AI Chips Improve Performance of Homomorphic Encryption
/ 1 min read
🧠⚡ AI Chips Enhance Homomorphic Encryption Performance. Researchers have developed a method to leverage existing ASIC AI accelerators, such as TPUs, to improve the efficiency of homomorphic encryption (HE), which is crucial for secure cloud-based services. The proposed CROSS compiler adapts HE primitives into AI operators, enabling significant performance gains by implementing modular multiplication and high-precision arithmetic. Evaluations on Google TPUv4 show up to 161x speedup compared to many-core CPUs and 5x compared to V100 GPUs. This advancement addresses the latency issues associated with HE, making it more viable for practical applications while ensuring data privacy. The kernel-level codes are available as open-source, promoting further research and development in this area.
