Introduction
As data-driven analytics expand, organizations increasingly outsource computation to cloud environments. However, this shift creates a significant privacy gap: the challenge of processing sensitive data on remote, untrusted servers. While cryptographic solutions like Fully Homomorphic Encryption (FHE) and Secure Multi-Party Computation (MPC) provide strong theoretical guarantees, their computational overhead often renders them impractical for large-scale, real-world applications.
Trusted Execution Environments (TEEs), such as Intel SGX and ARM TrustZone, offer a hardware-based alternative that promises near-native CPU performance. Yet, TEEs are not a panacea; they remain vulnerable to side-channel and state-rewind attacks. This survey provides a systematic review of TEE-based secure computation protocols, proposing a unified taxonomy and rigorous assessment criteria to evaluate the current landscape.
Taxonomy of Secure Computing Frameworks
The authors categorize TEE-based secure computation into three primary architectural frameworks, distinguished by how data is distributed and who controls the computation:
- Secure Outsourced Computation (SOC): A single client sends encrypted data to an untrusted cloud server equipped with a TEE. The server performs the task and returns the result.
- Secure Distributed Computation (SDC): Data is processed across multiple servers (master and slave nodes), often utilizing parallel processing frameworks like MapReduce to handle large-scale datasets.
- Secure Multi-party Computation (SMC): Multiple parties with private inputs jointly compute a result without a central coordinator, using TEEs to ensure that no party learns anything about the others' inputs beyond the final output.
Evaluation Criteria and Methodology
To compare state-of-the-art solutions across general-purpose computation, machine learning, and encrypted database queries, the survey establishes a multi-dimensional assessment framework. Protocols are evaluated based on four core dimensions: Setting, Methodology, Security, and Performance.
Beyond these frameworks, the survey identifies three distinct TEE utilization modes:
- Pure TEE: Utilizing the TEE as a "black box" for general confidentiality.
- TEE + Oblivious Primitives (OP): Integrating techniques like Oblivious RAM (ORAM) to prevent side-channel leaks stemming from memory access patterns.
- TEE + Trusted Public Ledger (PL): Using blockchains or ledgers to prevent "rewind attacks," where a malicious host resets the TEE to an older state to extract information or bypass security logic.
Key Insights and Performance Benchmarks
The survey highlights a critical trade-off between performance and security. While TEEs are inherently fast, making them "oblivious" to stop side-channels introduces significant overhead. For example, adding ORAM can turn a near-native execution into a multi-factor slowdown. However, hybrid approaches—combining hardware isolation with classic cryptographic primitives like Garbled Circuits or Secret Sharing—show great promise in mitigating hardware-specific vulnerabilities without the full cost of pure crypto.
Key results from the summarized benchmarks include:
- Machine Learning: TEE-based training (e.g., OMPML) can achieve 1.7x–2.5x speedups over traditional MPC-based solutions. Secure inference (SOI) typically sees a practical 1.6x–2.5x overhead compared to plaintext.
- Database Queries: Advanced protocols can improve performance by 2,000x–18,000x compared to pure cryptographic baselines for complex operations like select-join.
Future Frontiers
The authors identify several challenges for the next generation of TEE research. Scalability remains a concern due to the limited size of the Enclave Page Cache (EPC). There is also a growing need for real-time incremental computation and energy-efficient TEE protocols for mobile offloading. Finally, trust management remains a core philosophical issue: users must currently trust the hardware manufacturer. Research into multi-vendor TEEs or protocols that weaken this trust dependency is a vital future direction.
Summary updated