Appendix 4: Memristor Applications
91
Memory Applications
The use of emerging memristor materials for advanced electrical devices such
as multi-valued logic is expected to outperform today’s binary logic digital technologies. While conventional memory cells can store only 1 bit, memristor-based
multi-bit cells can store more information within single device thus increasing
the information storage density. Such devices can potentially utilize the nonlinear
resistance of memristor bio-inspired materials for efficient information storage
[98]. Unlike other passive, two-terminal devices such as resistors or capacitors, the
hysteresis of a memristive device can be used for information storage (e.g., resistive
memory RRAM), with low resistance and reconfigurable signal routing (opening
and closing the connections between the nanowire electrodes). The memristor-based
crossbar network structure [99], being used as memory, can offer the following
advantages: (1) it allows ultra-high density memory storage with relatively small
number of control electrodes and cross points can be accessed by n-rows and ncolumns in the crossbar; (2) it offers large connectivity between devices; and (3)
it facilitates reconfigurable circuits by changing the conductance of the memristive
devices at selected cross points [51].
In-memory Computing
Conventional computers are based on Von Neumann architecture, where processing
and storing of the data are performed by different units (namely, CPU and memory).
Over the last few decades, the performance of processors has improved in a much
higher pace than that of memories, which has led to today several orders of
magnitude performance gap between processor and memory. This gap causes a
bottleneck for transferring data between memory and processor, which is usually
called the memory wall. One of today’s critical challenge is data-intensive and
big-data problems in data storage and analysis. The increase of the data size has
already surpassed the capabilities of today’s computation architectures, which suffer
from the limited bandwidth, programmability overhead, energy inefficiency, and
limited scalability. Memristor-based architectures for data-intensive applications
have been reported with the potential to solve data-intensive problems by increasing
the computation efficiency, solving the communication bottleneck and reducing the
leakage currents. Attempts for reducing the memory wall problem by getting the
memory closer to the processing unit have been tried before. One of the leading trials
is processing near memory (PNM) architecture. In PNM, processing units, on which
part of the program is executed, are added to the off-chip DRAM memory and a part
of the program is executed in the off-chip memory by a co-processor. One wellknown implementation of PNM is Berkeley’s Intelligent RAM (IRAM) project.
The IRAM project had not become commercially successful, mainly due to the
bad integration between DRAM and logic technologies. Processing within memory
91
Memory Applications
The use of emerging memristor materials for advanced electrical devices such
as multi-valued logic is expected to outperform today’s binary logic digital technologies. While conventional memory cells can store only 1 bit, memristor-based
multi-bit cells can store more information within single device thus increasing
the information storage density. Such devices can potentially utilize the nonlinear
resistance of memristor bio-inspired materials for efficient information storage
[98]. Unlike other passive, two-terminal devices such as resistors or capacitors, the
hysteresis of a memristive device can be used for information storage (e.g., resistive
memory RRAM), with low resistance and reconfigurable signal routing (opening
and closing the connections between the nanowire electrodes). The memristor-based
crossbar network structure [99], being used as memory, can offer the following
advantages: (1) it allows ultra-high density memory storage with relatively small
number of control electrodes and cross points can be accessed by n-rows and ncolumns in the crossbar; (2) it offers large connectivity between devices; and (3)
it facilitates reconfigurable circuits by changing the conductance of the memristive
devices at selected cross points [51].
In-memory Computing
Conventional computers are based on Von Neumann architecture, where processing
and storing of the data are performed by different units (namely, CPU and memory).
Over the last few decades, the performance of processors has improved in a much
higher pace than that of memories, which has led to today several orders of
magnitude performance gap between processor and memory. This gap causes a
bottleneck for transferring data between memory and processor, which is usually
called the memory wall. One of today’s critical challenge is data-intensive and
big-data problems in data storage and analysis. The increase of the data size has
already surpassed the capabilities of today’s computation architectures, which suffer
from the limited bandwidth, programmability overhead, energy inefficiency, and
limited scalability. Memristor-based architectures for data-intensive applications
have been reported with the potential to solve data-intensive problems by increasing
the computation efficiency, solving the communication bottleneck and reducing the
leakage currents. Attempts for reducing the memory wall problem by getting the
memory closer to the processing unit have been tried before. One of the leading trials
is processing near memory (PNM) architecture. In PNM, processing units, on which
part of the program is executed, are added to the off-chip DRAM memory and a part
of the program is executed in the off-chip memory by a co-processor. One wellknown implementation of PNM is Berkeley’s Intelligent RAM (IRAM) project.
The IRAM project had not become commercially successful, mainly due to the
bad integration between DRAM and logic technologies. Processing within memory