In August 2024, NIST finalized the first three outputs of its Post-Quantum Cryptography (PQC) Standardization project, running since 2016: FIPS 203 (ML-KEM, key establishment), FIPS 204 (ML-DSA, digital signatures) and FIPS 205 (SLH-DSA, backup digital signatures). All three protect against a future large-scale quantum computer capable of breaking the RSA and elliptic-curve cryptography that secures most of today's TLS, firmware signing and secure-boot chains. NIST's own framing of urgency is blunt: these standards "can and should be put into use now," and a follow-on document, NIST IR 8547, has since attached hard 2030/2035 dates to the migration away from the algorithms ML-KEM, ML-DSA and SLH-DSA are meant to replace.
Key takeaways
- FIPS 203 — ML-KEM (Table 2/3): three parameter sets, named for their matrix dimension k, in order of increasing security strength and decreasing performance.
- The open-source pqm4 benchmarks measure the standards on an Arm Cortex-M4 (NUCLEO-L4R5ZI by default).
- IR 8547 notes that SP 800-57 Part 1 had projected disallowing 112-bit public-key schemes on 1 January 2031; NIST now deprecates them after 2030 instead.
- Three points matter to vendors.
What does each of the three standards do?
| Standard | Algorithm | Function | Based on | Published |
|---|---|---|---|---|
| FIPS 203 | ML-KEM | Key-encapsulation mechanism (establishing a shared secret key) | CRYSTALS-Kyber | 13 Aug 2024 |
| FIPS 204 | ML-DSA | Digital signatures | CRYSTALS-Dilithium | 13 Aug 2024 |
| FIPS 205 | SLH-DSA | Digital signatures (structurally independent backup) | SPHINCS+ | 13 Aug 2024 |
Per FIPS 203's own Abstract, a key-encapsulation mechanism (KEM) is "a set of algorithms that, under certain conditions, can be used by two parties to establish a shared secret key over a public channel," with that shared key then used by conventional symmetric-key cryptography for encryption and authentication. ML-KEM's security rests on the Module Learning With Errors (MLWE) problem, which is "believed to be secure, even against adversaries who possess a quantum computer." FIPS 204 and FIPS 205 both address digital signatures but deliberately rest on unrelated hardness assumptions — ML-DSA on structured lattice problems like ML-KEM, SLH-DSA purely on the security of its underlying hash functions — so a cryptanalytic advance against lattice-based math would not simultaneously compromise both NIST-approved signature options. All three are FIPS publications issued by NIST's Information Technology Laboratory with no individually credited authors; each is approved by the Secretary of Commerce and, per its own Applicability clause, available for voluntary adoption by private and commercial organizations even though FIPS formally governs U.S. federal information systems. NIST's PQC project page lists two more algorithms in standardization: Falcon (as FN-DSA) and HQC, a second KEM selected on 11 March 2025.
What parameter sets, key sizes and signature sizes do they define?
FIPS 203 — ML-KEM (Table 2/3): three parameter sets, named for their matrix dimension k, in order of increasing security strength and decreasing performance.
| Parameter set | Security category | Encapsulation key | Decapsulation key | Ciphertext |
|---|---|---|---|---|
| ML-KEM-512 | 1 | 800 bytes | 1,632 bytes | 768 bytes |
| ML-KEM-768 (NIST's recommended default) | 3 | 1,184 bytes | 2,400 bytes | 1,088 bytes |
| ML-KEM-1024 | 5 | 1,568 bytes | 3,168 bytes | 1,568 bytes |
NIST's own guidance in Section 8 of FIPS 203 is explicit: "NIST recommends using ML-KEM-768 as the default parameter set, as it provides a large security margin at a reasonable performance cost." Other sets may be used "in cases where this is impractical or even higher security is required." All three produce a 32-byte shared secret.
FIPS 204 — ML-DSA (Table 1/2): three parameter sets, named ML-DSA-kl for the dimensions of the matrix they use.
| Parameter set | Security category | Private key | Public key | Signature |
|---|---|---|---|---|
| ML-DSA-44 | 2 | 2,560 bytes | 1,312 bytes | 2,420 bytes |
| ML-DSA-65 | 3 | 4,032 bytes | 1,952 bytes | 3,309 bytes |
| ML-DSA-87 | 5 | 4,896 bytes | 2,592 bytes | 4,627 bytes |
NIST IR 8547 adds one caveat for device makers: ML-DSA-44 drops to category 1 if it runs with an approved random bit generator of less than 192 bits of security.
FIPS 205 — SLH-DSA (Table 2): twelve parameter sets — six security-parameter/variant combinations, each available in a SHA2-based or SHAKE-based hash instantiation. The "s" suffix optimizes for a smaller signature; the "f" suffix optimizes for faster signing at the cost of a larger signature. Public keys are small and constant per security level (32, 48 or 64 bytes; private keys are double that), but signatures are the largest of any of the three standards:
| Parameter set family | Security category | Signature size |
|---|---|---|
| SLH-DSA-*-128s / 128f | 1 | 7,856 bytes / 17,088 bytes |
| SLH-DSA-*-192s / 192f | 3 | 16,224 bytes / 35,664 bytes |
| SLH-DSA-*-256s / 256f | 5 | 29,792 bytes / 49,856 bytes |
FIPS 205 §11 flags one nuance worth knowing before picking a variant: all twelve parameter sets meet their claimed EUF-CMA security category, "except SLH-DSA-SHA2-128f and SLH-DSA-SHAKE-128f" for the specific property of message-bound signatures (resistance to one key/signature pair validating two different messages) — applications that need that property should either account for it when choosing a parameter set or apply NIST's referenced BUFF transformation.
Why do these sizes matter for firmware and constrained devices?
Every one of these algorithms produces keys, ciphertexts or signatures that are larger than the RSA/ECC equivalents they replace — an ML-KEM-768 ciphertext (1,088 bytes) dwarfs a typical ECDH exchange, and even the smallest SLH-DSA signature (7,856 bytes) is an order of magnitude larger than a 64-128 byte ECDSA signature. For firmware images, secure-boot chains, TLS handshakes over constrained links, and any protocol with fixed-size packet or header budgets, this is not a drop-in swap: image/partition layouts, bootloader buffer sizes, handshake budgets and OTA-update payload sizes may all need to be re-evaluated before a PQC migration can ship. The secure boot guide covers where those signatures and keys sit in the boot chain.
How much CPU, RAM and flash do they need on a microcontroller?
The open-source pqm4 benchmarks measure the standards on an Arm Cortex-M4 (NUCLEO-L4R5ZI by default). Selected results, averaged over the project's runs:
| Algorithm (pqm4 implementation) | Operation | Cycles | Stack | Code size |
|---|---|---|---|---|
| ML-KEM-768 (m4fstack) | Encapsulate / decapsulate | 664,654 / 714,194 | 2,860 / 2,844 bytes | 13,320 bytes |
| ML-KEM-768 (m4fspeed) | Encapsulate / decapsulate | 658,754 / 707,827 | 6,468 / 6,452 bytes | 16,016 bytes |
| ML-DSA-65 (m4f) | Sign (average) / verify | 6,193,171 / 2,415,944 | 68,872 / 9,888 bytes | 19,328 bytes |
| ML-DSA-65 (m4fstack) | Sign (average) / verify | 24,421,526 / 5,732,397 | 6,616 / 2,712 bytes | 24,120 bytes |
| SPHINCS+-SHA2-128s-simple (clean) | Sign / verify | 7,657,558,168 / 7,471,794 | 2,556 / 1,968 bytes | 5,264 bytes |
Three things stand out. ML-KEM fits comfortably on a mid-range microcontroller. ML-DSA signing time varies widely between runs (the m4f implementation ranged from about 2.9 to 26 million cycles), so a device that signs needs a worst-case budget, not an average. And the SPHINCS+ design SLH-DSA comes from verifies at a reasonable cost but signs in billions of cycles, which suits a build server, not a device.
What are the final NIST IR 8547 deadlines?
NIST IR 8547, "Transition to Post-Quantum Cryptography Standards," reached final publication in October 2026 (authors Dustin Moody, Ray Perlner, Andrew Regenscheid, Angela Robinson and David Cooper of NIST's Computer Security Division). The final text is simpler than the 2024 draft: every quantum-vulnerable public-key scheme gets the same schedule, whatever its security strength.
| Algorithm family (IR 8547 Tables 2 and 4) | Parameters | Transition |
|---|---|---|
| ECDSA, RSA signatures | ≥112 bits of classical security | Deprecated after 2030, disallowed after 2035 |
| EdDSA | ≥128 bits | Deprecated after 2030, disallowed after 2035 |
| Finite-field and elliptic-curve DH and MQV | ≥112 bits | Deprecated after 2030, disallowed after 2035 |
| RSA key establishment | ≥112 bits | Deprecated after 2030, disallowed after 2035 |
IR 8547 notes that SP 800-57 Part 1 had projected disallowing 112-bit public-key schemes on 1 January 2031; NIST now deprecates them after 2030 instead. It ties the 2035 end date to National Security Memorandum 10. These dates bind NIST standards and U.S. federal use, but they reach vendors through FIPS 140-3 validation and buyers who cite NIST. IR 8547 also confirms that a hybrid key-establishment combination can be FIPS 140-validated as long as at least one input comes from an approved scheme. On code signing it is direct: where signature-verification code cannot be updated after manufacture, devices should be designed to require quantum-resistant signatures if they may still be in use once a cryptographically relevant quantum computer exists.
What do EO 14412 and OMB M-26-15 add?
Executive Order 14412, "Securing the Nation Against Advanced Cryptographic Attacks" (22 June 2026), and OMB Memorandum M-26-15, "Execution of the Migration to Post-Quantum Cryptography" (24 June 2026), set earlier dates for federal high-impact systems and High Value Assets. Agencies had 120 days to submit a PQC migration plan, structured in phases:
| M-26-15 phase | Years | Focus |
|---|---|---|
| 1 Strategy, planning, discovery | 2026–2027 | Cryptographic inventory, risk assessment |
| 2 Pilots and early migration | 2027–2028 | Pilots, first prioritized systems |
| 3 Prioritized migration | 2028–2030 | PQC key establishment on prioritized systems (by 31 Dec 2030) |
| 4 Signature migration | 2031 | PQC signatures on prioritized systems (by 31 Dec 2031) |
| 5 Full migration | 2035 | Remaining systems |
Three points matter to vendors. Agencies "should ensure that any requirements for products used by the agency" in CISA's PQC product categories "include PQC integration". Systems "incapable of supporting PQC or hybrid cryptography" are to be prioritized for replacement. And agencies must support TLS 1.3 by 2 January 2030; the memo's hybrid key-exchange example pairs x25519 with ML-KEM-768.
Which algorithm should sign firmware?
| Option | Signature / public key | Fits when | Watch out for |
|---|---|---|---|
| ML-DSA-65 | 3,309 / 1,952 bytes | General firmware and update signing, fast verification | Public key needs room in ROM or OTP |
| SLH-DSA-128s | 7,856 / 32 bytes | Long-lived root key, hedge against lattice breaks | Slow signing; large signatures |
| LMS / XMSS (SP 800-208) | Depends on parameters | Firmware signing with a well-controlled signer | Signer must never reuse a one-time key; state tracking |
IR 8547 says stateful hash-based schemes "are not intended for general use but may be useful for applications like software and firmware signing," because each one-time key must never sign twice.
Worked example: budgeting post-quantum secure boot
Synthetic example, for illustration only. A sensor maker ships a Cortex-M4 at an assumed 80 MHz with a 2 MB image, an ECDSA P-256 boot check and a 64-byte OTP slot for the root key hash. Products stay in the field 15 years, so units sold in 2027 run past 2035.
Pick the verifierML-DSA-65 m4f verify: 2,415,944 cycles, about 30 ms at 80 MHz, with 9,888 bytes of stack. The m4fstack variant takes about 72 ms but needs only 2,712 bytes. Fit the keyA 1,952-byte public key does not fit in 64 bytes of OTP. Store its SHA-256 hash (32 bytes) in OTP and ship the key in the image header. Fit the signatureThe image header grows from 64 to 3,309 signature bytes; the OTA update format and bootloader buffer need the same change. Keep a hedgeAdd SLH-DSA-SHA2-128s as a second root (32-byte key; about 93 ms to verify at 80 MHz, using pqm4's SPHINCS+-SHA2-128s figure) so a lattice break does not brick the update path. Ship dual-signedDuring the 2030–2035 window, sign with both ECDSA and ML-DSA so older bootloaders keep working.
Where should a device team start?
Inventory the cryptography you ship, per product and release; a CBOM records algorithms and keys, not only library versions. Default to ML-KEM-768, which FIPS 203 names as the default, and pick an ML-DSA set by measured cost; category 3 (ML-DSA-65) matches ML-KEM-768. Fix the boot chain firstCode that verifies signatures in ROM cannot be patched later, which is IR 8547's own code-signing point. Move TLS to 1.3 with hybrid key exchangewhere your stack supports it; federal buyers need TLS 1.3 by 2 January 2030. Treat 2030 as the product deadlineDeprecation starts then, and federal high-impact systems need PQC key establishment by 31 December 2030. Replace the weak algorithms already in the field as well; see fixing weak cryptography in embedded firmware.
Can AI flag quantum-vulnerable crypto in a firmware image?
Partly, through the component list. SoftWSP's firmware pipeline extracts the software components in an image into a CycloneDX 1.6 SBOM, including cryptographic libraries such as OpenSSL, mbed TLS or wolfSSL by name and version, and matches them against NVD, the CISA KEV catalog, FIRST EPSS and CISA CSAF ICS advisories. AI drafts a plain-language report that cites the source for each finding, for an engineer to review. The pipeline does not detect which algorithms or key sizes the firmware actually uses, does not label findings "quantum-vulnerable", does not write a CBOM and does not choose parameter sets; that analysis stays with the engineer. The free SBOM viewer shows which crypto libraries a CycloneDX file lists. The firmware security testing methodology guide covers the manual review steps.
A post-quantum plan starts with knowing which cryptographic libraries and versions ship in each firmware release, and which of them already carry known exploited vulnerabilities. SoftWSP produces that SBOM and the cited vulnerability evidence for engineer review: see Device & Firmware Security.
This guide summarizes NIST's published FIPS 203, 204 and 205 texts, NIST's August 2024 announcement, NIST IR 8547's final (October 2026) transition guidance, OMB M-26-15 and the pqm4 benchmark results; it is not a substitute for the full standards or for cryptographic implementation review.
Put this guide to work
Frequently asked questions
What are FIPS 203, 204 and 205, and when were they published?
Three Federal Information Processing Standards NIST published on 13 August 2024, the first finalized outputs of NIST's Post-Quantum Cryptography Standardization project (running since 2016). FIPS 203, "Module-Lattice-Based Key-Encapsulation Mechanism Standard," specifies ML-KEM (derived from CRYSTALS-Kyber) for establishing a shared secret key, stating ML-KEM "is believed to be secure, even against adversaries who possess a quantum computer." FIPS 204, "Module-Lattice-Based Digital Signature Standard," specifies ML-DSA (derived from CRYSTALS-Dilithium) for digital signatures. FIPS 205, "Stateless Hash-Based Digital Signature Standard," specifies SLH-DSA (based on SPHINCS+) as a structurally different backup signature scheme. All three are published by NIST's Information Technology Laboratory; as FIPS publications they carry no named individual authors, are approved by the Secretary of Commerce, and — per each standard's own Applicability clause — their "adoption and use ... are available to private and commercial organizations" even though FIPS formally applies to U.S. federal systems.
What parameter sets does each standard define, and how large are the keys/signatures?
Each standard defines three (ML-KEM, ML-DSA) or twelve (SLH-DSA) parameter sets spanning NIST security categories 1, 3 and 5. Per FIPS 203 Table 3: ML-KEM-512 uses an 800-byte encapsulation key, 1,632-byte decapsulation key and 768-byte ciphertext (category 1); ML-KEM-768 uses 1,184 / 2,400 / 1,088 bytes (category 3, and NIST's own recommended default); ML-KEM-1024 uses 1,568 / 3,168 / 1,568 bytes (category 5). Per FIPS 204 Table 2: ML-DSA-44 uses a 1,312-byte public key, 2,560-byte private key and 2,420-byte signature (category 2); ML-DSA-65 uses 1,952 / 4,032 / 3,309 bytes (category 3); ML-DSA-87 uses 2,592 / 4,896 / 4,627 bytes (category 5). Per FIPS 205 Table 2, SLH-DSA's twelve parameter sets (SHA2- or SHAKE-based, each in a small-signature "s" or fast-signing "f" variant) have much smaller keys (32-64 bytes) but far larger signatures: 128-bit variants run 7,856 bytes (128s) or 17,088 bytes (128f); 192-bit variants run 16,224 or 35,664 bytes; 256-bit variants run 29,792 or 49,856 bytes.
Why does NIST need three different standards instead of one?
Because they serve different functions and rest on different mathematical hardness assumptions, which is itself a risk-management choice. FIPS 203 (ML-KEM) solves key establishment; FIPS 204 (ML-DSA) and FIPS 205 (SLH-DSA) both solve digital signatures but are intentionally built on unrelated math: ML-DSA, like ML-KEM, relies on structured (module) lattice problems, while SLH-DSA relies only on the security of underlying hash functions. NIST's own August 2024 announcement frames FIPS 205 explicitly as a hedge — "an alternative to ML-DSA ... based on a different math approach ... in the event that ML-DSA turns out to be vulnerable" — so a future cryptanalytic break of lattice-based assumptions would not take down every NIST-approved post-quantum signature scheme at once. NIST computer scientist Dustin Moody was quoted encouraging early adoption: "We encourage system administrators to start integrating them into their systems immediately, because full integration will take time."
What are the migration deadlines, and do they apply outside the U.S. federal government?
NIST IR 8547, "Transition to Post-Quantum Cryptography Standards" (final, October 2026; authors Dustin Moody, Ray Perlner, Andrew Regenscheid, Angela Robinson and David Cooper, NIST Computer Security Division), sets concrete dates for today's quantum-vulnerable algorithms: RSA, (EC)DSA/EdDSA, and finite-field/elliptic-curve Diffie-Hellman and MQV at the 112-bit security level are **deprecated after 2030 and disallowed after 2035** under NIST standards and guidance, and the final text applies the same schedule to schemes at ≥128-bit security (the 2024 draft had let those skip deprecation). IR 8547 ties this to National Security Memorandum 10, which set 2035 as the government-wide target, and notes OMB direction for agencies' most sensitive systems to adopt PQC for key establishment by 31 December 2030 and for signatures by 31 December 2031. These dates formally bind NIST publications and U.S. federal procurement, not private industry directly — but because FIPS 140-3/CMVP validation and many national/sector regulatory regimes reference NIST's approved-algorithm lists, vendors selling into regulated or government-adjacent supply chains should treat the same 2030/2035 horizon as a de facto product-planning deadline.
Sources
- NIST: FIPS 203, Module-Lattice-Based Key-Encapsulation Mechanism Standard (final, 13 August 2024) — full text, Abstract, Tables 1-3
- NIST: FIPS 204, Module-Lattice-Based Digital Signature Standard (final, 13 August 2024) — full text, Table 1 (parameters), Table 2 (sizes)
- NIST: FIPS 205, Stateless Hash-Based Digital Signature Standard (final, 13 August 2024) — full text, Table 2 (parameter sets)
- NIST: "NIST Releases First 3 Finalized Post-Quantum Encryption Standards" (13 August 2024 announcement — rationale, Moody/Locascio quotes, migration advice)
- NIST IR 8547: Transition to Post-Quantum Cryptography Standards (final, October 2026; Moody, Perlner, Regenscheid, Robinson, Cooper) — deprecation/disallowal tables
- NIST: Post-Quantum Cryptography project page (timeline, FN-DSA/FIPS 206 and HQC status, "can and should be put into use now")
- NIST IR 8547 (final, October 2026): PDF, Tables 2-5 and Sec. 3.1.1 code signing
- OMB Memorandum M-26-15, Execution of the Migration to Post-Quantum Cryptography (24 June 2026)
- govinfo: Executive Order 14412, Securing the Nation Against Advanced Cryptographic Attacks (22 June 2026, DCPD-202600417)
- NIST SP 800-208: Recommendation for Stateful Hash-Based Signature Schemes (LMS, XMSS; October 2020)
- pqm4: post-quantum cryptography benchmarks on the Arm Cortex-M4 (cycles, stack, code size)
About this guide
SoftWSP Firmware Security Desk
Covers the Firmware vulnerabilities & fixes cluster: 131 guides.

In this cluster
Questions about this page?
