Unlocking Private DeFi: 20,000× Throughput and 37× Lower Latency in Threshold FHE Decryption

Intro

Until now, threshold decryption in FHE was a bottleneck. The process was so slow and resource-intensive that building real-world applications on encrypted computation was virtually impossible. Competing approaches, such as noise flooding, required massive parameter inflation, heavy bootstrapping, and resulted in only 11–14 decryptions per second, with latencies of 71–90 milliseconds per operation.

That changes today.

Fhenix’s new Threshold (FHE) Decryption eliminates this bottleneck — and the breakthrough has already been recognized by the academic cryptography community. Our paper detailing this innovation has been accepted to @acm_ccs 2025 (Conference on Computer and Communications Security) — one of the most prestigious global conferences in security, privacy, and applied cryptography. This acknowledgment places Fhenix’s research alongside the work of Microsoft, Google, Meta, Stanford, and MIT — validating that our progress isn’t just fast… it’s foundational.

Problem vs Solution

For years, privacy in crypto has been more of a promise than a product. Everyone agrees that the next generation of DeFi — institutional-grade, compliant, yet censorship-resistant — needs to have privacy baked-in as a first class citizen. But there’s been a catch: performance. Even the most advanced Fully Homomorphic Encryption (FHE) systems, though mathematically elegant, have struggled to run fast enough for real-world finance. A critical bottleneck with FHE-based systems relates to decryption. Sometimes, users need to decrypt certain data. This process of unlocking encrypted data is known as Threshold (FHE) Decryption.

Threshold (FHE) Decryption is probably the most prevalent operation in confidential DeFi. For example, since all token balances are encrypted via FHE, any request to read a user’s balance requires performing a decryption operation. Now compound this: what happens when a 1,000 users each open up their encrypted wallet trying to read their 10-100 encrypted token balances at the same time? The network needs to perform 10,000-100,000 decryptions in a single second!

The problem? Up until today, threshold FHE decryption algorithms were capped at a few decryptions per second (on a single server). This limitation is inherent to the technique most leading FHE providers used, which required noiseflooding. As a result, these techniques, though useful, impose large parameters, high latency, and throughput measured only in the tens of operations per second at best, leaving real-world Private DeFi out of reach.

Fhenix’s Threshold (FHE) Decryption takes a different path. Instead of masking noise with more randomness, Threshold (FHE) Decryption securely removes it via a lightweight MPC-based rounding procedure. The protocol splits into an offline preprocessing phase and a low-communication online phase, preserving simulation-based security while eliminating the bottlenecks that have held FHE back.

That’s why this breakthrough matters. Fhenix’s new Threshold Decryption doesn’t just improve FHE performance — it makes Private DeFi possible. The same cryptographic principles that once existed only in whitepapers can now run at the speed of trading, lending, and settlement.

Results* (online phase, 4 parties, 1 ms ping):

Compared to the previous state-of-the-art currently used by others, we were able to achieve:

• Latency: 315.6 ms → 8.48 ms (~37× lower)
• Throughput: 3.18/s → 64,319/s (~20,000× higher)
• Scales to more parties: 8.48 ms (4p) → 10.09 ms (8p) → 24.23 ms (16p) at 1 ms ping
• Robust to network delay: ~56 ms online latency at 10 ms ping for 4–16 parties

These results aren’t just abstract benchmarks — they translate directly into user experience. For developers, it means building privacy-preserving dApps that actually respond in real time. For traders, it means encrypted order books that feel as smooth as centralized exchanges. And for institutions, it means fine-grained access control, compliance and usability.

Why it’s fast:

• No noise flooding. We avoid parameter inflation and the CPU-bound bootstrapping path.
• MPC-based rounding. Noise and message live in different “neighborhoods”; we extract and cancel noise before opening.
• Offline/online split. Precompute moderate-size comparison gates once; then decrypt in a small constant number of fast online rounds
• Rounds: 3 logical openings in the protocol; 6 rounds in our current instantiation (includes authenticity checks).

What this enables (examples):

• Sealed-bid auctions: 100 bids decrypt in tens of milliseconds (vs seconds), enabling real-time reveal phases.
• Encrypted order-book DEX: Per-order decrypts in single-digit to tens of milliseconds (network-dependent), approaching CEX-like responsiveness for the decryption step.Confidential stablecoins + private risk engines: High-rate decryption for limits, audits, and proofs across many vaults.

*Because the leading prior work didn’t release code, our comparison is based on a similar setup and focuses on the online phase only, so the numbers show directionally fair performance gains rather than one-to-one benchmarks.

Next Steps

1. @GuyZys will showcase this breakthrough at ACM CSS 2025 on October 15 in Taipei, Taiwan, with more content and confidential product demos built with CoFHE coming soon.‍
2. Build private dApps now:

In practical terms, this is what unlocks Private DeFi: the ability to execute complex logic — auctions, lending markets, cross-chain settlements — over encrypted data, at production speed. Until now, builders had to choose between privacy and usability. Fhenix’s work closes that gap. It turns FHE from a research curiosity into infrastructure that can underpin the next evolution of finance.

Read the paper.

Join the developer community.

Explore CoFHE docs and start shipping encrypted contracts.

The future of privacy on Ethereum needs encryption that works at production speed. This is that step.