Back to DashboardHardware InfrastructurePhase 6 · Infrastructure & Performanceintermediate60 min

Homomorphic Encryption (FHE)

Understand fully homomorphic encryption before choosing the optional advanced path for HSM custody, controlled decryption and assurance evidence.

Why this matters: FHE can keep inputs and results encrypted while an evaluator computes, but it does not by itself provide result integrity, hide every form of metadata, or make decryption authorization safe. Scheme choice, parameters and key custody determine whether the design meets its real threat model.

Start here: Start with the noise budget: apply additions and multiplications, watch CKKS levels and TFHE noise change, and explain when bootstrapping is required.

For your role

Executive / Business Leader
Use the FHE foundations path to separate the business promise from the limits: FHE can reduce evaluator trust, but cost, metadata leakage, result integrity and key-custody decisions remain.
GRC / Risk & Compliance
Use the FHE foundations path to test vendor claims against the actual boundary: FHE is not anonymisation, result integrity, FIPS validation or proof that a computation was correct.
Developer / Engineer
Start with scheme choice, noise and circuit shape, then use the optional advanced path to trace the calls made by the owner, evaluator and custodian.
Security Architect
Use the foundations path to choose the scheme and parameter envelope; the advanced path then maps secrets, public evaluation keys, ciphertexts and controlled decryption across trust zones.
Researcher / Academic
Compare the four ISO/IEC 28033 draft scheme families and the measured implementation evidence, then use the advanced path for decryption-oracle, flooding and threshold limitations.
IT Ops / DevOps
Start with the operational cost and failure boundary, then use the advanced path for controlled decryption, audit, rate limits, rotation, backup and recovery.
Curious Explorer
Follow the foundations path for the central idea: a server can calculate on encrypted data, but it pays a large performance cost and still reveals some metadata. The HSM custody material is intentionally left out of this path.
Practice in the Simulation

Choose your learning path

For: Anyone evaluating, designing, implementing or operating an FHE workload. This path completes on 8 Learn sections and 1 workshop step. Shared content counts toward every path; optional references are never required.

Key takeaways
  • FHE lets a server compute on data it only ever sees encrypted: the trust moves from hardware to mathematics.
  • “Fully” means any computation can in principle be evaluated, not that every computation is exact, fast or verified.

A TEE protects data in use by trusting hardware. Fully homomorphic encryption (FHE) takes the other route: it trusts mathematics. The server computes directly on ciphertexts and returns an encrypted result. It never sees the input or the output, and there is no enclave to attest. The cost is speed: FHE is orders of magnitude slower than computing in the clear.

TEEFHE
Root of trustCPU vendor, firmware, attestation chainHardness of lattice problems
Data inside the serverPlaintext, inside the enclaveAlways ciphertext
Side channelsMain attack surfacePlaintext is absent, but timing, traffic, access patterns, circuit shape, dimensions and implementation leakage remain
SpeedNear native10³–10⁶× slower; needs GPU/FPGA at scale
Integrity of the resultAttested codeNot provided — ciphertexts are malleable
Quantum statusAttestation signatures are classical todayCore scheme is lattice-based

Threat model before scheme choice

FHE primarily hides plaintext from the evaluator. It does not automatically prove that the evaluator ran the right circuit, authenticate ciphertexts, hide request timing or size, conceal every access pattern, protect a compromised client, or authorize who may receive a decrypted result. Record the evaluator, client, custodian and result recipient separately, then state which may be malicious or collude.

If correctness against a malicious evaluator matters, add an integrity mechanism such as a verifiable-computation proof or an application-specific check. Circuit privacy and model privacy are separate requirements from input confidentiality.

What “fully homomorphic” promises, and what it does not

Gentry proposed a fully homomorphic encryption scheme, “a scheme that allows one to evaluate circuits over encrypted data without being able to decrypt” (Gentry, STOC 2009). “Fully” means there is no limit on which circuit can be evaluated: with bootstrapping, additions and multiplications can be chained without end. It is a statement about what is possible, not about precision, speed, integrity or who sees the result. The name invites expectations that the sources qualify:

The name invitesWhat the sources show
Any computation can run on encrypted dataIn principle yes, but each scheme favours a data type. Microsoft SEAL: “For applications where exact values are necessary, the BFV and BGV schemes are more suitable”, CKKS suits real numbers, and with these schemes “encrypted comparison, sorting, or regular expressions, are in most cases not feasible”. TFHE handles comparisons and lookup logic, on small integers.
Results come back exactDepends on the scheme. CKKS is “a homomorphic encryption scheme for approximate arithmetic” (Cheon, Kim, Kim & Song 2017): results are close to the true value, correct to a number of bits. BFV, BGV and TFHE return exact integers as long as the noise stays within bounds.
The data is protectedWhoever holds the secret key, or may ask for decryptions, decides who sees results (section 8).
It runs at a usable speedAt a cost. The ICO describes HE as “several thousand times slower than processing plaintext”; DARPA estimated in 2021 that it was “about a million times slower” to compute in the FHE world; NIST SP 800-226 says that in most cases secure computation “is several orders of magnitude slower than native execution” and that “its performance is improving quickly” (section 4) (not sourced in the Library).
Decrypted results can be shared freelyNot without care. Decrypted CKKS results seen by others can lead to key recovery (Li & Micciancio 2021); the countermeasure, noise flooding, costs precision (Li, Micciancio, Schultz & Sorrell 2022). Exact schemes need a negligible decryption-failure probability (Cheon et al. 2024).

References

Definitions, schemes and data types

Decryption limits

Use cases

Overhead and acceleration

Standards and compliance

Operations, encoding and compilers

Market and adoption

Ready to test the concepts?

Start with noise, encoding and circuit design. The advanced path then follows HSM custody, controlled decryption, rotation and shared decryption.

Related Resources

Check off all sections and mark this reading done.

Check your understanding

14 questions on Homomorphic Encryption (FHE), each with its answer and the reason.

Take the quiz

Next step

Produce the artifact: Infrastructure Modernization Planner

This module belongs to phase 6 (Infrastructure & Performance); Infrastructure Modernization Planner produces a deliverable of that phase in the Command Center.

Production Tool Capabilities & Verification Checklist

Capabilities to require in your production tools:

  • Ring-LWE / RLWE parameter selection (BGV, BFV, CKKS, TFHE) compliant with the HomomorphicEncryption.org security standard parameter tables (ISO/IEC 28033 is still a Draft International Standard) and ≥128-bit security estimates.
  • Strict noise-budget monitoring and circuit-depth bounds preventing decryption failures or key-recovery attacks from IND-CPA^D oracle leakage.
  • Authenticated PQC transport (`ML-KEM` + `ML-DSA`) wrapping FHE evaluation keys and ciphertexts between client and compute nodes.

How to evaluate in your environment

Validate lattice dimension (N) and modulus (q) against the Lattice Estimator for at least 128-bit quantum security and test IND-CPA^D decryption error boundaries.

Production evidence to collect

FHE parameter security-level estimation report, noise-budget test log, and ciphertext serialization benchmark.

Completing this Hub module and its browser-based sample exercises teaches the technical and governance decision criteria for your role; it does not test, validate, or certify your deployed production systems.

Record evidence for your environment

Identify the tested build and configuration. Use reference IDs for observations. For a proposed exclusion, record the system boundary or rule and why it excludes this build.

Ring-LWE / RLWE parameter selection (BGV, BFV, CKKS, TFHE) compliant with the HomomorphicEncryption.org security standard parameter tables (ISO/IEC 28033 is still a Draft International Standard) and ≥128-bit security estimates.
Strict noise-budget monitoring and circuit-depth bounds preventing decryption failures or key-recovery attacks from IND-CPA^D oracle leakage.
Authenticated PQC transport (`ML-KEM` + `ML-DSA`) wrapping FHE evaluation keys and ciphertexts between client and compute nodes.

Evidence missing. 0 of 3 requirements have a reference or scope rationale.

Learning module content can be inaccurate. Please double-check its information. Report inaccuracies in PQC Today GitHub Discussions.