终端安全是物联网安全的基础。基于存储器的物理不可克隆函数（PUF）是终端设备天然指纹，作为信任根可由内增强终端安全能力。但目前处理存储器PUF不稳定响应依赖纠错操作，执行开销大，难适应终端资源紧约束特性。本项目拟从改进存储器PUF响应产生机制新思路出发，研究多比特原始响应重组产生一比特新响应方法，从而直接测量新响应错误率。然后基于新响应相异错误率指导，因地制宜处理每一新响应，构建轻量安全密钥提取法，避免纠错操作，具体为：对多数新响应充足终端类，筛选高稳定新响应直接提取密钥，量化密钥重构故障率及新响应筛选率与筛选准则关系；对个别新响应不足终端类，拟借助服务平台丰富计算资源，对新响应错误率排序，融合试错机制间接提取密钥，量化误识率及误拒率与试错次数关系。最后基于提取密钥支持关键安全应用包括身份认证、数据加密、远程证明，在终端系统集成测试，优化执行开销与安全性，达到轻量化安全化目的。

Security of Internet of Thing (IoT) devices is paramount as IoT devices are responsible for sensing, aggregating and transmitting (sensitive) data. Memories pervasively embedded within IoT devices can be treated as intrinsic physical unclonable function (PUF), alike inseparable ‘fingerprint’. In this context, memory PUF as hardware root-of-trust can inherently enhance the security of IoT devices requiring neither hardware modification nor extra hardware overhead. However, it is still non-trival to directly mount memory PUF technique on resource-constraint IoT devices due to limitations posed by reliance on costly error correction to reconcile PUF unreliable responses. This project will address above challenge by reconsidering the memory PUF response generation mechanism. Firstly, we will investigate means of forming new memory PUF response by reforming/reorganizing a group of raw response bits to produce a 1-bit new response bit. The main benefit is that the bit-specific reliability of this new response bit can be accurately captured. Secondly, we will derive cryptographic keys from new bits assisted with bit-specific reliability to ultimately eschew the error correction. Specifically, whenever the reformed bits are sufficient for most IoT devices, highly reliable reformed bits can be selected to directly derive a key. In this context, the key failure rate and selection rate of high reliable reformed bits as a function of the selection criteria will be formalized and analysed. In case that the reformed bits are insufficient for rare IoT devices due to too restricted memory volume, resource-rich server will be exploited to firstly grade the reformed bits according to the bit-specific reliability and then perform the trial-and-error to indirectly restore the key. In this context, the false rejection rate and false acceptance rate as a function of the number of trials will be formalized and analysed. Thirdly, security applications including authentication, data encryption and remote attestation that build upon the derived key will be fully implemented on various IoT devices to comprehensively evaluate the system overhead and security. Overall, by constructively deriving a lightweight secure key that supports various IoT security applications from the intrinsic memory PUF, this project can enhance the security of IoT devices from internal physical layer.