SLH-DSA — formerly known as SPHINCS+, standardised by NIST as FIPS 205 — is the conservative signature scheme in the post-quantum portfolio. Conservative meaning: its security depends only on the properties of a hash function. No lattices. No discrete logs. No structured number theory that has decades of cryptanalytic history but still produces uncomfortable surprises. If a hash function holds, SLH-DSA holds.
That security simplicity is paid for in size. An SLH-DSA-128s signature is 7,856 bytes, an SLH-DSA-128f signature 17,088, and an SLH-DSA-256f signature 49,856 — roughly 2 to 15 times the 3,309 bytes of ML-DSA-65, against 64 bytes for ECDSA P-256. That is the trade. Almost every decision about when to use SLH-DSA flows from one question: is the bandwidth tolerable here?
Where the trade is right
Three categories where SLH-DSA earns its cost cleanly:
1. Firmware signing for long-deployed hardware
A POS terminal deployed at an Indian merchant site is going to run the same firmware for 5–10 years before it gets refreshed. The signature on that firmware has to remain unbreakable for that whole period. If lattice-based signatures were ever cryptanalytically weakened, a firmware-signing root using ML-DSA would have to be rotated, which means re-issuing every signed firmware image in deployment. With SLH-DSA, the trust anchor's security tracks the hash function — and SHA-2 / SHA-3 have a much longer cryptanalytic track record than any lattice scheme.
The extra signature bytes don't matter here. Firmware images are typically megabytes; a 17 KB signature is rounding error.
2. Code signing for slow-moving software
The same logic applies to operating system update images, driver packages, and platform-level software bundles. The relying parties verifying these signatures are doing so as a step in a download that's already in the tens-to-hundreds of megabytes. The signature size is invisible. The security assumption simplification is operationally valuable.
3. Roots of code-signing, firmware and offline hierarchies
A 20-year root certificate signed with ML-DSA is a 20-year bet that lattice cryptanalysis won't advance to break ML-DSA. Reasonable bet, but a non-trivial one. A 20-year root signed with SLH-DSA is a 20-year bet that the underlying hash function will continue to resist attack. That bet has much more historical evidence behind it.
Where that root sits matters. In a TLS-serving hierarchy the root's signature is not read rarely: it sits on the intermediate certificate, and the intermediate is sent in every handshake. An SLH-DSA root there adds its full signature to every connection. So SLH-DSA roots suit code-signing, firmware and offline hierarchies, where the chain travels with an artefact or is verified occasionally. For TLS-serving hierarchies, prefer ML-DSA-87 or ML-DSA-65 at the root — or keep an SLH-DSA root only for chains that are not sent on the wire (see the PKI rotation piece).
Where the trade is wrong
Three categories where ML-DSA (or FN-DSA) is the right default, not SLH-DSA:
- TLS leaf certificates. Served on every handshake, multiple per page-load on a complex site. A 17 KB signature in this position would put 30+ KB of certificate chain on every TLS connection setup. Use ML-DSA-44, or FN-DSA-512 once FIPS 206 is published.
- JWT and short-lived bearer tokens. Small payloads (typically a few hundred bytes) wrapping a signature. A 17 KB signature is 30× the payload. ML-DSA, or FN-DSA if signature size really matters.
- mTLS client certificates. Sent on every connection, often from constrained clients. Same argument.
SLH-DSA parameter sets, briefly
FIPS 205 defines twelve parameter sets — six size/speed profiles, each available in SHA2 and SHAKE variants. The first dimension is security level (NIST levels 1, 3, 5 corresponding to roughly AES-128, AES-192, AES-256 against classical attack). The second is the "s" / "f" choice: "s" for small signatures and slow signing (with faster verification); "f" for fast signing at the cost of larger signatures and slower verification.
For BFSI use:
- SLH-DSA-128s: the default for code and firmware signing. Releases are signed infrequently, so the slower signing is affordable, and the signature is less than half the size of the "f" variant.
- SLH-DSA-128f: only for high-volume signing pipelines where signing latency genuinely matters.
- SLH-DSA-256f: NIST level 5 for code-signing, firmware and offline roots where the conservative ceiling matters most.
What to migrate to SLH-DSA, in order
For an Indian bank running through the PQC programme, the SLH-DSA targets in priority order:
- Code-signing and firmware roots. Once. Long validity. Cross-sign with the legacy root for trust continuity. (For the root of a TLS-serving internal CA, use ML-DSA-87 or ML-DSA-65 instead — its chain is sent in every handshake.)
- Code-signing intermediate. Sign your internal software releases under an SLH-DSA-128s intermediate.
- Firmware-signing intermediate. POS, ATM, IoT firmware chains.
- Document-signing high-assurance tier. Only if your regulatory category requires hash-based conservatism — most don't.
Everything else stays on ML-DSA (or FN-DSA where signature size dominates). The pattern, across the migration: SLH-DSA at the top of the hierarchy where it signs rarely and lives long; ML-DSA / FN-DSA at the bottom where it signs constantly and lives briefly.