Why existing FHE cannot run on Solana
Lattice-based FHE schemes — the entire current generation, including Zama's TFHE, Fhenix's Bonsai, Inco's evm-tfhe — share a common architectural constraint: every encrypted operation introduces noise into the ciphertext. After a small number of operations, the noise reaches a threshold beyond which decryption fails. To continue computing, the ciphertext must be bootstrapped — homomorphically decrypted and re-encrypted, refreshing the noise. Bootstrapping is computationally expensive. The fastest production implementations measure bootstrapping in tens of milliseconds per gate, with realistic dApp workloads requiring hundreds of bootstraps per transaction.
A 400ms Solana block cannot accommodate this. The math forces compromise: either the chain slows to FHE's pace, or the FHE workload is offloaded entirely off-chain with trust assumptions that defeat the purpose.
LUT-FHE: bootstrap-free encrypted compute
AURA's scheme replaces ring-based ciphertext arithmetic with precomputed lookup tables defined over the multivariate quadratic structure. Rather than refreshing noise after each operation, the scheme is structurally noise-free — the algebraic structure does not accumulate error. This eliminates bootstrapping entirely.
The result, on standardized benchmarks, is approximately 1000x speedup over TFHE-based alternatives across typical arithmetic workloads. Specific operations show greater asymptotic speedups; integer addition, for example, executes at roughly 0.04 microseconds versus Zama's ~50ms gate evaluation. Independent benchmark verification is in progress and will be published before the Series A close.
Security reduces to the Multivariate Quadratic (MQ) problem, proven NP-hard by Garey and Johnson (1979). Unlike lattice-based schemes whose hardness rests on conjectured assumptions about Learning With Errors (LWE), MQ-hardness has a stronger theoretical foundation and no known efficient quantum attack. AURA is post-quantum secure not as a future upgrade path but as a structural property.
The coprocessor architecture
AURA's runtime executes off the host chain. A Solana program built with the AURA SDK does not perform FHE computation directly — it submits an encrypted compute request to the AURA coprocessor network, receives a verifiable ciphertext result, and proceeds with on-chain settlement. This separation is deliberate and load-bearing for the entire design.
It preserves chain throughput. Solana validators do not spend block time on FHE computation. They process transactions at native speed and verify ciphertext outputs as they would verify any other input.
It makes the architecture chain-agnostic. The coprocessor network is a separate compute substrate. Any chain can integrate the AURA SDK and route encrypted compute requests to the coprocessor.
It creates the economic substrate for the mining protocol. The coprocessor network is operated by independent miners running FHE compute against encrypted requests. Miners earn AURA tokens in proportion to verified work delivered.
It enables verifiability without trust. Each coprocessor result is accompanied by a cryptographic correctness proof. Validators on the host chain verify the proof, not the computation. Trust is replaced by mathematics.