Protocol Architecture

Silica Deep Dive: Speed, Performance & Reliability

A closer look at how Silica achieves fast, secure, and resilient finality through Tephra ingress, Tephra, Crystallite, Lithification, and Cali.

Silica Network Team • 2024-12-04 • 8 min read

⏱ Pre-Final Signals vs. Finalized Apply

Silica targets fast confirmation through two layers of finality:

  • Pre-final admission (100–300ms): As soon as a batch is referenced by enough validator activity in the DAG, wallets and explorers can show it as "Pre-Final." This gives the user an immediate confirmation signal while full finality is pending.
  • Finalized apply (0.8–1.5s): Once Tephra order commits and Crystallite applies the block locally, transactions are finalized permanently — no rollbacks, no forks. Lithification then packages proofs and checkpoints for monitoring, sync, and audit.
pub struct Lineation {
    ordered_batches: Vec<BatchId>,
    validator_signatures: Vec<MlDsaSignature>,
    finality_proof: FinalityProof,
    timestamp: u64,
}

Finality Source of Truth:

A block is final when it is committed by the DAG commit rule and finalized/applied locally. Certificates are proofs (reports) that enable downstream monitoring—they do not gate finality.

This blend of fast UX signals + deterministic local apply is what makes Silica designed to feel fast without sacrificing safety.

🧩 Layered Pipeline: Batch, Order, Apply, Prove

At the core, Silica uses a branded layered architecture:

  • Tephra ingress: Transactions are grouped into proposal payloads for efficient network fanout.
  • Tephra: Peers exchange missing events and record who saw what via self/other-parent links, then derive deterministic order.
  • Crystallite: Execution applies that order locally in deterministic lanes.
  • Lithification: Anchors and certificates package proofs for monitoring and recovery.
  • Cali: Recovery watchdogs observe the network and help restart cleanly when liveness degrades.
async fn create_batch(&mut self, txs: Vec<Transaction>) -> Result<MicroBatch>;
async fn gossip_batch_digest(&self, batch: &MicroBatch) -> Result<()>;

Ordering Source of Truth:

Tephra consensus order is the only ordering truth. Crystallite applies the committed result, Lithification emits proofs and checkpoints, and Cali watches health and recovery.

This targets high throughput with fast deterministic finality.

🧵 Parallelism & Sharding

Silica's execution engine is designed for scale:

  • 20–40k TPS per shard with parallel deterministic execution lanes.
  • High throughput across shards when scaled out — exact numbers pending benchmarks.
  • Access lists: Transactions declare touched keys, enabling conflict-free parallel execution.
  • Cross-shard transactions: Coordinated with lightweight two-phase receipts.
async fn execute_lane(&mut self, lane_id: LaneId, txs: Vec<Transaction>) -> Result<LaneResult>;

Result: predictable performance even at global scale.

🛡 Adversarial Validator Safety

Validator sets rotate every ~30 minutes using a randomness-driven lottery weighted by stake × reputation × Geographic Proof of Distribution (GPoD).

Byzantine thresholds:

  • <1/3 adversarial: Safety + liveness preserved
  • 1/3–2/3 adversarial: Safety preserved, liveness may stall until rotation
  • ≥2/3 adversarial: Safety breach possible (requires extreme adversarial control)

Mitigations:

  • Rolling churn: 10–25% per epoch
  • Reputation weighting: Uptime, quality, geo distribution
  • Monitoring / recovery watchdog: Slashing, emergency re-rolls
  • Emergency restart: At last clean lineation

🌍 Geographic Proof of Distribution (GPoD)

Silica enforces geographic fairness with GPoD:

  • H3 grids: Quotas at state level, verification at city level
  • Probes: Each epoch, nodes run VRF-sampled near/far latency checks
  • Rewards: Boosts in under-provisioned regions, linger bonuses for early movers
  • Anti-gaming: Warmups, cooldowns, rep velocity throttles prevent hop-farming

placement_c = clamp(1 + α*(1 - n_c/target_c) - β*max(0, n_c/target_c - 1), 0.5, 3.0)

This ensures no region can dominate and rewards follow global distribution.

🔒 Post-Quantum Security

Silica is one of the few blockchains designed for a post-quantum world:

  • ML-DSA signatures (FIPS 204, release requirement) secure validator certificates
  • PQ VRFs (future-ready): Today, VRFs run on EC curves with optional PQ attestation; roadmap includes native lattice-based VRFs

This means Silica finality remains safe even against next-gen computing attacks.

🔁 Reliability Mechanisms

Silica adds redundancy at multiple layers:

  • Quorum-seen pre-final signals → UX feels instant
  • Rolling churn + overlapping validator sets → reduce adversarial lottery risk
  • Automated restart protocol: Detects stalls, feeds evidence to Cali judgement, and resumes finality (slashing is enforced by Cali via staking/bank, not recovery)

Automated Restart Protocol:

  1. Detection: No lineation >2× expected interval
  2. Challenge: Evidence of missed signatures collected
  3. Action: Slashing + reseed + partial validator re-roll
  4. Resume: Backlog finalized cleanly

📊 Performance Targets

Finality: 0.8–1.5s per lineation
Pre-final admission: 100–300ms
Throughput: Target: high throughput across shards
Validator set size: 64–96
Rotation: Every ~30m with rolling churn

✅ Takeaway

Silica targets fast UX, deterministic finality, and global fairness through:

  • Tephra ingress + Tephra + Crystallite + Lithification + Cali.
  • Parallel sharding with access lists.
  • GPoD-driven geographic balance.
  • Cali-driven liveness guarantees.
  • Post-quantum cryptography for future safety.

This is what makes Silica not just fast — but reliably fast.