Memristive devices are electrical resistance switches that can retain a state of internal resistance based on the history of applied voltage and current. These devices can store and process information, and offer several key performance characteristics that exceed conventional integrated circuit technology. An important class of memristive devices are two-terminal resistance switches based on ionic motion, which are built from a simple conductor/insulator/conductor thin-film stack. These devices were originally conceived in the late 1960s and recent progress has led to fast, low-energy, high-endurance devices that can be scaled down to less than 10 nm and stacked in three dimensions. However, the underlying device mechanisms remain unclear, which is a significant barrier to their widespread application. Here, we review recent progress in the development and understanding of memristive devices. We also examine the performance requirements for computing with memristive devices and detail how the outstanding challenges could be met.

Memristive devices for computing

In this paper, recent progress of binary metal-oxide resistive switching random access memory (RRAM) is reviewed. The physical mechanism, material properties, and electrical characteristics of a variety of binary metal-oxide RRAM are discussed, with a focus on the use of RRAM for nonvolatile memory application. A review of recent development of large-scale RRAM arrays is given. Issues such as uniformity, endurance, retention, multibit operation, and scaling trends are discussed.

Metal–Oxide RRAM

Despite much progress in semiconductor integrated circuit technology, the extreme complexity of the human cerebral cortex, with its approximately 10(14) synapses, makes the hardware implementation of neuromorphic networks with a comparable number of devices exceptionally challenging. To provide comparable complexity while operating much faster and with manageable power dissipation, networks based on circuits combining complementary metal-oxide-semiconductors (CMOSs) and adjustable two-terminal resistive devices (memristors) have been developed. In such circuits, the usual CMOS stack is augmented with one or several crossbar layers, with memristors at each crosspoint. There have recently been notable improvements in the fabrication of such memristive crossbars and their integration with CMOS circuits, including first demonstrations of their vertical integration. Separately, discrete memristors have been used as artificial synapses in neuromorphic networks. Very recently, such experiments have been extended to crossbar arrays of phase-change memristive devices. The adjustment of such devices, however, requires an additional transistor at each crosspoint, and hence these devices are much harder to scale than metal-oxide memristors, whose nonlinear current-voltage curves enable transistor-free operation. Here we report the experimental implementation of transistor-free metal-oxide memristor crossbars, with device variability sufficiently low to allow operation of integrated neural networks, in a simple network: a single-layer perceptron (an algorithm for linear classification). The network can be taught in situ using a coarse-grain variety of the delta rule algorithm to perform the perfect classification of 3 × 3-pixel black/white images into three classes (representing letters). This demonstration is an important step towards much larger and more complex memristive neuromorphic networks.

Training and operation of an integrated neuromorphic network based on metal-oxide memristors

Numerous candidates attempting to replace Si-based flash memory have failed for a variety of reasons over the years. Oxide-based resistance memory and the related memristor have succeeded in surpassing the specifications for a number of device requirements. However, a material or device structure that satisfies high-density, switching-speed, endurance, retention and most importantly power-consumption criteria has yet to be announced. In this work we demonstrate a TaO(x)-based asymmetric passive switching device with which we were able to localize resistance switching and satisfy all aforementioned requirements. In particular, the reduction of switching current drastically reduces power consumption and results in extreme cycling endurances of over 10(12). Along with the 10 ns switching times, this allows for possible applications to the working-memory space as well. Furthermore, by combining two such devices each with an intrinsic Schottky barrier we eliminate any need for a discrete transistor or diode in solving issues of stray leakage current paths in high-density crossbar arrays.

A fast, high-endurance and scalable non-volatile memory device made from asymmetric Ta2O5−x/TaO2−x bilayer structures

Processing-in-memory (PIM) is a promising solution to address the "memory wall" challenges for future computer systems. Prior proposed PIM architectures put additional computation logic in or near memory. The emerging metal-oxide resistive random access memory (ReRAM) has showed its potential to be used for main memory. Moreover, with its crossbar array structure, ReRAM can perform matrix-vector multiplication efficiently, and has been widely studied to accelerate neural network (NN) applications. In this work, we propose a novel PIM architecture, called PRIME, to accelerate NN applications in ReRAM based main memory. In PRIME, a portion of ReRAM crossbar arrays can be configured as accelerators for NN applications or as normal memory for a larger memory space. We provide microarchitecture and circuit designs to enable the morphable functions with an insignificant area overhead. We also design a software/hardware interface for software developers to implement various NNs on PRIME. Benefiting from both the PIM architecture and the efficiency of using ReRAM for NN computation, PRIME distinguishes itself from prior work on NN acceleration, with significant performance improvement and energy saving. Our experimental results show that, compared with a state-of-the-art neural processing unit design, PRIME improves the performance by ~2360× and the energy consumption by ~895×, across the evaluated machine learning benchmarks.

PRIME: a novel processing-in-memory architecture for neural network computation in ReRAM-based main memory

Highly reliable TaOx ReRAM has been successfully demonstrated. The memory cell shows stable pulse switching with endurance over 109 cycles, sufficient retention exceeding 10 years at 85degC. TaOx exhibits stable high and low resistance states based on the redox reaction mechanism, confirmed by HX-PES directly for the first time. An 8 kbit 1T1R memory array with a good operating window has been fabricated using the standard 0.18 mum CMOS process.

Highly reliable TaOx ReRAM and direct evidence of redox reaction mechanism

Nonvolatile memories with fast write operation at low voltage are required as storage devices to exceed flash memory performance [1–3]. We develop an 8Mb multi-layered cross-point ReRAM macro with 443MB/s write throughput (64b parallel write per 17.2ns cycle), which is almost twice as fast as existing methods, using the fast-switching performance of TaO x ReRAM [4] and the following three techniques to reduce the sneak current in bipolar type cross-point cell array structure in an 0.18μm process. First, memory cell and array technologies reduce the sneak current with a newly developed bidirectional diode as a memory cell select element for the first time. Second, we use a hierarchical bitline (BL) structure for multi-layered cross-point memory with fast and stable current control. Third, we implement a multi-bit write architecture that realizes fast write operation and suppresses sneak current. This work is applicable to both high-density stand-alone and embedded memory with more stacked memory arrays and/or scaling memory cells.

An 8 Mb Multi-Layered Cross-Point ReRAM Macro With 443 MB/s Write Throughput

A nonvolatile memory element including a resistance variable element configured to reversibly change between a low-resistance state and a high-resistance state in response to electric signals with different polarities; and a current controlling element configured such that when a current flowing when a voltage whose absolute value is a first value which is larger than 0 and smaller than a predetermined voltage value and whose polarity is a first polarity is applied is a first current and a current flowing when a voltage whose absolute value is the first value and whose polarity is a second polarity is applied is a second current, the first current is higher than the second current, and the resistance variable element is connected with the current controlling element such that the first polarity voltage is applied to the current controlling element when the resistance variable element changes from the low-resistance to the high-resistance state.

Nonvolatile memory element, and nonvolatile memory device

A cross point nonvolatile memory device capable of suppressing sneak-current-caused reduction in sensitivity of detection of a resistance value of a memory element is provided. The device includes perpendicular bit and word lines; a cross-point cell array including memory cells each having a resistance value reversibly changing between at least two resistance states according to electrical signals, arranged on cross-points of the word and bit lines; an offset detection cell array including an offset detection cell having a resistance higher than that of the memory cell in a high resistance state, the word lines being shared by the offset detection cell array; a read circuit (a sense amplifier) that determines a resistance state of a selected memory cell based on a current through the selected bit line; and a current source which supplies current to the offset detection cell array in a read operation period.

Nonvolatile semiconductor memory device and read method for the same

A non-volatile semiconductor memory device comprises a plurality of memory cell holes ( 101 ) formed through an interlayer insulating layer ( 80 ) at respective cross-points of a plurality of first wires ( 10 ) of a stripe shape and a plurality of second wires ( 20 ) of a stripe shape when viewed from above such that the memory cell holes ( 101 ) expose upper surfaces of the plurality of first wires, respectively, a plurality of dummy holes ( 111 ) formed on the plurality of first wires in the interlayer insulating layer such that the dummy holes reach the upper surfaces of the plurality of first wires, respectively, and stacked-layer structures formed inside the memory cell holes and inside the dummy holes, respectively, each of the stacked-layer structures including a first electrode ( 30 ) and a variable resistance layer ( 40 ); an area of a portion of the first wire which is exposed in a lower opening of one of the dummy holes being greater than an area of a portion of the first wire which is exposed in a lower opening of one of the memory cell holes; and one or more of the dummy holes being formed on each of the first wires.

Kiyotaka Tsuji

Papers

Highly reliable TaOx ReRAM and direct evidence of redox reaction mechanism

An 8 Mb Multi-Layered Cross-Point ReRAM Macro With 443 MB/s Write Throughput

Nonvolatile memory element, and nonvolatile memory device

Nonvolatile semiconductor memory device and read method for the same

Non-volatile semiconductor memory device and manufacturing method thereof