A Novel STT-MRAM-Based Massive Parallelism Strong PUF with High Throughput and Low BER for IoT Security

The lightweight design and reliability of physical unclonable functions (PUFs) are critical for IoT security. In this work, we propose a strong PUF based on spin-transfer torque magnetoresistive random access memory (STT-MRAM). For the first time, we adopt a logic-in-memory (LIM) structure known as Computational Random Access Memory (CRAM) in order to promote massive parallelism and low area overhead. In our simulations of the results, the proposed PUF macro occupies 2960.4 μm2 in total, 5.77 μm2 per cell, and presents a high bit rate of 1.42 Gb/s. In addition, we introduce the TRIM circuit and algorithm into the response generation process and develop two TRIM algorithms to optimize its reliability. The proposed PUF presents an excellent average intra-chip Hamming distance of 0.288% under the temperature range from −10 °C to 150 °C. The PUF also achieves an inter-chip Hamming distance of 50.003% and generates 22N/3 challenge–response pairs (CRPs), which surpass other state-of-the-art MRAM strong PUFs. More importantly, the proposed PUF shows excellent machine-learning resistance. This work provides a new strong PUF solution for resource-limited system-on-chip (SoC) chips for edge applications.