Resistive Random Access Memory Device …
335
PCRAM crosspoint array using OTS selector that exhibited very low write (<300 ns)
and read (<100 ns) latencies, good write high endurance (>10
7 ) and excellent thermal
retention (>10
4 h at 85 °C) [148]. In terms of integration between OTS selector and
RRAM, Lee et al. from Samsung reported in 2012, a 1S1R memory cell using a
TaO x -based RRAM and a AsGeTeSiN-based OTS select device [149]. The nanoscale
selector showed high selectivity (10
3 ), endurance (10
8 ) and on-state current density
(>10MA/cm
2 ).
Similarly, in 2017, Alayan et al. reported a 1S1R memory where HfO x -based
RRAM was stacked with a GeSe OTS selector [150]. A stable and novel reading
strategy was proposed and up to 10
6 read cycles have been achieved in their 1S1R
memory cell.
In 2014, Jo et al. from Crossbar Inc. reported a Field Assisted Superlinear
Threshold (FAST) Selector based on a superlinear threshold layer (SLT) whereby a
conduction channel is created at the threshold electric field [151]. By themselves,
these Type II threshold selectors exhibited excellent selector performance such as
very high non-linearity (>10
10 ), large drive current density (>5 MA/cm
2 ), high
endurance (>10
8 ) and fast operation speeds (<50 ns). When the FAST selectors are
integrated with low-current, forming-free RRAM cells in a 4 Mb 1S1R crosspoint
array, the integrated 1S1R device demonstrated >10
2 memory on/off ratio and >10
6
non-linearity over 10
5 cycles [151].
Despite intense development of selector research in recent years, there is still
no consensus on the best selector candidate as it must match the characteristics of
the non-volatile memory device. Thus, the selector device still remains as a crucial
challenge in the implementation of ultra-high density crosspoint memory architectures. Ultimately, a self-rectifying RRAM device would be ideal as it eliminates the
requirement of an additional selector element.
4 Conclusions and Future Outlook
In the past decade, RRAM has seen much progress from single device up to 3D crosspoint array demonstrations. Nevertheless, challenges still persist such as detailed clarification of the resistive switching dynamics, the perfect matching of RRAM with
selector or a reliable self-rectifying RRAM for true 3D vertical crosspoint implementation. Furthermore, new computing architectures are emerging, facilitated by
the unique physical properties of RRAM and the perfect pairing between RRAM
crosspoint arrays and the matrix-vector computation.
In the near future, RRAM technology is expected to have great potential in both
data storage and computing. For instance, RRAM can be leveraged as a storage-class
memory, which fills the gap between main memory and storage memory. In addition,
the high-density, low-power and multi-level capabilities of RRAM has also motivated
research into neuromorphic computing systems as a synaptic device, which would
induce a paradigm shift in computing technology and a revolution in future electronic
devices.
335
PCRAM crosspoint array using OTS selector that exhibited very low write (<300 ns)
and read (<100 ns) latencies, good write high endurance (>10
7 ) and excellent thermal
retention (>10
4 h at 85 °C) [148]. In terms of integration between OTS selector and
RRAM, Lee et al. from Samsung reported in 2012, a 1S1R memory cell using a
TaO x -based RRAM and a AsGeTeSiN-based OTS select device [149]. The nanoscale
selector showed high selectivity (10
3 ), endurance (10
8 ) and on-state current density
(>10MA/cm
2 ).
Similarly, in 2017, Alayan et al. reported a 1S1R memory where HfO x -based
RRAM was stacked with a GeSe OTS selector [150]. A stable and novel reading
strategy was proposed and up to 10
6 read cycles have been achieved in their 1S1R
memory cell.
In 2014, Jo et al. from Crossbar Inc. reported a Field Assisted Superlinear
Threshold (FAST) Selector based on a superlinear threshold layer (SLT) whereby a
conduction channel is created at the threshold electric field [151]. By themselves,
these Type II threshold selectors exhibited excellent selector performance such as
very high non-linearity (>10
10 ), large drive current density (>5 MA/cm
2 ), high
endurance (>10
8 ) and fast operation speeds (<50 ns). When the FAST selectors are
integrated with low-current, forming-free RRAM cells in a 4 Mb 1S1R crosspoint
array, the integrated 1S1R device demonstrated >10
2 memory on/off ratio and >10
6
non-linearity over 10
5 cycles [151].
Despite intense development of selector research in recent years, there is still
no consensus on the best selector candidate as it must match the characteristics of
the non-volatile memory device. Thus, the selector device still remains as a crucial
challenge in the implementation of ultra-high density crosspoint memory architectures. Ultimately, a self-rectifying RRAM device would be ideal as it eliminates the
requirement of an additional selector element.
4 Conclusions and Future Outlook
In the past decade, RRAM has seen much progress from single device up to 3D crosspoint array demonstrations. Nevertheless, challenges still persist such as detailed clarification of the resistive switching dynamics, the perfect matching of RRAM with
selector or a reliable self-rectifying RRAM for true 3D vertical crosspoint implementation. Furthermore, new computing architectures are emerging, facilitated by
the unique physical properties of RRAM and the perfect pairing between RRAM
crosspoint arrays and the matrix-vector computation.
In the near future, RRAM technology is expected to have great potential in both
data storage and computing. For instance, RRAM can be leveraged as a storage-class
memory, which fills the gap between main memory and storage memory. In addition,
the high-density, low-power and multi-level capabilities of RRAM has also motivated
research into neuromorphic computing systems as a synaptic device, which would
induce a paradigm shift in computing technology and a revolution in future electronic
devices.