte in each codeword
Each codeword can correct its 1 byte error
Result: Can correct 3 consecutive byte errors

4. FEC Encoding

Encoding Process

Assemble Flit (TLP data, DLP, CRC)
Interleave bytes into 3 groups
Calculate RS parity for each group
Append 2-byte parity to each codeword
De-interleave to form final Flit with ECC
Transmit Flit

Interleaving Pattern

    Input bytes: B0, B1, B2, B3, B4, B5, B6, B7, B8, ...
    
    Codeword 0: B0, B3, B6, B9, ... + P0a, P0b
    Codeword 1: B1, B4, B7, B10, ... + P1a, P1b
    Codeword 2: B2, B5, B8, B11, ... + P2a, P2b
    
    Output (interleaved parity):
    ... B246, B247, B248, B249, P0a, P1a, P2a, P0b, P1b, P2b
                               └─────── 6 ECC bytes ───────┘

5. FEC Decoding

Decoding Process

Receive Flit
Separate bytes into 3 interleaved groups
For each codeword:
- Calculate syndrome
- If syndrome ≠ 0, error detected
- Locate error position
- Calculate error value
- Correct byte
De-interleave corrected data
Verify CRC (additional check)

Syndrome Calculation

Syndrome indicates presence and location of errors:

S = 0: No error (or undetectable)
S ≠ 0: Error detected, use for correction

6. Error Correction Capability

Single Codeword Capability

Correct: 1 byte error per codeword
Detect: 2 byte errors per codeword

Flit-Level Capability (with interleaving)

Error Pattern	Result
1 byte error anywhere	Corrected
2 byte burst error	Corrected
3 byte burst error	Corrected
3 scattered errors (1 per codeword)	Corrected
4+ byte burst error	Detected (CRC fails)
2 errors in same codeword	Detected (not corrected)

BER Improvement

Effective BER Improvement


Pre-FEC BER: ~10⁻⁴ to 10⁻⁵ (typical for PAM4)
Post-FEC BER: < 10⁻¹² (specification target)

Coding Gain: ~7-8 orders of magnitude improvement

7. FEC Latency

Latency Components

Encoding latency: Parity calculation
Decoding latency: Syndrome + correction
Total: ~10-20 ns typical (implementation dependent)

Latency vs Retry

Approach	Latency	Bandwidth Impact
FEC Correction	~20 ns	None (inline)
Retry (ACK/NAK)	~1-2 μs	Replay buffer overhead

FEC + CRC + Retry

FEC corrects most errors. CRC detects uncorrectable errors. Retry handles uncorrectable cases via Flit retransmission. This layered approach achieves high reliability with low overhead.

FEC (Forward Error Correction)