UC Berkeley

The performance of semiconductor logic and memory devices has improved significantly with advancements in complementary metal-oxide-semiconductor (CMOS) manufacturing technology to miniaturize transistors, resulting in increased integrated circuit (“chip”) processing speed, energy efficiency, and cost per function. The proliferation of the Internet of Things (IoT) and the generation and processing of large data sets, commonly known as “big data,” have increased demand for non-volatile (NV) information storage that can be embedded with energy-efficient logic switches so that certain computational tasks can be performed in the memory itself. To achieve ultra-low-power electronics, alternative switching devices that can be operated with much smaller voltages than CMOS transistors are required. Micro-electro-mechanical (MEM) switches have been shown to be promising for ultra-low-power digital computing applications due to their negligible IOFF and abrupt switching behavior across a wide range of operating temperatures. Therefore, they have attracted growing interest for embedded energy-efficient logic and NVM applications. In addition to energy efficiency, it is also crucial to maintain the security and confidentiality of information collected and shared by IoT devices. To address this issue, memory devices that can store data and function as hardware security key generator are desirable. Emerging Resistive Random Access Memory (ReRAM) is a promising technology for hardware security applications due to its inherent variability.

This dissertation focuses on novel non-volatile memory devices and their applications. First, logic MEM switches are demonstrated to be operable as NV memory devices using controlled welding and unwelding of the contacting electrodes. Reprogrammability with consistently low programmed state resistance, and excellent (essentially infinite) retention time at elevated temperature, are experimentally demonstrated. These results indicate that MEM switches are promising for low-cost implementation of ultra-low-power integrated systems.

Next, this dissertation presents a prototype MEM switch design incorporating a floating gate (FG) for non-volatile charge storage. The FG-MEM NV switch potentially can achieve much longer data retention time than a conventional floating-gate MOS transistor, since there exists an air gap between conductive electrodes when the switch is in the OFF-state. Initial experimental results and FG-MEM NV switch design improvements are discussed.

Finally, this dissertation proposes a new method of generating Physically Unclonable Function (PUF) encryption keys that leverage the inherent random variability of ReRAM device programming time, for hardware security applications. The design and operation of a ReRAM device is presented, followed by a detailed discussion of the PUF key generation scheme. The randomness and reliability of the generated keys are then evaluated. The randomness of the ReRAM-based PUF is found to be of high quality compared to previous PUF implementations. Therefore, ReRAM technology enables the incorporation of both NVM and PUF functions within the same chip.