Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 277 of the 290 references it cites were found in a registry or an index; 94 of the 273 sources its claims cite were checked for support.

Hardware implementation of memristor-based artificial neural networks

Open access

Abstract

Artificial Intelligence (AI) is currently experiencing a bloom driven by deep learning (DL) techniques, which rely on networks of connected simple computing units operating in parallel. The low communication bandwidth between memory and processing units in conventional von Neumann machines does not support the requirements of emerging applications that rely extensively on large sets of data. More recent computing paradigms, such as high parallelization and near-memory computing, help alleviate the data communication bottleneck to some extent, but paradigm- shifting concepts are required. Memristors, a novel beyond-complementary metal-oxide-semiconductor (CMOS) technology, are a promising choice for memory devices due to their unique intrinsic device-level properties, enabling both storing and computing with a small, massively-parallel footprint at low power. Theoretically, this directly translates to a major boost in energy efficiency and computational throughput, but various practical challenges remain. In this work we review the latest efforts for achieving hardware-based memristive artificial neural networks (ANNs), describing with detail the working principia of each block and the different design alternatives with their own advantages and disadvantages, as well as the tools required for accurate estimation of performance metrics. Ultimately, we aim to provide a comprehensive protocol of the materials and methods involved in memristive neural networks to those aiming to start working in this field and the experts looking for a holistic approach.

Overview

CoChat Summary

TL;DR

This paper is a review and practical protocol for hardware implementation of memristor-based artificial neural networks, with a focus on image-classification systems. It walks through the full stack from input encoding and crossbar VMM to training, peripheral circuits, simulation, and chip integration, while emphasizing device/circuit non-idealities and design trade-offs.

Key findings

  • Memristor crossbars are presented as the core hardware for vector-matrix multiplication, since VMM/MAC is the most demanding ANN operation and can be parallelized in crossbar arrays.
  • For MNIST-style benchmarking, the paper uses a 60,000-image training set and 10,000-image test set; in the worked example, images are downsized from 28 × 28 to 8 × 8 and an ANN with one 64 × 10 single-layer perceptron is trained.
  • In the example training study on 100 images, Scaled Conjugate Gradient and Levenberg–Marquardt achieved the highest accuracy, but Scaled Conjugate Gradient was chosen because it was faster.
  • The paper states that ADCs are often the largest and most power-hungry circuit block in memristive neural networks, consuming up to 70–90% of on-chip area and up to 80–88% of energy.

Methods

  • Review/protocol article: synthesizes prior work on memristive ANN hardware and organizes the system into blocks (image capture, input driving, VMM core, sensing, activation, softargmax, ADCs, and training/update).
  • Worked example: a single-layer perceptron for MNIST image classification; MATLAB is used to train the software model and generate a SPICE netlist, and SPICE simulation is used to evaluate the hybrid CMOS/memristor circuit during inference and write-verify programming.
  • Analysis emphasis: compares ex situ vs in situ training, discusses conductance mapping to memristors, and surveys simulation abstraction levels from neural-network-level tools to circuit/SPICE-level models.

Limitations

  • Stated in the paper: many published studies are not true full-hardware implementations; some only characterize a few devices or rely on software ANN simulations, so reported performance can be weakly connected to fabricated hardware.
  • Stated in the paper: full on-chip memristive ANN implementation is limited by peripheral circuit overhead, ADC/DAC trade-offs, device non-idealities, and the difficulty of integrating memristor stacks with CMOS.
  • Evident from the text: the paper is not an experimental validation of a new memristor ANN chip; it is primarily a review plus a simulation-based worked example, so its conclusions about performance are illustrative rather than a direct new hardware result.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41467-024-45670-9

Cites

277 of 290 references verified13 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Computing power demand increase and platform transition from Von-Neumann towards highly parallelized architectures. a The increase in computing power demands over the past four decades expressed in petaFLOPS per days. Until 2012, computing power demand doubled every 24 months; recently this has shortened to approximately every 2 months. The colour legend indicates different application domains10. Mehonic, A., Kenyon, A.J. Brain-inspired computing needs a master plan. Nature 604, 255–260 (2022), reproduced with permission from SNCSC. b A comparison of neural network accelerators for FPGA, ASIC, and GPU devices in terms of speed and power consumption. GOP/s giga operations per second, TOP/s tera operations per second.
Fig. 1 Computing power demand increase and platform transition from Von-Neumann towards highly parallelized architectures. a The increase in computing power demands over the past four decades expressed in petaFLOPS per days. Until 2012, computing power demand doubled every 24 months; recently this has shortened to approximately every 2 months. The colour legend indicates different application domains10. Mehonic, A., Kenyon, A.J. Brain-inspired computing needs a master plan. Nature 604, 255–260 (2022), reproduced with permission from SNCSC. b A comparison of neural network accelerators for FPGA, ASIC, and GPU devices in terms of speed and power consumption. GOP/s giga operations per second, TOP/s tera operations per second.CC BY · PubMed Central
Fig. 2 Generalized block diagram indicating the required circuital blocks to implement a memristive ANN for pattern classification. Green blocks (3, 5, 7 and 8) indicate the required mathematical operations (such as the VMM or activation functions). Red blocks (1, 2, 4, 6, 9, 11, 12, 13, 14, 15, 16) identify the required circuits for signal adaptation and/or conversion. The data path followed during the inference (or forward pass) is indicated by the red arrows/lines. The data path followed for in-situ training is indicated by the blue arrows/lines. The data path followed under ex-situ training is shown by the yellow arrows/lines. For each box, the upper (colored) part indicates the name of the function to realize by the circuital block, and the bottom part indicates the type of hardware required. The box titled successive neural layers would encompass multiple sub-blocks with a structure similar to the group titled First neural layer. 1S1R stands for 1Selector 1 Resistor while 1R stands for 1 Resistor. UART, SPI and I2C are well known communication standards. RISC stands for Reduced Instruction Set Computer.
Fig. 2 Generalized block diagram indicating the required circuital blocks to implement a memristive ANN for pattern classification. Green blocks (3, 5, 7 and 8) indicate the required mathematical operations (such as the VMM or activation functions). Red blocks (1, 2, 4, 6, 9, 11, 12, 13, 14, 15, 16) identify the required circuits for signal adaptation and/or conversion. The data path followed during the inference (or forward pass) is indicated by the red arrows/lines. The data path followed for in-situ training is indicated by the blue arrows/lines. The data path followed under ex-situ training is shown by the yellow arrows/lines. For each box, the upper (colored) part indicates the name of the function to realize by the circuital block, and the bottom part indicates the type of hardware required. The box titled successive neural layers would encompass multiple sub-blocks with a structure similar to the group titled First neural layer. 1S1R stands for 1Selector 1 Resistor while 1R stands for 1 Resistor. UART, SPI and I2C are well known communication standards. RISC stands for Reduced Instruction Set Computer.CC BY · PubMed Central
Fig. 3 Non-Von Neumann vector-matrix-multiplication (VMM) cores reported in the literature. a Full-CMOS SRAM (Static Random Access Memory) crossbar array, b Hybrid memristor/CMOS 1T1R crossbar array and c Full-memristive passive crossbar array. All cases assume a crossbar array integration structure which performs the Multiply-and-Accumulate (MAC) by exploiting the Kirchhoff’s law of currents. The use of memristors allows a smaller footprint per synapse as a lower number of smaller devices is employed. Passive crossbar arrays of memristors allow the highest possible integration density, yet they are still an immature technology with plenty of room for optimization. a290 Yamaoka, M. Low-Power SRAM. In: Kawahara, T., Mizuno, H. (eds) Green Computing with Emerging Memory. Springer, New York, NY (2013), reproduced with permission from SNCSC. b is adapted with permission under CC BY 4.0 license from ref. 54. c is adapted with permission under CC BY 4.0 license from ref. 93. F is the feature size of the litography and the energy estimation is on the cell-level. FEOL and BEOL stands for Front End Of Line and Back End Of Line, respectively.
Fig. 3 Non-Von Neumann vector-matrix-multiplication (VMM) cores reported in the literature. a Full-CMOS SRAM (Static Random Access Memory) crossbar array, b Hybrid memristor/CMOS 1T1R crossbar array and c Full-memristive passive crossbar array. All cases assume a crossbar array integration structure which performs the Multiply-and-Accumulate (MAC) by exploiting the Kirchhoff’s law of currents. The use of memristors allows a smaller footprint per synapse as a lower number of smaller devices is employed. Passive crossbar arrays of memristors allow the highest possible integration density, yet they are still an immature technology with plenty of room for optimization. a290 Yamaoka, M. Low-Power SRAM. In: Kawahara, T., Mizuno, H. (eds) Green Computing with Emerging Memory. Springer, New York, NY (2013), reproduced with permission from SNCSC. b is adapted with permission under CC BY 4.0 license from ref. 54. c is adapted with permission under CC BY 4.0 license from ref. 93. F is the feature size of the litography and the energy estimation is on the cell-level. FEOL and BEOL stands for Front End Of Line and Back End Of Line, respectively.CC BY · PubMed Central
Fig. 4 Example of a widely popular image database used for ANNs training and test, and how they are feed to the network. a Samples of the MNIST dataset of handwritten numeric digits considered in this article. In all cases images are represented in 28 × 28 px. Pixel brightness (or intensity) is codified in 256 levels ranging from 0 (fully OFF, black) to 1 (fully ON, white). b Readability loss as the resolution decreases from 28 × 28 pixels (case I) to 8 × 8 (case IV). c Schematic representation of the unrolling of the image pixels. Note that each of the n image columns of pixels are vertically concatenated to reach a n2 × 1 column vector. It is then scaled by VREAD to produce a vector of analogue voltages that is fed to the ANN.
Fig. 4 Example of a widely popular image database used for ANNs training and test, and how they are feed to the network. a Samples of the MNIST dataset of handwritten numeric digits considered in this article. In all cases images are represented in 28 × 28 px. Pixel brightness (or intensity) is codified in 256 levels ranging from 0 (fully OFF, black) to 1 (fully ON, white). b Readability loss as the resolution decreases from 28 × 28 pixels (case I) to 8 × 8 (case IV). c Schematic representation of the unrolling of the image pixels. Note that each of the n image columns of pixels are vertically concatenated to reach a n2 × 1 column vector. It is then scaled by VREAD to produce a vector of analogue voltages that is fed to the ANN.CC BY · PubMed Central
Fig. 5 Schematic diagrams of DAC circuits conventionally used in the literature to bias the rows of the memristive crossbar. a N-bit weighted Binary, b Current-steering DAC, c Memristive-DAC d N-bit R-2R DAC and e Pulse Width Modulation (PWM)-based DAC.
Fig. 5 Schematic diagrams of DAC circuits conventionally used in the literature to bias the rows of the memristive crossbar. a N-bit weighted Binary, b Current-steering DAC, c Memristive-DAC d N-bit R-2R DAC and e Pulse Width Modulation (PWM)-based DAC.CC BY · PubMed Central
Fig. 6 Memristor crossbar structure and electrical connection diagram for signed weights representation. a Sketch of the crossbar array structure. Red and blue arrows exemplify the electron flow through the memristors connecting the top (Word lines -WL-) and bottom lines (Bit lines -BL-). Different memristor resistance states are schematically represented (High Resistance State -HRS- to Low Resistance State -LRS-). The dashed blue line depicts the so-called sneak path problem. The parasitic wire resistance is indicated for WLi and BLi. Reproduced with permission under CCBY 4.0 license from ref. 253. b Equivalent circuit representation of the CPA sketched in a, showing the input voltages, output currents and TIA blocks that translates the output CPA current to a vector of analogue voltages. In this case the circuit was simplified by ignoring the line resistances. Finally, two different realizations of the memristive-based ANN synaptic layer are shown in c – unbalanced – and d – balanced –.
Fig. 6 Memristor crossbar structure and electrical connection diagram for signed weights representation. a Sketch of the crossbar array structure. Red and blue arrows exemplify the electron flow through the memristors connecting the top (Word lines -WL-) and bottom lines (Bit lines -BL-). Different memristor resistance states are schematically represented (High Resistance State -HRS- to Low Resistance State -LRS-). The dashed blue line depicts the so-called sneak path problem. The parasitic wire resistance is indicated for WLi and BLi. Reproduced with permission under CCBY 4.0 license from ref. 253. b Equivalent circuit representation of the CPA sketched in a, showing the input voltages, output currents and TIA blocks that translates the output CPA current to a vector of analogue voltages. In this case the circuit was simplified by ignoring the line resistances. Finally, two different realizations of the memristive-based ANN synaptic layer are shown in c – unbalanced – and d – balanced –.CC BY · PubMed Central
Fig. 7 Circuit schematics for the sensing electronics placed in at the output of every column of the memristive crossbar. In all cases, the goal is to translate a current signal into a voltage signal. a The sensing resistor is the simplest case, as it translates current into voltage directly by the Ohm’s law. b The use of a TIA allows to connect the crossbar columns to 0 volts and operate with lower output currents. As well as in the resistor-based approach, the current voltage conversion is linear when operating the TIA within its linear range and the output voltage signal is immediately available as soon as the output of the TIA settles. c For currents below the nano-ampere regime, charge integration is the most suitable option for current-voltage conversion. This can be achieved by using a capacitor. As such, the measurement is not instantaneous as a constant, controllable integration time is required before the measurement. d To minimize the area requirements of the integration capacitor, the use of a current divider allows to further reduce the current and, with it, the size of the required capacitor. The tradeoff in this case is with precision (mainly due to transistor mismatch) and output voltage dynamic range.
Fig. 7 Circuit schematics for the sensing electronics placed in at the output of every column of the memristive crossbar. In all cases, the goal is to translate a current signal into a voltage signal. a The sensing resistor is the simplest case, as it translates current into voltage directly by the Ohm’s law. b The use of a TIA allows to connect the crossbar columns to 0 volts and operate with lower output currents. As well as in the resistor-based approach, the current voltage conversion is linear when operating the TIA within its linear range and the output voltage signal is immediately available as soon as the output of the TIA settles. c For currents below the nano-ampere regime, charge integration is the most suitable option for current-voltage conversion. This can be achieved by using a capacitor. As such, the measurement is not instantaneous as a constant, controllable integration time is required before the measurement. d To minimize the area requirements of the integration capacitor, the use of a current divider allows to further reduce the current and, with it, the size of the required capacitor. The tradeoff in this case is with precision (mainly due to transistor mismatch) and output voltage dynamic range.CC BY · PubMed Central
Fig. 8 Equivalent electrical circuit of the topology used to implement the mathematical difference between two electrical signals. a Assuming that voltage inputs are unipolar (that is, only negative or positive), it is required to first transduce the current signals into voltage and then add an operational amplifier in a subtractor configuration. b If bipolar signals can be applied in the inputs, by biasing the negative synaptic weights with a voltage or opposite polarity, summing the resulting currents in a common node (Kirchhoff’s Law for Current) already solves the subtraction operation, and only one transimpedance amplifier is required per column.
Fig. 8 Equivalent electrical circuit of the topology used to implement the mathematical difference between two electrical signals. a Assuming that voltage inputs are unipolar (that is, only negative or positive), it is required to first transduce the current signals into voltage and then add an operational amplifier in a subtractor configuration. b If bipolar signals can be applied in the inputs, by biasing the negative synaptic weights with a voltage or opposite polarity, summing the resulting currents in a common node (Kirchhoff’s Law for Current) already solves the subtraction operation, and only one transimpedance amplifier is required per column.CC BY · PubMed Central
Fig. 9 Circuital implementations of the analogue activation functions used in memristive neural networks. Full-CMOS implementations of the a sigmoid and b ReLU activation functions. Aiming to minimize the area footprint of the activation function, c presents a ReLU implementation based on a VO2 Mott insulator device.
Fig. 9 Circuital implementations of the analogue activation functions used in memristive neural networks. Full-CMOS implementations of the a sigmoid and b ReLU activation functions. Aiming to minimize the area footprint of the activation function, c presents a ReLU implementation based on a VO2 Mott insulator device.CC BY · PubMed Central
Fig. 10 Analogue CMOS implementation of the Winner-Takes-All (WTA) function. a WTA CMOS block with voltage input291. The gate terminal of transistor Q5, and the source terminals of transistors Q6 and Q7 are common to all WTA cells. b WTA CMOS block with current input148. Node Vcom is common to all WTA cells. In both cases, the output voltage of the WTA cell with the highest input voltage/current is driven to the positive reference voltage (VDD), while the output voltage of the remaining WTA cells is driven to ground. The number of cells in the WTA module is the same to the number of classes of images to identify by the ANN.
Fig. 10 Analogue CMOS implementation of the Winner-Takes-All (WTA) function. a WTA CMOS block with voltage input291. The gate terminal of transistor Q5, and the source terminals of transistors Q6 and Q7 are common to all WTA cells. b WTA CMOS block with current input148. Node Vcom is common to all WTA cells. In both cases, the output voltage of the WTA cell with the highest input voltage/current is driven to the positive reference voltage (VDD), while the output voltage of the remaining WTA cells is driven to ground. The number of cells in the WTA module is the same to the number of classes of images to identify by the ANN.CC BY · PubMed Central
Fig. 11 Schematic diagrams of ADC circuits conventionally used in the literature. a SAR-ADC, b ΔΣ-ADC, c CCO-ADC, d VCO-based ADC and e Flash ADC.
Fig. 11 Schematic diagrams of ADC circuits conventionally used in the literature. a SAR-ADC, b ΔΣ-ADC, c CCO-ADC, d VCO-based ADC and e Flash ADC.CC BY · PubMed Central
Fig. 12 Basic concepts of neural network training. a Simplified organization of the most common terms reported in the literature, differentiating between gradient based and gradient free training tools. For the gradient-based tools, we propose an organization of the algorithms for (i) gradient computation, (ii) optimization and (iii) learning rate. b Illustration of the gradient descent method, for a trivial 2 × 1 neural network trained with supervised learning.
Fig. 12 Basic concepts of neural network training. a Simplified organization of the most common terms reported in the literature, differentiating between gradient based and gradient free training tools. For the gradient-based tools, we propose an organization of the algorithms for (i) gradient computation, (ii) optimization and (iii) learning rate. b Illustration of the gradient descent method, for a trivial 2 × 1 neural network trained with supervised learning.CC BY · PubMed Central
Fig. 13 k-fold cross validation with 10 repeats considering 11 different learning algorithms. 165,172–178 The accuracy obtained in each repeat is plotted against the CPU run-time of the learning algorithm when trained for the MNIST dataset for two different resolutions: a 8 × 8 and b 28 × 28 px. images. Although the Levenberg-Marquardt algorithm shows the higher mean accuracy, it is also the slowest to converge in our implementation, especially when considering large-size networks, as those required for classifying the 28 × 28 px. images. As a trade-off between accuracy and learning time, we have considered for the example to be described in later in this article, the Scaled Conjugate Gradient, as the accuracy difference with the Levenberg-Marquardt method is not statistically relevant: i.e., the observed difference might be due to a data fluctuation in the test dataset.
Fig. 13 k-fold cross validation with 10 repeats considering 11 different learning algorithms. 165,172–178 The accuracy obtained in each repeat is plotted against the CPU run-time of the learning algorithm when trained for the MNIST dataset for two different resolutions: a 8 × 8 and b 28 × 28 px. images. Although the Levenberg-Marquardt algorithm shows the higher mean accuracy, it is also the slowest to converge in our implementation, especially when considering large-size networks, as those required for classifying the 28 × 28 px. images. As a trade-off between accuracy and learning time, we have considered for the example to be described in later in this article, the Scaled Conjugate Gradient, as the accuracy difference with the Levenberg-Marquardt method is not statistically relevant: i.e., the observed difference might be due to a data fluctuation in the test dataset.CC BY · PubMed Central
Fig. 14 Typical figures-of-merit used to quantify the performance of ANNs intended for pattern recognition. In this case, they are plotted as a function of the training epochs. a Accuracy, b confusion matrix, c Loss function (cross-entropy), d Sensitivity, e Specificity, f Precision, g F1-score, h κ-coefficient.
Fig. 14 Typical figures-of-merit used to quantify the performance of ANNs intended for pattern recognition. In this case, they are plotted as a function of the training epochs. a Accuracy, b confusion matrix, c Loss function (cross-entropy), d Sensitivity, e Specificity, f Precision, g F1-score, h κ-coefficient.CC BY · PubMed Central
Fig. 15 Schematic representation of the trade-off between simulation speed and accuracy across the different tools reported in the literature for memristive ANNs evaluation. For each case, we list the main programming languages involved and some examples.
Fig. 15 Schematic representation of the trade-off between simulation speed and accuracy across the different tools reported in the literature for memristive ANNs evaluation. For each case, we list the main programming languages involved and some examples.CC BY · PubMed Central
Fig. 16 Detail of the different stages of the transaction level modelling, with the addition of the Neural Network and transistor (circuit) level simulation. Modelling approaches are arranged based on how accurately (untimed, approximate, cycle-accurate) the timing of the computation and communication aspects are captured. Transaction level models then expand from B to G, with B being the specification models (which uses considers the communication and computation to be untimed) and G the implementation models (which considers both cycle-accurate timing for both computation and communication). As we approach B, the model can be regarded as a System Level Simulation, while if it approaches G, it is regarded as an architecture-level simulation. Outside this group, we find those models simulated in Python or similar tools which focus on the network topology (A) and the circuital models which materializes the implementation models (G) in the transistor or register transfer level.
Fig. 16 Detail of the different stages of the transaction level modelling, with the addition of the Neural Network and transistor (circuit) level simulation. Modelling approaches are arranged based on how accurately (untimed, approximate, cycle-accurate) the timing of the computation and communication aspects are captured. Transaction level models then expand from B to G, with B being the specification models (which uses considers the communication and computation to be untimed) and G the implementation models (which considers both cycle-accurate timing for both computation and communication). As we approach B, the model can be regarded as a System Level Simulation, while if it approaches G, it is regarded as an architecture-level simulation. Outside this group, we find those models simulated in Python or similar tools which focus on the network topology (A) and the circuital models which materializes the implementation models (G) in the transistor or register transfer level.CC BY · PubMed Central
Fig. 17 Different representations of the Vector Matrix Multiplication operation typical from a synaptic layer. (a) Unitless mathematic VMM operation. (b) Mathematic VMM operation involving electrical magnitudes. (c) Electrical circuit representation of the memristive crossbar-based analogue VMM operation. (d) Realistic memristor crossbar representation considering the line resistance (RL) and the interline capacitances (see the inset showing a circuit schematic of a memristive cell in a CPA structure considering the associated wire parasitic resistance and capacitance). Aspects such as device variability are captured by the memristor model employed.
Fig. 17 Different representations of the Vector Matrix Multiplication operation typical from a synaptic layer. (a) Unitless mathematic VMM operation. (b) Mathematic VMM operation involving electrical magnitudes. (c) Electrical circuit representation of the memristive crossbar-based analogue VMM operation. (d) Realistic memristor crossbar representation considering the line resistance (RL) and the interline capacitances (see the inset showing a circuit schematic of a memristive cell in a CPA structure considering the associated wire parasitic resistance and capacitance). Aspects such as device variability are captured by the memristor model employed.CC BY · PubMed Central
Fig. 18 Connections schemes used to feed the CPA with the input pattern. a Single Side Connect (SSC) and (b) Dual Side Connect (DSC). On the SSC case, the input stimuli are applied only to the inputs of one side of the CPA, while the other is connected to high impedances (or remain disconnected). b In the DSC case, both terminals of a given wordline (horizontal lines in the CPA) are connected to the same input voltage, which thereby reduces the voltage drop along the wordlines.
Fig. 18 Connections schemes used to feed the CPA with the input pattern. a Single Side Connect (SSC) and (b) Dual Side Connect (DSC). On the SSC case, the input stimuli are applied only to the inputs of one side of the CPA, while the other is connected to high impedances (or remain disconnected). b In the DSC case, both terminals of a given wordline (horizontal lines in the CPA) are connected to the same input voltage, which thereby reduces the voltage drop along the wordlines.CC BY · PubMed Central
Fig. 19 Detail of the control circuits used for the dual inference/write procedures. a complete circuit schematic for a 4×8 1T1R crossbar array. b Detail of the synchronizers including the sense amplifiers used to detect the correct programming of a given memristor. c Address block, essentially a counter which sequentially addresses each memristor in the crossbar. d Row and column decoders, used to enable the memristor addressed by the address block. e Row and column driver, used to bias the rows with the voltage input or with the programming signal, and to connect the columns to the output neurons (during inference) or the sense amplifier (during write-verify).
Fig. 19 Detail of the control circuits used for the dual inference/write procedures. a complete circuit schematic for a 4×8 1T1R crossbar array. b Detail of the synchronizers including the sense amplifiers used to detect the correct programming of a given memristor. c Address block, essentially a counter which sequentially addresses each memristor in the crossbar. d Row and column decoders, used to enable the memristor addressed by the address block. e Row and column driver, used to bias the rows with the voltage input or with the programming signal, and to connect the columns to the output neurons (during inference) or the sense amplifier (during write-verify).CC BY · PubMed Central
Fig. 20 Schematic representation of the n2×1 column vectors of analogue voltages being fed to the SLP. 4 cases are represented: a–c corresponds to the correct classification of images from classes k, k+1 and k-1, respectively (for instance, in the case of the MNIST database, they might be images of the ‘5’, ‘6’ and ‘4’ digits). d Depicts the case of misclassification, as the highest current corresponds to the k+1 output for an image from class k.
Fig. 20 Schematic representation of the n2×1 column vectors of analogue voltages being fed to the SLP. 4 cases are represented: a–c corresponds to the correct classification of images from classes k, k+1 and k-1, respectively (for instance, in the case of the MNIST database, they might be images of the ‘5’, ‘6’ and ‘4’ digits). d Depicts the case of misclassification, as the highest current corresponds to the k+1 output for an image from class k.CC BY · PubMed Central
Fig. 21 Write-verify approach for conductance programming. a Schematic representation of the Write-Verify loop approach for programming the memristors in the CPA to a given conductance value. Reproduced with permission under CCBY 4.0 license from ref. 253. b Sensed output current for a SLP partition (one small CPA) during the programming phase in a Write-Verify loop procedure. The greater the peak, the higher the conductance level being programmed. The inset in the center shows a schematic representation of the current measured during the verify and write pulses, as well as the current target. The inset in the right of both panels shows a schematic of the equivalent circuit using during the verify phase. Reproduced with permission under CCBY 4.0 license from ref. 254.
Fig. 21 Write-verify approach for conductance programming. a Schematic representation of the Write-Verify loop approach for programming the memristors in the CPA to a given conductance value. Reproduced with permission under CCBY 4.0 license from ref. 253. b Sensed output current for a SLP partition (one small CPA) during the programming phase in a Write-Verify loop procedure. The greater the peak, the higher the conductance level being programmed. The inset in the center shows a schematic representation of the current measured during the verify and write pulses, as well as the current target. The inset in the right of both panels shows a schematic of the equivalent circuit using during the verify phase. Reproduced with permission under CCBY 4.0 license from ref. 254.CC BY · PubMed Central

Verification

Claims the paper makes

47 Valid · 14 Not valid · 58 Not checked · 121 Own results and interpretations

88 more claims · 22 with their cited sources checked

Create a free account to see all 240 claims and to verify the 1 reference still open.

See all claims — free account
  • Probably not a substantial claim. 79
  • It may state two separate claims as one. 91
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Citations

Cites

  • Deep Residual Learning for Image Recognition

    He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 1×

    “For benchmarking a device or a chip, it is essential to evaluate the accuracy of standard deep neural network models like VGG 89 and ResNet 90 on CIFAR and ImageNet dataset by utilizing architecture-levelsimulationandrealistichardwarestatistics 91,92 .”
  • Very Deep Convolutional Networks for Large-Scale Image Recognition

    Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148 · cited 1×

    “For benchmarking a device or a chip, it is essential to evaluate the accuracy of standard deep neural network models like VGG 89 and ResNet 90 on CIFAR and ImageNet dataset by utilizing architecture-levelsimulationandrealistichardwarestatistics 91,92 .”
  • Adam: A Method for Stochastic Optimization

    Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946 · cited 2×

    “…employed 165 , and for stochastic evaluation of the loss function using a mini-batch of images diverse methods have been employed, including Adaptive Gradient Algorithm (or AdaGrad) 182 , Root Mean Square Propagation (or RMSProp) 183 , Adaptive Moment Estimation (or Adam) 184 and Adadelta 185 .”
  • ImageNet: A large-scale hierarchical image database

    Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009) · 2009 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2009.5206848 · Cited by 65,333 · cited 1×

    “…86 ; 2) CIFAR (Canadian Institute for Advanced Research),whichcontains60,000colorimagesdividedinto10classes for CIFAR-10 and 100 classes for CIFAR-100 87 ; 3) ImageNet, one of the largest image datasets, which consists of over 1.2 million labelled from 1000 classes for the ImageNet competition 88 .”
    Influential
  • Particle swarm optimization

    Kennedy, J., & Eberhart, R. (2002) · Proceedings of ICNN'95 - International Conference on Neural Networks · doi:10.1109/icnn.1995.488968 · Cited by 58,192 · cited 1×

    “Gradient-free methods such as the Particle Swarm optimization 168 , Genetic Algorithms 169 and Simulated Annealing 170 algorithms are more demanding from a computational point of view, andhence,theyarerarelyemployedforANNtraining,bywhichtheylie beyond the scope of this article.”
  • Optimization by Simulated Annealing

    Kirkpatrick, S., Gelatt, C. D., & Vecchi, M. P. (1983) · Science · doi:10.1126/science.220.4598.671 · Cited by 45,134

  • Learning representations by back-propagating errors

    Rumelhart, D. E., Hinton, G. E., & Williams, R. J. (1986) · Nature · doi:10.1038/323533a0 · Cited by 31,892 · cited 1×

    “In most ANNs, the gradient of the loss function is normally computed by the backpropagation algorithm 171 .”
  • An Algorithm for Least-Squares Estimation of Nonlinear Parameters

    Marquardt, D. W. (1963) · Journal of the Society for Industrial and Applied Mathematics · doi:10.1137/0111030 · Cited by 30,728 · cited 6×

    “…evaluation, common optimization algorithms are the following: (i) Gradient Descent 165 (the simplest one and closest to the previous (cid:129) Manha(cid:3)an UpdateRule [181] (cid:129) Lavenberg-Marquardt[176] (cid:129) BFGS Quasi-Newton[172] Howdo we updatethe weightsduringeachitera(cid:2)on?”
    Influential
  • The missing memristor found

    Strukov, D. B., Snider, G. S., Stewart, D. R., & Williams, R. S. (2008) · Nature · doi:10.1038/nature06932 · Cited by 11,828

  • Untitled work

    (2017) · Cited by 11,625

Show all 282 references

Research works printed but not linked: 2

  1. “Microchips that mimic the human brain could make AI far more energy efficient | Science | AAAS.” Accessed: May 15, 2023. [Online]. Available: https://www.science.org/content/article/microchips-mimic-human-brain-could-make-ai-far-more-energy-efficient
  2. Ielmini, D. and Waser, R. Resistive Switching. Weinheim, Germany: Wiley-VCH Verlag GmbH & Co. KGaA, 2016.

Cited by

See all 564

Metadata

Authors
Aguirre, F., Sebastian, A., Le Gallo, M., Song, W., Wang, T., Yang, J. J., Lu, W., Chang, M.-F., Ielmini, D., Yang, Y., Mehonic, A., Kenyon, A., Villena, M. A., Roldán, J. B., Wu, Y., Hsu, H.-H., Raghavan, N., Suñé, J., Miranda, E., Eltawil, A., Setti, G., Smagulova, K., Salama, K. N., Krestinskaya, O., Yan, X., Ang, K.-W., Jain, S., Li, S., Alharbi, O., Pazos, S., & Lanza, M.
Affiliations
  • Universitat Autònoma de Barcelona (ES)
  • IBM Research - Zurich (CH)
  • University of Southern California (US)
  • University of Michigan (US)
  • National Tsing Hua University (TW)
  • Consorzio Nazionale Interuniversitario per la Nanoelettronica (IT)
  • Peking University (CN)
  • University College London (GB)
  • King Abdullah University of Science and Technology (SA)
  • Universidad de Granada (ES)
  • Singapore University of Technology and Design (SG)
  • Hebei University (CN)
  • National University of Singapore (SG)
Corresponding author
Lanza, Mario
Venue
Nature Communications
Published
2024-03-04
Volume
15
Issue
1
Pages
1974-1974
ISSN
2041-1723 (ISSN-L)
OpenAlex
W4392367648
Cited by
564
References
282
Citations by year
2026: 246 · 2025: 247 · 2024: 71
FWCI
11.53
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PORTO@Iris (submittedVersion, cc-by)
  • National University of Singapore (submittedVersion, cc-by-nc-nd)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
  • UCL Discovery (University College London) (submittedVersion)
  • Virtual Community of Pathological Anatomy (University of Castilla La Mancha) (submittedVersion)
  • Greater South Information System
Funders
  • King Abdullah University of Science and Technology
  • Universitat Autònoma de Barcelona
Topics
  • Physical Sciences › Engineering › Electrical and Electronic Engineering › Advanced Memory and Neural Computing
  • Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Applications
  • Physical Sciences › Engineering › Electrical and Electronic Engineering › Ferroelectric and Negative Capacitance Devices
Keywords
memristors, hardware implementation, in-memory computing, neuromorphic hardware, energy efficiency, deep learning hardware, beyond-CMOS computing, performance estimation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-09-29, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-09-29 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Hardware implementation of memristor-based artificial neural networks. https://app.cochat.ai/papers/10.1038/s41467-024-45670-9
The paper
Aguirre, F., Sebastian, A., Le Gallo, M., Song, W., Wang, T., Yang, J. J., Lu, W., Chang, M.-F., Ielmini, D., Yang, Y., Mehonic, A., Kenyon, A., Villena, M. A., Roldán, J. B., Wu, Y., Hsu, H.-H., Raghavan, N., Suñé, J., Miranda, E., . . . Lanza, M. (2024). Hardware implementation of memristor-based artificial neural networks. Nature Communications. https://doi.org/10.1038/s41467-024-45670-9

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.