FHE · Glossary

Words we use, defined.

Decryption gap

The moment useful data must be decrypted to be computed on. Umbra is built to close it at production speed.

Ciphertext

Data in encrypted form. Homomorphic schemes compute on ciphertext and return ciphertext.

Fully homomorphic encryption (FHE)

Encryption that allows arbitrary computation on ciphertext. Decrypting the result matches computing on plaintext.

Lattice FHE

The dominant software family (BGV, BFV, CKKS and related). Correct, and historically too slow for a rack because of noise and bootstrapping.

Bootstrapping

Refreshing noisy ciphertext so more operations can run. Correct and expensive. Umbra runtime does not bootstrap.

TEE

Trusted execution environment. Shrinks who can see plaintext; it does not remove plaintext at the point of use.

Encrypted logic operation

The unit ZipLogic measures on Umbra silicon. Card throughput is measured; node throughput is estimated until end-to-end measurement completes. See /performance.

EPU

Encrypted Processing Unit: ZipLogic’s name for a purpose-built accelerator that runs on ciphertext. Not a claim that the category is already established.

Why FHE is slow →

Further reading, not ours.