Silent Two-party Computation Assisted by Semi-trusted Hardware

Yibiao Lu, Bingsheng Zhang, Weiran Liu, Lei Zhang, Kui RenIACR ePrint 2020

Introduction

Secure Multi-party Computation (MPC) is a cornerstone of modern cryptography, enabling parties to compute functions over private inputs without revealing the inputs themselves. Despite its theoretical elegance, practical deployment of MPC is often hindered by significant communication overhead. Traditional protocols, such as those based on garbled circuits, require the transmission of massive amounts of data—often gigabytes—over the network. This bottleneck is particularly acute in Wide Area Networks (WAN), where latency and limited bandwidth drastically slow down computation.

Recent advancements have explored the use of Trusted Execution Environments (TEEs), like Intel SGX, to offload heavy cryptographic tasks. However, these "trusted hardware" solutions typically rely on the assumption that the hardware manufacturer is fully honest. Given the discovery of architectural vulnerabilities and potential backdoors, this "full trust" model is increasingly viewed as a security risk. This paper presents a novel approach: using semi-trusted hardware to achieve "silent" MPC, where communication costs are minimized without requiring absolute trust in the hardware provider.

Methodology: Correlated Randomness Teleportation

The authors introduce the concept of Correlated Randomness Teleportation. Instead of transmitting large pre-computed tables or Oblivious Transfer (OT) pairs, parties leverage a TEE enclave to generate this data locally.

The Semi-Trusted Hardware Model

The proposed model assumes the hardware is "semi-trusted." An adversary might passively leak the enclave's inputs or even maliciously replace its internal logic. To maintain security under these conditions, the protocol ensures that private user inputs never enter the enclave. Instead, the enclave is used solely to generate input-independent correlated randomness.

Protocol Steps

  1. Seed-based Generation: Parties send small, 128-bit random seeds to the TEE enclave.
  2. Local Expansion: The enclave uses these seeds to expand the correlated randomness (e.g., garbled tables or OT pairs) within its secure memory. This transforms a network-bound task into a memory-bound one.
  3. Silent Execution: Because the heavy cryptographic material is generated locally on both sides (or "teleported" via the seed), the network communication becomes independent of the circuit size, scaling only with the size of the actual inputs.
  4. Verification: To guard against a maliciously corrupted TEE that might produce incorrect data, the protocol employs a lightweight hashing mechanism. Parties exchange and verify hashes of the generated randomness to ensure consistency before proceeding.

Experimental Evaluation

The researchers benchmarked their "silent" protocols against the industry-standard EMP-toolkit using common cryptographic circuits: AES-128, SHA-256, and SHA-512. These benchmarks were conducted in both Local Area Network (LAN) and Wide Area Network (WAN) environments to assess scalability.

Random Oblivious Transfer (ROT) Performance

One of the most significant bottlenecks in MPC is the generation of Oblivious Transfer pairs. As shown in the results, the "teleportation" method offers dramatic improvements.

Performance comparison for Random Oblivious Transfer (ROT) generation across different network conditions.

For generating $10^8$ ROTs, the new protocol is significantly faster. In WAN settings, where network latency is high, the speedup reaches up to 450X compared to traditional software-based OT extension (EMP-IKNP-ROT).

Two-Party Computation (2PC) Results

The efficiency gains extend to full 2PC evaluations of complex circuits. By removing the need to send garbled circuits over the wire, the protocol shifts the bottleneck from the network interface to the CPU and RAM.

Comparison of 2PC running times for AES and SHA circuits between the baseline EMP-AG2PC and the proposed silent protocol.

As illustrated in the table above, the proposed protocol achieves a 5.3–5.7X speedup in LAN and a 43–47X speedup in WAN for AES and SHA circuits. This demonstrates that "silent" MPC can make complex secure computations practical even over standard internet connections.

Key Insights

  • Decoupling Trust from Performance: By using the TEE only for randomness generation, the protocol gains the performance of hardware acceleration without exposing sensitive user data to potential hardware vulnerabilities.
  • Network Independence: The "silent" nature of the protocol means that as circuits grow in size (e.g., from SHA-256 to SHA-512), the communication cost remains constant, making it highly scalable for large-scale data processing.

Conclusion and Future Work

This work demonstrates that semi-trusted hardware can revolutionize the efficiency of MPC. By "teleporting" correlated randomness, the authors effectively bypass the communication bottleneck that has long plagued secure computation.

Future research directions include extending this model to multi-party settings (beyond 2PC) and addressing the lack of "accountability"—currently, if a protocol aborts due to a hash mismatch, it is difficult to determine if the hardware or a specific party was at fault. Additionally, while the model accounts for "leaky" hardware, ongoing research into TEE side-channels remains vital for ensuring long-term security.

Summary updated

BibTeX
@article{lu2020silent,
  author = {Yibiao Lu and Bingsheng Zhang and Weiran Liu and Lei Zhang and Kui Ren},
  title = {{Silent Two-party Computation Assisted by Semi-trusted Hardware}},
  journal = {IACR Cryptol. ePrint Arch.},
  year = {2020},
}
Figure