Homomorphic encryption lets a cloud provider compute on encrypted data without ever decrypting it, but choosing a scheme is fundamentally a trade-off decision. Partially homomorphic encryption (PHE) handles only one operation type and runs fastest; somewhat homomorphic encryption (SHE) supports limited combinations of addition and multiplication before noise breaks the ciphertext; fully homomorphic encryption (FHE) supports arbitrary computation but pays for that flexibility with heavy computational overhead. The right choice depends on how complex your computation is and how much latency and resource consumption you can tolerate.

Understanding Homomorphic Encryption: A Foundation for Secure Computation

Conventional encryption protects data at rest and in transit, but it forces a dilemma the moment you want to process that data in the cloud: either decrypt it on the provider's infrastructure, exposing plaintext to a third party, or keep it encrypted and lose the ability to compute on it. Homomorphic encryption resolves this by allowing certain mathematical operations to be performed directly on ciphertexts, producing an encrypted result that, when decrypted by the key holder, matches what the same operations would have produced on plaintext.

The practical implication, as the general reference literature on homomorphic encryption notes, is that a program built on such a scheme "need never decrypt its inputs, so it can be run by an untrusted party without revealing its inputs and internal state." That property makes homomorphic encryption especially relevant to outsourcing private computations to cloud environments. Entrust, which publishes learning resources on the technology, frames the three main types as points on a single spectrum: each "offers a trade-off between computational capacity and efficiency," and because encryption consumes real infrastructure and energy resources, understanding all three types matters before committing to one.

Partially Homomorphic Encryption (PHE): Single Operations, High Efficiency

PHE is the simplest and fastest of the three types. It permits an unlimited number of computations, but all of them must be of a single kind — either addition or multiplication on ciphertexts, never both. Classic examples cited in technical glossaries include RSA, which supports multiplication, and Paillier, which supports addition.

This constraint sounds severe, but it maps neatly onto a surprising range of real tasks. Any workload whose core logic reduces to repeated addition — summing values, tallying votes, computing basic aggregate statistics across a dataset — fits an additive PHE scheme like Paillier without any compromise. Because the scheme does so little cryptographic work per operation, it carries the lowest computational overhead of the three types. The limitation is equally clear: PHE cannot support complex multi-step computations that mix operations, so anything beyond single-operation aggregation requires a more capable scheme.