Falcon-Rust

Unofficial rust implementation of the Falcon post-quantum digital signature scheme.

Falcon was submitted to the NIST PQC standardization project and was selected for standardization. The final standard is still outstanding. We do anticipate slight changes between the standard and the submission, and these changes might break compatibility.

Falcon comes in two variants. Falcon512 claims at least 108 bits of security, and Falcon1024 claims at least 252 bits of security, both against quantum computers.

This implementation adheres to the specification. It was originally written following the the official python implementation, but has since deviated.

Example

use rand::rng;
use falcon_rust::falcon512;

let mut rng = rng();
let mut msg : [u8; 5] = rng.random();
let (sk, pk) = falcon512::keygen(rng.random());
let sig = falcon512::sign(&msg, &sk);
assert!(falcon512::verify(&msg, &sig, &pk));

Performance

If you are after performance, you are probably better off with one of the implementations by the inventors, either the foreign function interface (FFI) into the optimized C code (pqcrypto-falcon), or the optimized rust crate (fn-dsa). The following benchmark was produced by my 12th Gen Intel(R) Core(TM) i9-12900K (which supports AVX2). You can make your own by running cargo bench.

	Keygen	Sign	Verify
falcon-rust 512	226.84 ms	329.26 µs	20.277 µs
falcon-rust 1024	1.3106 s	665.00 µs	41.933 µs
C FFI 512	4.0688 ms	129.49 µs	25.555 µs
C FFI 1024	12.048 ms	255.59 µs	50.142 µs
FN DSA 512	2.0587 ms	143.86 µs	10.486 µs
FN DSA 1024	9.0216 ms	278.71 µs	20.793 µs

Features

• key generation
• signature generation
• signature verification
• derandomized algorithms
• (de)serialization
• better algorithms (e.g. RNS)
• uncompressed signature format
• signed-message interface
• hardware optimizations
• message-recovery mode
• constant-time (?)

To-do's

• NIST KATs
• make LdlTree straightforward
• optimize representation of secret key, signature, public key
• test interoperability against the reference implementation
• negative tests
• profile, and fix bottlenecks
• Montgomery representation for field elements
• Residue number system (RNS) for big integer arithmetic
• streaming (de)serialization
• investigate secret-dependent time variability

Contributing

Contributions are welcome! If accepted, contributions will be released under the same license.