328
V. Y. Zhuo et al.
(SE) sandwiched between an electrochemically active electrode (AE) and an electrochemically inert electrode (CE). Candidates for the AE are limited to some metals
that are mobile in the SE, and are usually Cu [90], Ag [91] or an alloy such as CuTe
[35]. CE materials are less limited and include Pt, Ru, Ir, Au, W, Mo, Co, Cr, TiW,
TaN or poly-Si [34]. Conventionally, the SE materials have been amorphous Si, C,
doped organic semiconductors, iodides, methylsilesquioxane, selenides, sulphides,
tellurides, ternary chalcogenides, or even non-solid electrolytes like water and even
vacuum gaps [92–95]. Recently, more and more metal oxides and nitrides have been
used, including Al 2 O 3 , CuO x , HfO 2 , GdO 2 , MoO x , SiO 2 , Ta 2 O 5 , TiO 2 , WO 3 , ZnO,
ZrO 2 , and AlN [96–98]. The transition from traditional electrolytes to the complementary metal-oxide-semiconductor (CMOS) compatible and inexpensive oxides
have enhanced the retention and operation voltages of CBRAM, making them more
suitable for certain applications such as switches in large-scale integrated circuits
[96]. More details on the material systems available for CBRAM can be obtained
from the comprehensive review by Valov et al. [34].
Switching mechanisms in CBRAM are similar to that in OxRRAM. Electroforming in CBRAM also causes structural alterations to the SE and creates nanoscale
CFs to accommodate the electrochemically active metal ions for the subsequent resistive switching [35]. Most CBRAM devices exhibit bipolar switching where by the
electric field is the dominant driving force. However, there are also reports of unipolar
CBRAM, which suggests the possibility of Joule heating effects [99, 100]. Unlike
OxRRAM, the switching mechanisms are better understood since the metal ions are
more easily observed using microscopy techniques, thus the kinetic processes of
the CF can be more easily verified [101]. Since ion migration and associated redox
processes during resistive switching occur at the nanoscale, the TEM, specifically the
in situ TEM, is currently one of the best techniques used to understand the dynamic
processes of resistive switching as it can yield details on the chemical state, composition, morphology, size and trace the evolution of the CF growth/dissolution down
to individual metal nanoclusters. As summarized by Yang et al. [102], there are
mainly four different electrochemical metallization processes in CBRAM devices
depending on the cation mobility and the redox reaction rate as shown in Fig. 6. The
cation mobility influences the nucleation site of the CF and the direction of the CF
growth whereas the redox reaction rate determines the ion supply which decides the
CF morphology [102].
As schematically shown in Fig. 6a, when cation mobility and redox reaction rate
are high in the SE, it leads to inverted cone-shaped CFs that initiate from the inert CF.
This has been directly verified using in situ TEM for Ag/H 2 O/Pt (see Fig. 6e) [101],
Cu/Al 2 O 3 /Pt [103] and Cu/Cu-GeTe/Pt–Ir [104] devices. On the other hand, when
both cation mobility and redox reaction rate are low, the cations traverse a short
distance from the AE to attain critical nucleation conditions within the dielectric
which leads to metal cluster formation. These metal clusters then form a CF that
grows from the AE to the CE as shown schematically in Fig. 6b and has been
reported in Ag/SiO 2 /W (see Fig. 6f) [105], Ag/ZrO 2 /Pt [106], Ag/a-Si/Pt [107], and
TiN/Al 2 O 3 /Cu [108].
V. Y. Zhuo et al.
(SE) sandwiched between an electrochemically active electrode (AE) and an electrochemically inert electrode (CE). Candidates for the AE are limited to some metals
that are mobile in the SE, and are usually Cu [90], Ag [91] or an alloy such as CuTe
[35]. CE materials are less limited and include Pt, Ru, Ir, Au, W, Mo, Co, Cr, TiW,
TaN or poly-Si [34]. Conventionally, the SE materials have been amorphous Si, C,
doped organic semiconductors, iodides, methylsilesquioxane, selenides, sulphides,
tellurides, ternary chalcogenides, or even non-solid electrolytes like water and even
vacuum gaps [92–95]. Recently, more and more metal oxides and nitrides have been
used, including Al 2 O 3 , CuO x , HfO 2 , GdO 2 , MoO x , SiO 2 , Ta 2 O 5 , TiO 2 , WO 3 , ZnO,
ZrO 2 , and AlN [96–98]. The transition from traditional electrolytes to the complementary metal-oxide-semiconductor (CMOS) compatible and inexpensive oxides
have enhanced the retention and operation voltages of CBRAM, making them more
suitable for certain applications such as switches in large-scale integrated circuits
[96]. More details on the material systems available for CBRAM can be obtained
from the comprehensive review by Valov et al. [34].
Switching mechanisms in CBRAM are similar to that in OxRRAM. Electroforming in CBRAM also causes structural alterations to the SE and creates nanoscale
CFs to accommodate the electrochemically active metal ions for the subsequent resistive switching [35]. Most CBRAM devices exhibit bipolar switching where by the
electric field is the dominant driving force. However, there are also reports of unipolar
CBRAM, which suggests the possibility of Joule heating effects [99, 100]. Unlike
OxRRAM, the switching mechanisms are better understood since the metal ions are
more easily observed using microscopy techniques, thus the kinetic processes of
the CF can be more easily verified [101]. Since ion migration and associated redox
processes during resistive switching occur at the nanoscale, the TEM, specifically the
in situ TEM, is currently one of the best techniques used to understand the dynamic
processes of resistive switching as it can yield details on the chemical state, composition, morphology, size and trace the evolution of the CF growth/dissolution down
to individual metal nanoclusters. As summarized by Yang et al. [102], there are
mainly four different electrochemical metallization processes in CBRAM devices
depending on the cation mobility and the redox reaction rate as shown in Fig. 6. The
cation mobility influences the nucleation site of the CF and the direction of the CF
growth whereas the redox reaction rate determines the ion supply which decides the
CF morphology [102].
As schematically shown in Fig. 6a, when cation mobility and redox reaction rate
are high in the SE, it leads to inverted cone-shaped CFs that initiate from the inert CF.
This has been directly verified using in situ TEM for Ag/H 2 O/Pt (see Fig. 6e) [101],
Cu/Al 2 O 3 /Pt [103] and Cu/Cu-GeTe/Pt–Ir [104] devices. On the other hand, when
both cation mobility and redox reaction rate are low, the cations traverse a short
distance from the AE to attain critical nucleation conditions within the dielectric
which leads to metal cluster formation. These metal clusters then form a CF that
grows from the AE to the CE as shown schematically in Fig. 6b and has been
reported in Ag/SiO 2 /W (see Fig. 6f) [105], Ag/ZrO 2 /Pt [106], Ag/a-Si/Pt [107], and
TiN/Al 2 O 3 /Cu [108].