**Preface to "Semiconductor Memory Devices for Hardware-Driven Neuromorphic Systems"**

Artificial intelligence (AI) is a technological area that has been under development for half a century. It is a term familiar to many people. AI has been exquisitely shaped into machine learning, and more recently, deep neural networks. As its evolution progresses, AI technology is infiltrating our daily lives more profoundly. Although AI technology has predominantly grown in computer and software engineering so far, further developments can be made in a hardware sense for higher system energy efficiency and more portable end-user-friendly edge applications. In order to more effectively mimic "our way of thinking" , mathematical analogy and the hardware-sense realization should go together hand in hand. AI can be more specifically termed as neuromorphic when the mathematical/algorithmic essences are realized by AI-oriented, specially designed hardware components. Hardware-sense AI can appear in general-purpose processing units made of conventional transistors. Integration of a large number of processing units can eventually mimic our way of thinking and can perform better depending on area. However, a lack of real AI may be perceived if volume and energy consumption are not considered. In order to address this deficiency, renovations should be realized at the device level. More synapse-like electron devices in terms of integration density, completeness in realizing biological synaptic behaviors, and energy-efficient operations are considered to be central for next-generation neuromorphic chips. The most important distinguishable feature between conventional AI chips and advanced neuromorphic systems is energy efficiency. However, only recently has this revolutionary synaptic device technology been implemented with semiconductor memory devices, materials, and processing technologies, with the aim of device scaling, data storage and processing, and low-power operation capabilities. It is the right time to investigate how the neuromorphic system and its building component technologies are developing. It cannot be underestimated that neural networks representing mathematical frames, energy-efficient memory-based synaptic devices, neurons and relevant circuits need to accompany one-another in good balance and harmony for ultra-low-power and super-light neuromorphic systems. This book will help the readership understand the evolutionary direction of neuromorphic systems, which is made in more hardware-driven ways, and provides perspectives in the relevant fields.

I deeply thank all the authors who have contributed the research articles with the best recency and also would like to give my sincere gratitude to my colleages, collaborators, family members, and my lifetime advisor, Prof. Byung-Gook Park. Also, the support for the research on neuromorphic devices and systems by Nano Material Technology Development Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (MSIT) (NRF-2016M3A7B4910348) is acknowledged.

> **Seongjae Cho** *Editor*

*Article*
