Changing Memory Requirements for IoT

Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), and Flash memory were invented in 1964, 1966, 1980, respectively. They have been the primary memory solutions over the last a few decades. Continuous development of higher performance digital systems has demanded faster memory solutions so that the overall system performance, usually limited by memory performance, can be improved to meet the market needs. Among the aforementioned memory solutions, DRAM and SRAM are faster than flash memory, which makes them employed as primary computing memories. Flash memory was invented as a technology for data storage without power supply targeting to replace various hard disks. Even though the progress in the flash memory speed has been demonstrated, it is still slower than DRAM and SRAM, and is mainly used as a non-volatile data storage solution.

However, the introduction of Internet of Things (IoT) has changed the memory requirements. The topmost goal of many IoT devices is to minimize overall power and energy to maximize the system lifetime when powered by batteries or energy harvesting devices. Various emerging non-volatile devices such as Magnetic Random Access Memory (MRAM) and Resistive Random Access Memory (ReRAM) show good potential as a memory solution for IoT devices. MRAM has been commercialized by companies like GlobalFoundries (with Everspin) and Samsung Electronics. While MRAM provides good performance for IoT devices, it is difficult to fabricate because of many thin layers requiring precise fabrication control. This will increase the overall fabrication cost, which limits its proliferation in the IoT era.

Advantages and Limitations of ReRam

ReRAM is another promising candidate for IoT devices because of its simple structure and low-cost fabrication. Like MRAM, ReRAM also shows different resistance levels depending on how it is programmed (Fig. 1). Higher resistance is formed when no conductive filament is formed between two electrodes (Fig. 1 (left)). Lower resistance occurs when the two electrodes are electrically connected through a conductive filament (Fig. 1 (right)). In spite of various advantages of ReRAM, it is not mature yet and still has various challenges like technology compatibility, endurance, and variability to be overcome for commercialization. Currently, most of the demonstrated ReRAM chips use higher voltage for programming, which cannot be accepted in most main-stream semiconductor fabrication technologies. Besides, the ReRAM device characteristics changes significantly over usage time, limiting the device lifetime.

Fig. 1. Two resistance states of ReRAM

Research Team to Develop Industry-level ReRAM Technologies

Recently, Nanyang Technological University (NTU) formed a research team (led by Prof. Lew Wen Siang in School of Physical and Mathematical Sciences) with GlobalFoundries to develop industry-level ReRAM technologies. The NTU team consists of professors, researchers, and Ph.D students with various research expertise covering from materials to circuits and architecture. The NTU research team dreams of developing ReRAM technologies ready for mass production after leveraging the manufacturing technologies of GlobalFoundries.

Fig. 2. Simplified ReRAM architecture for explaining data sensing

ABOUT THE AUTHOR
PROF. TONY TAEHYOUNG KIM
Associate Professor of School of Electrical & Electronic Engineering, Deputy Director of Centre for Integrated Circuits and Systems (CICS) & Program Lead of NTU-TUM Integrated Circuits Design (ICD) Programme