Skip to content

ML-DSA and SLH-DSA explained

ML-DSA is the general-purpose post-quantum signature scheme, based on lattice assumptions and suitable for most signing. SLH-DSA is a hash-based alternative with larger, slower signatures whose security depends only on hash functions — a deliberate hedge against advances in lattice cryptanalysis.

ML-DSA, the default

Built on the same module-lattice foundations as ML-KEM, ML-DSA is the recommendation for most signing: certificates, protocol messages, authentication. Signatures are larger than ECDSA's but the operations are fast, and it is the scheme most implementations will reach for.

It has two modes. Hedged signing incorporates fresh randomness per signature; deterministic signing derives the required randomness from the message and the key. Both are specified.

SLH-DSA, the hedge

Hash-based signatures rest only on the properties of the hash function. That is a far more conservative assumption than lattice hardness, because hash functions have been studied longer and their failure modes are better understood.

The price is size and speed: signatures are substantially larger than ML-DSA's and signing is considerably slower. Verification is reasonable.

Choosing between them

  • High-volume signing — ML-DSA. Anything signing continuously cannot absorb SLH-DSA's cost.
  • Root and firmware signing — SLH-DSA is worth considering. Signatures are rare, long-lived, and the conservative assumption matters most where the key is hardest to rotate.
  • Constrained verifiers — check the verification cost against the device, not just the signing cost against the server.
  • Regulatory — some national schemes express preferences. Check the applicable one rather than assuming NIST's recommendation carries.

Why the randomness question is sharper for signatures

ECDSA fails catastrophically on nonce reuse: two signatures under one key with the same nonce expose the private key algebraically. That is not a theoretical property — it is how real wallet compromises happened.

ML-DSA in hedged mode consumes fresh randomness per signature, so a high-volume signing service consumes entropy continuously rather than only at key generation. That turns entropy from a provisioning concern into a throughput one, and it is why the budgeting arithmetic matters for signing estates.

Deterministic mode removes the per-signature dependency and introduces a different consideration: determinism can expose a signer to fault attacks on physically accessible hardware. Neither mode is universally correct.