Here is the question that sounds like science fiction can a server add up your salaries, or run a search over your records, without ever being able to read them? Homomorphic encryption says yes, and the mechanism is genuinely strange the first time you meet it. You encrypt your data, hand the ciphertext to an untrusted server, the server does arithmetic directly on the encrypted values, and hands back an encrypted answer that only you can decrypt. At no point does it hold the plaintext.
The reason this was a research curiosity for decades and not a product is noise. Every homomorphic operation adds a little error to the ciphertext; do enough operations and the error swamps the signal and decryption fails. The breakthrough is in Stanford's grant US8515058B1, "Bootstrappable homomorphic encryption method, computer program and apparatus," issued August 2013 to Craig Gentry — the researcher whose 2009 work first showed fully homomorphic encryption was possible at all. Bootstrapping is the trick the title names: homomorphically evaluating the decryption circuit itself to produce a fresh ciphertext with the noise reset, so computation can continue indefinitely.
Read that claim slowly, because it is recursive in a beautiful way the scheme decrypts a ciphertext while it is still encrypted, using an encrypted copy of the secret key, and the output is a cleaner encryption of the same value. That self-referential refresh is what turns a scheme limited to a handful of operations into one that can run arbitrary computations. Everything practical in homomorphic encryption descends from getting bootstrapping to work.
The later grants in this space are about making it fast enough to use. Microsoft's US10541805B2 ("Variable relinearization in homomorphic encryption," 2020) attacks one of the expensive housekeeping steps; Samsung's US11575502B2 (2023) claims a hardware processing device for homomorphic encryption, which tells you the bottleneck moved from "is it possible" to "can we accelerate it in silicon." All three sit under CPC H04L 9/008, the subclass for homomorphic encryption specifically.
There is also an applied layer worth naming. Enveil's US10880275B2 ("Secure analytics using homomorphic and injective format-preserving encryption," 2020) claims running analytics — searches and aggregations — over data that stays encrypted end to end. That is the commercial promise made concrete query a dataset you are not allowed to see.
Why it matters: the regulatory and trust pressure to compute without exposing data — health records, financial data, cross-organization analytics — keeps rising, and homomorphic encryption is the cryptographic answer that needs no trusted middleman. The Gentry bootstrapping grant is the priority-dated root of the whole field, and the newer grants are the industry trying to make the impossible also be fast.