320
V. Y. Zhuo et al.
[3]. Despite its dominance, the long-term scalability of flash memory is very uncertain
due to inherent physical limits and device reliability [4].
Consequently, there is a need for a new nonvolatile memory that can keep up with
the scalability and performance presented by future technology nodes. Many alternative memory technologies have emerged and are extensively studied and developed.
Amongst them, the resistive random-access memory (RRAM), which retains information in the form of distinct resistance states [5], is widely regarded as the most
promising for massive data storage.
Historically, the resistive switching phenomenon dates back more than two
centuries and was first demonstrated on the electric arc by Sir Humphry Davy in
1802 [6]. In the 1960s, advances in thin film technology enabled very high electric
fields in tri-layered structures which led to observable resistive switching behaviour
in ultrathin metal/oxide/metal films [7, 8]. However, research efforts in these devices
diminished after a decade of intensive study [8] due to the development of silicon
integrated circuits. In 1971, Leon Chua hypothesized the existence of memristors,
the fourth basic circuit element [9]. The recent intense surge of RRAM research
activities was renewed in the late 1990s by Asamitsu et al. [10], Beck et al. [11],
and Kozicki et al. [12] due to the declining progress of silicon technology. Then in
2008, Hewlett Packard Labs realized and termed analog switching RRAM devices
as memristors [13], thereby validating the fourth basic circuit element hypothesized
by Leon Chua.
In 2013, Panasonic produced the first commercialized embedded RRAM chip
[14]. Subsequently, Toshiba and Sandisk announced a 32 Gb high-density RRAM
chip [15]. In 2017, Taiwan Semiconductor Manufacturing Company (TSMC)
announced the production of embedded RRAM chips in 2019 with their 22 nm
technology process [16]. In addition, to achieve ultrahigh storage density, the twoterminal RRAM devices are implemented in a crosspoint array structure. In 2007,
Samsung demonstrated the first three-dimensional (3D) RRAM crosspoint array [17].
Notably, a 3D XPoint memory technology was released in 2015 by Intel Corporation
and Micron Technology Inc. Likewise, Crossbar Inc. also released 3D RRAM products with high scalability beyond the 10 nm node, as well as better read latency, energy
efficiency and write performance than the current NAND flash memory products.
Other than data storage, RRAM technology can also be used in the development
of human brain-like computing systems with very high energy efficiency, computing
capability and density scalability. For instance, matrix-vector multiplication or dot
product can be realized with RRAM crosspoint architecture [18, 19]. Functional
neuromorphic chips using binary or analog RRAM devices has been successfully
demonstrated [20–23] for online training.
Despite the advancements in RRAM technology, there is still insufficient breakthrough in the complete understanding of the RRAM physical switching mechanism
[24–26]. The main area of contention on RRAM mechanisms revolves around its
conductive filament (CF), especially on its composition and on how it connects and
ruptures.
In this chapter, we will first review the proposed RRAM mechanisms, followed by
the resistive switching materials. Thereafter, the RRAM crosspoint array architecture
V. Y. Zhuo et al.
[3]. Despite its dominance, the long-term scalability of flash memory is very uncertain
due to inherent physical limits and device reliability [4].
Consequently, there is a need for a new nonvolatile memory that can keep up with
the scalability and performance presented by future technology nodes. Many alternative memory technologies have emerged and are extensively studied and developed.
Amongst them, the resistive random-access memory (RRAM), which retains information in the form of distinct resistance states [5], is widely regarded as the most
promising for massive data storage.
Historically, the resistive switching phenomenon dates back more than two
centuries and was first demonstrated on the electric arc by Sir Humphry Davy in
1802 [6]. In the 1960s, advances in thin film technology enabled very high electric
fields in tri-layered structures which led to observable resistive switching behaviour
in ultrathin metal/oxide/metal films [7, 8]. However, research efforts in these devices
diminished after a decade of intensive study [8] due to the development of silicon
integrated circuits. In 1971, Leon Chua hypothesized the existence of memristors,
the fourth basic circuit element [9]. The recent intense surge of RRAM research
activities was renewed in the late 1990s by Asamitsu et al. [10], Beck et al. [11],
and Kozicki et al. [12] due to the declining progress of silicon technology. Then in
2008, Hewlett Packard Labs realized and termed analog switching RRAM devices
as memristors [13], thereby validating the fourth basic circuit element hypothesized
by Leon Chua.
In 2013, Panasonic produced the first commercialized embedded RRAM chip
[14]. Subsequently, Toshiba and Sandisk announced a 32 Gb high-density RRAM
chip [15]. In 2017, Taiwan Semiconductor Manufacturing Company (TSMC)
announced the production of embedded RRAM chips in 2019 with their 22 nm
technology process [16]. In addition, to achieve ultrahigh storage density, the twoterminal RRAM devices are implemented in a crosspoint array structure. In 2007,
Samsung demonstrated the first three-dimensional (3D) RRAM crosspoint array [17].
Notably, a 3D XPoint memory technology was released in 2015 by Intel Corporation
and Micron Technology Inc. Likewise, Crossbar Inc. also released 3D RRAM products with high scalability beyond the 10 nm node, as well as better read latency, energy
efficiency and write performance than the current NAND flash memory products.
Other than data storage, RRAM technology can also be used in the development
of human brain-like computing systems with very high energy efficiency, computing
capability and density scalability. For instance, matrix-vector multiplication or dot
product can be realized with RRAM crosspoint architecture [18, 19]. Functional
neuromorphic chips using binary or analog RRAM devices has been successfully
demonstrated [20–23] for online training.
Despite the advancements in RRAM technology, there is still insufficient breakthrough in the complete understanding of the RRAM physical switching mechanism
[24–26]. The main area of contention on RRAM mechanisms revolves around its
conductive filament (CF), especially on its composition and on how it connects and
ruptures.
In this chapter, we will first review the proposed RRAM mechanisms, followed by
the resistive switching materials. Thereafter, the RRAM crosspoint array architecture