Encrypted Processor Framework & Accelerator
Fully homomorphic encryption solves the right problem and charges too much for it. Arbitrary computation over ciphertext is a genuine result, and the overhead — orders of magnitude, depending on the scheme and the circuit depth — keeps it out of most places it would be useful.
This project takes the opposite starting point. Rather than accepting FHE’s generality and attacking its cost, we modify a well-understood cryptosystem to support the operations that matter, and build the accelerator that makes those operations fast. The target is deliberately in between: stronger than computing on plaintext, and cheap enough to be a real option where FHE is not.
Two halves, and neither works alone:
- The framework defines what can be computed under encryption, what the modification costs in security margin, and where a program must break out to plaintext. That last boundary is the security-critical one, and stating it precisely is most of the intellectual work.
- The accelerator makes the resulting primitives fast in hardware. A framework whose operations are too slow to use is a paper result.
The work is a sibling to the lab’s Side-Channel-Resistant Computer Architecture project, which arrives at overlapping territory from the architecture side.
Open questions
- What security margin does the modification actually cost, stated against a standard model rather than by analogy?
- Which real workloads fit inside the supported operation set without breaking out to plaintext so often that the guarantee stops meaning anything?
- What is the accelerator’s speedup over a software implementation, and how does it compare against FHE on the same problem?
Dark Matter Next project
Security in Automobile Systems