Browse
Submit Data
Databases
API
Help

Dataset Information

26 Views

0 Connections

0 Citations

0 Reanalyses

0 Downloads

Omics score: 0

In situ Parallel Training of Analog Neural Network Using Electrochemical Random-Access Memory.

ABSTRACT: In-memory computing based on non-volatile resistive memory can significantly improve the energy efficiency of artificial neural networks. However, accurate in situ training has been challenging due to the nonlinear and stochastic switching of the resistive memory elements. One promising analog memory is the electrochemical random-access memory (ECRAM), also known as the redox transistor. Its low write currents and linear switching properties across hundreds of analog states enable accurate and massively parallel updates of a full crossbar array, which yield rapid and energy-efficient training. While simulations predict that ECRAM based neural networks achieve high training accuracy at significantly higher energy efficiency than digital implementations, these predictions have not been experimentally achieved. In this work, we train a 3 × 3 array of ECRAM devices that learns to discriminate several elementary logic gates (AND, OR, NAND). We record the evolution of the network's synaptic weights during parallel in situ (on-line) training, with outer product updates. Due to linear and reproducible device switching characteristics, our crossbar simulations not only accurately simulate the epochs to convergence, but also quantitatively capture the evolution of weights in individual devices. The implementation of the first in situ parallel training together with strong agreement with simulation results provides a significant advance toward developing ECRAM into larger crossbar arrays for artificial neural network accelerators, which could enable orders of magnitude improvements in energy efficiency of deep neural networks.

SUBMITTER: Li Y

PROVIDER: S-EPMC8060477 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Json Xml

Similar Datasets

Random access parallel microscopy.

Project description:We introduce a random-access parallel (RAP) imaging modality that uses a novel design inspired by a Newtonian telescope to image multiple spatially separated samples without moving parts or robotics. This scheme enables near-simultaneous image capture of multiple petri dishes and random-access imaging with sub-millisecond switching times at the full resolution of the camera. This enables the RAP system to capture long-duration records from different samples in parallel, which is not possible using conventional automated microscopes. The system is demonstrated by continuously imaging multiple cardiac monolayer and Caenorhabditis elegans preparations.

| S-EPMC7843131 | biostudies-literature

Purely antiferromagnetic magnetoelectric random access memory.

Project description:Magnetic random access memory schemes employing magnetoelectric coupling to write binary information promise outstanding energy efficiency. We propose and demonstrate a purely antiferromagnetic magnetoelectric random access memory (AF-MERAM) that offers a remarkable 50-fold reduction of the writing threshold compared with ferromagnet-based counterparts, is robust against magnetic disturbances and exhibits no ferromagnetic hysteresis losses. Using the magnetoelectric antiferromagnet Cr2O3, we demonstrate reliable isothermal switching via gate voltage pulses and all-electric readout at room temperature. As no ferromagnetic component is present in the system, the writing magnetic field does not need to be pulsed for readout, allowing permanent magnets to be used. Based on our prototypes, we construct a comprehensive model of the magnetoelectric selection mechanisms in thin films of magnetoelectric antiferromagnets, revealing misfit induced ferrimagnetism as an important factor. Beyond memory applications, the AF-MERAM concept introduces a general all-electric interface for antiferromagnets and should find wide applicability in antiferromagnetic spintronics.

| S-EPMC5216083 | biostudies-other

A pneumatic random-access memory for controlling soft robots.

Project description:Pneumatically-actuated soft robots have advantages over traditional rigid robots in many applications. In particular, their flexible bodies and gentle air-powered movements make them more suitable for use around humans and other objects that could be injured or damaged by traditional robots. However, existing systems for controlling soft robots currently require dedicated electromechanical hardware (usually solenoid valves) to maintain the actuation state (expanded or contracted) of each independent actuator. When combined with power, computation, and sensing components, this control hardware adds considerable cost, size, and power demands to the robot, thereby limiting the feasibility of soft robots in many important application areas. In this work, we introduce a pneumatic memory that uses air (not electricity) to set and maintain the states of large numbers of soft robotic actuators without dedicated electromechanical hardware. These pneumatic logic circuits use normally-closed microfluidic valves as transistor-like elements; this enables our circuits to support more complex computational functions than those built from normally-open valves. We demonstrate an eight-bit nonvolatile random-access pneumatic memory (RAM) that can maintain the states of multiple actuators, control both individual actuators and multiple actuators simultaneously using a pneumatic version of time division multiplexing (TDM), and set actuators to any intermediate position using a pneumatic version of analog-to-digital conversion. We perform proof-of-concept experimental testing of our pneumatic RAM by using it to control soft robotic hands playing individual notes, chords, and songs on a piano keyboard. By dramatically reducing the amount of hardware required to control multiple independent actuators in pneumatic soft robots, our pneumatic RAM can accelerate the spread of soft robotic technologies to a wide range of important application areas.

| S-EPMC8284813 | biostudies-literature

Switching operation and degradation of resistive random access memory composed of tungsten oxide and copper investigated using in-situ TEM.

Project description:In-situ transmission electron microscopy (in-situ TEM) was performed to investigate the switching operation of a resistive random access memory (ReRAM) made of copper, tungsten oxide and titanium nitride (Cu/WOx/TiN). In the first Set (Forming) operation to initialize the device, precipitation appeared inside the WOx layer. It was presumed that a Cu conducting filament was formed, lowering the resistance (on-state). The Reset operation induced a higher resistance (the off-state). No change in the microstructure was identified in the TEM images. Only when an additional Reset current was applied after switching to the off-state could erasure of the filament be seen (over-Reset). Therefore, it was concluded that structural change relating to the resistance switch was localized in a very small area around the filament. With repeated switching operations and increasing operational current, the WOx/electrode interfaces became indistinct. At the same time, the resistance of the off-state gradually decreased. This is thought to be caused by Cu condensation at the interfaces because of leakage current through the area other than through the filament. This will lead to device degradation through mechanisms such as endurance failure. This is the first accelerated aging test of ReRAM achieved using in-situ TEM.

| S-EPMC4661473 | biostudies-literature

In-Memory Logic Operations and Neuromorphic Computing in Non-Volatile Random Access Memory.

Project description:Recent progress in the development of artificial intelligence technologies, aided by deep learning algorithms, has led to an unprecedented revolution in neuromorphic circuits, bringing us ever closer to brain-like computers. However, the vast majority of advanced algorithms still have to run on conventional computers. Thus, their capacities are limited by what is known as the von-Neumann bottleneck, where the central processing unit for data computation and the main memory for data storage are separated. Emerging forms of non-volatile random access memory, such as ferroelectric random access memory, phase-change random access memory, magnetic random access memory, and resistive random access memory, are widely considered to offer the best prospect of circumventing the von-Neumann bottleneck. This is due to their ability to merge storage and computational operations, such as Boolean logic. This paper reviews the most common kinds of non-volatile random access memory and their physical principles, together with their relative pros and cons when compared with conventional CMOS-based circuits (Complementary Metal Oxide Semiconductor). Their potential application to Boolean logic computation is then considered in terms of their working mechanism, circuit design and performance metrics. The paper concludes by envisaging the prospects offered by non-volatile devices for future brain-inspired and neuromorphic computation.

| S-EPMC7475900 | biostudies-literature

Wave physics as an analog recurrent neural network.

Project description:Analog machine learning hardware platforms promise to be faster and more energy efficient than their digital counterparts. Wave physics, as found in acoustics and optics, is a natural candidate for building analog processors for time-varying signals. Here, we identify a mapping between the dynamics of wave physics and the computation in recurrent neural networks. This mapping indicates that physical wave systems can be trained to learn complex features in temporal data, using standard training techniques for neural networks. As a demonstration, we show that an inverse-designed inhomogeneous medium can perform vowel classification on raw audio signals as their waveforms scatter and propagate through it, achieving performance comparable to a standard digital implementation of a recurrent neural network. These findings pave the way for a new class of analog machine learning platforms, capable of fast and efficient processing of information in its native domain.

| S-EPMC6924985 | biostudies-literature

Short-Term Memory Dynamics of TiN/Ti/TiO2/SiOx/Si Resistive Random Access Memory.

Project description:In this study, we investigated the synaptic functions of TiN/Ti/TiO2/SiOx/Si resistive random access memory for a neuromorphic computing system that can act as a substitute for the von-Neumann computing architecture. To process the data efficiently, it is necessary to coordinate the information that needs to be processed with short-term memory. In neural networks, short-term memory can play the role of retaining the response on temporary spikes for information filtering. In this study, the proposed complementary metal-oxide-semiconductor (CMOS)-compatible synaptic device mimics the potentiation and depression with varying pulse conditions similar to biological synapses in the nervous system. Short-term memory dynamics are demonstrated through pulse modulation at a set pulse voltage of -3.5 V and pulse width of 10 ms and paired-pulsed facilitation. Moreover, spike-timing-dependent plasticity with the change in synaptic weight is performed by the time difference between the pre- and postsynaptic neurons. The SiOx layer as a tunnel barrier on a Si substrate provides highly nonlinear current-voltage (I-V) characteristics in a low-resistance state, which is suitable for high-density synapse arrays. The results herein presented confirm the viability of implementing a CMOS-compatible neuromorphic chip.

| S-EPMC7559005 | biostudies-literature

Achieving delafossite analog by in situ electrochemical self-reconstruction as an oxygen-evolving catalyst.

Project description:Development of novel and robust oxygen evolution reaction (OER) catalysts with well-modulated atomic and electronic structure remains a challenge. Compared to the well-known metal hydroxides or (oxyhydr)oxides with lamellar structure, delafossites (ABO2) are characterized by alternating layers of A cations and edge-sharing BO2 octahedra, but are rarely used in OER due to their poor electron conductivity and intrinsic activity. Here, we propose a delafossite analog by mutation of metal oxyhydroxide and delafossite based on first-principles calculations. Modulation on the electronic structure due to distortion of the original crystal field of the BO2 layers is calculated to enhance electron conductivity and catalytic activity. Inspired by the theoretical design, we have experimentally realized the delafossite analog by electrochemical self-reconstruction (ECSR). Operando X-ray absorption spectroscopy and other experimental techniques reveal the formation of delafossite analog with Ag intercalated into bimetallic cobalt-iron (oxyhydr)oxide layers from a metastable precursor through amorphization. Benefitting from the featured local electronic and geometric structures, the delafossite analog shows superior OER activity, affording a current density of 10 mA⋅cm-2 at an overpotential of 187 mV and an excellent stability (300 h) in alkaline conditions.

| S-EPMC7486770 | biostudies-literature

Modeling of electron conduction in contact resistive random access memory devices as random telegraph noise.

Project description:The intense development and study of resistive random access memory (RRAM) devices has opened a new era in semiconductor memory manufacturing. Resistive switching and carrier conduction inside RRAM films have become critical issues in recent years. Electron trapping/detrapping behavior is observed and investigated in the proposed contact resistive random access memory (CR-RAM) cell. Through the fitting of the space charge limiting current (SCLC) model, and analysis in terms of the random telegraph noise (RTN) model, the temperature-dependence of resistance levels and the high-temperature data retention behavior of the contact RRAM film are successfully and completely explained. Detail analyses of the electron capture and emission from the traps by forward and reverse read measurements provide further verifications for hopping conduction mechanism and current fluctuation discrepancies.

| S-EPMC3331868 | biostudies-other

Conductive-bridging random access memory: challenges and opportunity for 3D architecture.

Project description:The performances of conductive-bridging random access memory (CBRAM) have been reviewed for different switching materials such as chalcogenides, oxides, and bilayers in different structures. The structure consists of an inert electrode and one oxidized electrode of copper (Cu) or silver (Ag). The switching mechanism is the formation/dissolution of a metallic filament in the switching materials under external bias. However, the growth dynamics of the metallic filament in different switching materials are still debated. All CBRAM devices are switching under an operation current of 0.1 μA to 1 mA, and an operation voltage of ±2 V is also needed. The device can reach a low current of 5 pA; however, current compliance-dependent reliability is a challenging issue. Although a chalcogenide-based material has opportunity to have better endurance as compared to an oxide-based material, data retention and integration with the complementary metal-oxide-semiconductor (CMOS) process are also issues. Devices with bilayer switching materials show better resistive switching characteristics as compared to those with a single switching layer, especially a program/erase endurance of >10(5) cycles with a high speed of few nanoseconds. Multi-level cell operation is possible, but the stability of the high resistance state is also an important reliability concern. These devices show a good data retention of >10(5) s at >85°C. However, more study is needed to achieve a 10-year guarantee of data retention for non-volatile memory application. The crossbar memory is benefited for high density with low power operation. Some CBRAM devices as a chip have been reported for proto-typical production. This review shows that operation current should be optimized for few microamperes with a maintaining speed of few nanoseconds, which will have challenges and also opportunities for three-dimensional (3D) architecture.

| S-EPMC4412874 | biostudies-literature

OmicsDI is part of the ELIXIR infrastructure

OmicsDI is an Elixir interoperability service. Learn more ›

Tweets

OmicsDI Databases

PRIDE
PeptideAtlas
MassIVE
JPOST Repository
Physiome Model Repository

EGA
EVA
ENA
LINCS
PAXDB
Cell Collective

MetaboLights
Metabolomics Workbench
MetabolomeExpress
GNPS
BioModels
FAIRDOMHub

ArrayExpress
dbGaP
ExpressionAtlas
GEO
NODE

Information

Databases
Help
API
Contact us
Code on GitHub
Terms of use
Submit Data