Novel Ferroelectric Tunnel FinFET based Encryption-embedded Computing-in-Memory for Secure AI with High Area-and Energy-Efficiency
Jin Luo, Hanyong Shao, Boyi Fu, Zhiyuan Fu, Weikai Xu, Kaifeng Wang, Mengxuan Yang, Yiqing Li, Xiao Lv, Qianqian Huang, Ru Huang
Abstract
In this work, novel ferroelectric tunnel FET (FeTFET) is proposed and experimentally demonstrated to implement encryption-embedded computing-in-memory with non-volatility (nvCIM), enabling both in-situ key authentication and multiply-accumulate (MAC) operation with high area-and energy-efficiency. For the first time, XOR-cipher for weight encryption is merged into XNOR-based MAC, eliminating explicit decryption process and simplifying the local multiplication directly on the encrypted weight with XNOR-operator. Furthermore, by exploiting and modulating the non-volatile ferroelectric polarization for encrypted weight storage and the unique feature of ambipolar tunneling current for input, the XNOR operator can be realized by the fabricated FeTFET based on 14-nm FinFET technology node with only one transistor, enabling the encryption-embedded MAC with multilevel weight without the need of extra decryption circuitry or complementary encrypted weight storage. Based on the proposed FeTFET-based encryption-embedded nVCIM design, security-enhanced neural network inference and one-shot learning are demonstrated with high energy efficiency, showing its great potential for secure AI.