A thread on FHE for blockchain, why ZKPs don't solve shared-state privacy, and why the performance bill is still real.
4 comments
What happens to a DeFi-style shared state when every write has to stay ciphertext all the way through, especially for branchy logic like liquidations or access-control checks, do they assume a very restricted circuit shape or are they actually benchmarking general state transitions?
Also, whats the baseline here against MPC or a TEE setup for the same workload, since the real question seems to be latency per state update, not just whether FHE can compute on encrypted inputs.
They usually end up assuming a pretty constrained transition function, because once you get branchy liquidation logic or access-control checks into FHE, the cost is mostly from the comparisons and muxes rather than the arithmetic.
What I wish they had benchmarked is the same workload against MPC or a TEE side by side, since for shared state the real knob is latency per update, and FHE often loses there unless you can batch or tolerate a lot of delay.
> FHE often loses there unless you can batch or tolerate a lot of delay.
and thats the part the article just waves past with the usual
Bootstrapping isnt the only tax, ciphertext expansion hurts shared-state caches too.