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

The DOI matches its Crossref record; 64 of the 65 references it cites were found in a registry or an index; 15 of the 35 sources its claims cite were checked for support.

A compute-in-memory chip based on resistive random-access memory

Open access

Abstract

Abstract Realizing increasingly complex artificial intelligence (AI) functionalities directly on edge devices calls for unprecedented energy efficiency of edge hardware. Compute-in-memory (CIM) based on resistive random-access memory (RRAM) 1 promises to meet such demand by storing AI model weights in dense, analogue and non-volatile RRAM devices, and by performing AI computation directly within RRAM, thus eliminating power-hungry data movement between separate compute and memory 2–5 . Although recent studies have demonstrated in-memory matrix-vector multiplication on fully integrated RRAM-CIM hardware 6–17 , it remains a goal for a RRAM-CIM chip to simultaneously deliver high energy efficiency, versatility to support diverse models and software-comparable accuracy. Although efficiency, versatility and accuracy are all indispensable for broad adoption of the technology, the inter-related trade-offs among them cannot be addressed by isolated improvements on any single abstraction level of the design. Here, by co-optimizing across all hierarchies of the design from algorithms and architecture to circuits and devices, we present NeuRRAM—a RRAM-based CIM chip that simultaneously delivers versatility in reconfiguring CIM cores for diverse model architectures, energy efficiency that is two-times better than previous state-of-the-art RRAM-CIM chips across various computational bit-precisions, and inference accuracy comparable to software models quantized to four-bit weights across various AI tasks, including accuracy of 99.0 percent on MNIST 18 and 85.7 percent on CIFAR-10 19 image classification, 84.7-percent accuracy on Google speech command recognition 20 , and a 70-percent reduction in image-reconstruction error on a Bayesian image-recovery task.

Overview

CoChat Summary

TL;DR

NeuRRAM is a 48-core RRAM compute-in-memory chip that co-optimizes device, circuit, architecture, system, and algorithm choices to support reconfigurable AI inference with hardware-measured accuracy. The paper reports better energy-delay product than prior RRAM-CIM chips and benchmark results of 99.0% on MNIST, 85.7% on CIFAR-10, 84.7% on Google speech commands, and a 70% reduction in image-reconstruction error on MNIST image recovery.

Key findings

  • The chip achieves an energy-delay product that is 1.6× to 2.3× lower than previous state-of-the-art RRAM-CIM chips, and 7× to 13× higher computational density at various MVM bit-precisions.
  • Fully hardware-measured benchmark results include 99.0% MNIST accuracy, 85.7% CIFAR-10 accuracy, 84.7% Google speech command accuracy, and 70% lower image-reconstruction error on the Bayesian image-recovery task.
  • Hardware-algorithm co-optimization improved CIFAR-10 classification from 25.34% accuracy without noise injection to 85.99% with noise injection; chip-in-the-loop progressive fine-tuning improved cumulative CIFAR-10 test-set accuracy by 1.99%.
  • The authors report that the measured CIFAR-10 accuracy for deep ResNet-20 inference on-chip was 83.67%, which remained 3.36% below the 87.03% of the corresponding 4-bit-weight software model.

Methods

  • Built a 48-core NeuRRAM chip with 3 million RRAM devices in a 130-nm CMOS process, using a bidirectional transposable neurosynaptic array (TNSA) and a voltage-mode neuron circuit.
  • Evaluated the chip on multiple AI tasks: MNIST and CIFAR-10 image classification, Google speech command recognition, and MNIST image recovery with CNNs, LSTMs, and an RBM; all benchmark results were obtained from hardware measurements on complete datasets.
  • Used model calibration, noise-resilient training, iterative write–verify programming, and chip-in-the-loop progressive fine-tuning; weights were mapped across cores with strategies for balancing load and mitigating voltage drop.

Limitations

  • The paper states that the reported throughput and energy-efficiency numbers are peak values at 100% array utilization and do not include energy/time for buffering or moving intermediate data; this is an explicit limitation of the measurement scope.
  • The paper notes that some on-chip accuracies are still below full-precision digital implementations, mainly because of ≤4-bit quantization of inputs and activations, especially sensitive input and output layers.
  • Evident from the text: the implementation relies on an FPGA board for partial-sum accumulation, average pooling, buffers, and other operations, so the system is not a fully self-contained chip-level AI system in the demonstrated setup.

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

Integrity signals

Registry

Crossref · 10.1038/s41586-022-04992-8

Cites

64 of 64 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Design methodology and main contributions of the NeuRRAM chip. a, Cross-layer co-optimizations across the full stack of the design enable NeuRRAM to simultaneously deliver high versatility, computational efficiency and software-comparable inference accuracy. b, Micrograph of the NeuRRAM chip. c, Reconfigurability in various aspects of the design enables NeuRRAM to implement diverse AI models for a wide variety of applications. d, Comparison of EDP, a commonly used energy-efficiency and performance metric among recent RRAM-based CIM hardware. e, Fully hardware-measured inference accuracy on NeuRRAM is comparable to software models quantized to 4-bit weights across various AI benchmarks.
Fig. 1 Design methodology and main contributions of the NeuRRAM chip. a, Cross-layer co-optimizations across the full stack of the design enable NeuRRAM to simultaneously deliver high versatility, computational efficiency and software-comparable inference accuracy. b, Micrograph of the NeuRRAM chip. c, Reconfigurability in various aspects of the design enables NeuRRAM to implement diverse AI models for a wide variety of applications. d, Comparison of EDP, a commonly used energy-efficiency and performance metric among recent RRAM-based CIM hardware. e, Fully hardware-measured inference accuracy on NeuRRAM is comparable to software models quantized to 4-bit weights across various AI benchmarks.CC BY · PubMed Central
Fig. 2 Reconfigurable architecture of the NeuRRAM chip. a, Multi-core architecture of the NeuRRAM chip, and various ways, labelled (1) to (6), to map neural-network layers onto CIM cores. b, Zoomed-in chip micrograph on a single CIM core. c, A cross-sectional transmission electron microscopy image showing the layer stack of the monolithically integrated RRAM and CMOS. d, Block diagram of a CIM core. A core consists of a TNSA, drivers for BLs, WLs, and SLs, registers that store MVM inputs and outputs, a LFSR pseudo-random number generator (PRNG), and a controller. During the MVM input stage, the drivers convert register inputs (REG) and PRNG inputs (PRN) to analogue voltages and send them to TNSA; during the MVM output stage, the drivers pass digital outputs from neurons back to registers through REG. e, The architecture of a TNSA consists of 16 × 16 corelets with interleaving RRAM weights and CMOS neurons. Each neuron integrates inputs from 256 RRAMs connecting to the same horizontal BL or vertical SL. f, Each corelet contains 16 × 16 RRAMs and 1 neuron. The neuron connects to 1 of the 16 BLs and 1 of the 16 SLs that pass through the corelet, and can use a BL and a SL for both its input and output. g, The TNSA can be dynamically configured for MVM in forwards, backwards or recurrent directions. h, Differential input and differential output schemes used to implement real-valued weights during forwards and backwards MVMs. Weights are encoded as the differential conductance between two RRAM cells on adjacent rows (G+ and G-).
Fig. 2 Reconfigurable architecture of the NeuRRAM chip. a, Multi-core architecture of the NeuRRAM chip, and various ways, labelled (1) to (6), to map neural-network layers onto CIM cores. b, Zoomed-in chip micrograph on a single CIM core. c, A cross-sectional transmission electron microscopy image showing the layer stack of the monolithically integrated RRAM and CMOS. d, Block diagram of a CIM core. A core consists of a TNSA, drivers for BLs, WLs, and SLs, registers that store MVM inputs and outputs, a LFSR pseudo-random number generator (PRNG), and a controller. During the MVM input stage, the drivers convert register inputs (REG) and PRNG inputs (PRN) to analogue voltages and send them to TNSA; during the MVM output stage, the drivers pass digital outputs from neurons back to registers through REG. e, The architecture of a TNSA consists of 16 × 16 corelets with interleaving RRAM weights and CMOS neurons. Each neuron integrates inputs from 256 RRAMs connecting to the same horizontal BL or vertical SL. f, Each corelet contains 16 × 16 RRAMs and 1 neuron. The neuron connects to 1 of the 16 BLs and 1 of the 16 SLs that pass through the corelet, and can use a BL and a SL for both its input and output. g, The TNSA can be dynamically configured for MVM in forwards, backwards or recurrent directions. h, Differential input and differential output schemes used to implement real-valued weights during forwards and backwards MVMs. Weights are encoded as the differential conductance between two RRAM cells on adjacent rows (G+ and G-).CC BY · PubMed Central
Fig. 3 Voltage-mode MVM with multi-bit inputs and outputs. a, Conventional current-mode-sensing scheme needs to activate a small fraction of total N rows each cycle to limit total current ISL and time-multiplex ADCs across multiple columns to amortize ADC area, thus limiting its computational parallelism. b, Voltage-mode sensing employed by NeuRRAM can activate all the rows and all the columns in a single cycle, enabling higher parallelism. c, MVM output distribution from a CNN layer and from an LSTM layer (weights normalized to the same range). Voltage-mode sensing intrinsically normalizes wide variation in output dynamic range. d, Schematic of the voltage-mode neuron circuit, where BLsel, SLsel, Sample, Integ, Reset, Latch, Decr, and WR are digital signals controlling state of the switches. e, Sample waveforms to perform MVM and 4-bit signed inputs digital-to-analogue conversion. WLs are pulsed once per magnitude-bit; sampling and integration are performed 2n−1 times for the nth LSB. f, Two-phase MVM: for input precision greater than 4 bits, inputs are divided into a MSB segment and a LSB segment. MVMs and ADCs are performed separately for each segment, followed by a shift-and-add to obtain final outputs. g, Sample waveforms to perform 5-bit signed outputs analogue-to-digital conversion. The sign-bit is first generated by a comparison operation. The magnitude-bits are generated through a binary search process realized by adding/subtracting charge on Cinteg. From MSB to LSB, added/subtracted charge is halved every bit. h, Chip-measured 64 × 64 MVM outputs versus ideal outputs under 4-bit input and 6-bit output.
Fig. 3 Voltage-mode MVM with multi-bit inputs and outputs. a, Conventional current-mode-sensing scheme needs to activate a small fraction of total N rows each cycle to limit total current ISL and time-multiplex ADCs across multiple columns to amortize ADC area, thus limiting its computational parallelism. b, Voltage-mode sensing employed by NeuRRAM can activate all the rows and all the columns in a single cycle, enabling higher parallelism. c, MVM output distribution from a CNN layer and from an LSTM layer (weights normalized to the same range). Voltage-mode sensing intrinsically normalizes wide variation in output dynamic range. d, Schematic of the voltage-mode neuron circuit, where BLsel, SLsel, Sample, Integ, Reset, Latch, Decr, and WR are digital signals controlling state of the switches. e, Sample waveforms to perform MVM and 4-bit signed inputs digital-to-analogue conversion. WLs are pulsed once per magnitude-bit; sampling and integration are performed 2n−1 times for the nth LSB. f, Two-phase MVM: for input precision greater than 4 bits, inputs are divided into a MSB segment and a LSB segment. MVMs and ADCs are performed separately for each segment, followed by a shift-and-add to obtain final outputs. g, Sample waveforms to perform 5-bit signed outputs analogue-to-digital conversion. The sign-bit is first generated by a comparison operation. The magnitude-bits are generated through a binary search process realized by adding/subtracting charge on Cinteg. From MSB to LSB, added/subtracted charge is halved every bit. h, Chip-measured 64 × 64 MVM outputs versus ideal outputs under 4-bit input and 6-bit output.CC BY · PubMed Central
Fig. 4 Hardware-algorithm co-optimization techniques to improve NeuRRAM inference accuracy. a, Various device and circuit non-idealities (labelled (1) to (7)) of in-memory MVM. b, Model-driven chip calibration technique to search for optimal chip operating conditions and record offsets for subsequent cancellation. c, Noise-resilient neural-network training technique to train the model with noise injection. The noise distribution is obtained from hardware characterization. The trained weights are programmed to the continuous analogue conductance of RRAMs without quantization as shown by the continuous diagonal band at the bottom. d, Chip-in-the-loop progressive fine-tuning technique: weights are progressively mapped onto the chip one layer at a time. The hardware-measured outputs from layer n are used as inputs to fine-tune the remaining layers n + 1 to N.
Fig. 4 Hardware-algorithm co-optimization techniques to improve NeuRRAM inference accuracy. a, Various device and circuit non-idealities (labelled (1) to (7)) of in-memory MVM. b, Model-driven chip calibration technique to search for optimal chip operating conditions and record offsets for subsequent cancellation. c, Noise-resilient neural-network training technique to train the model with noise injection. The noise distribution is obtained from hardware characterization. The trained weights are programmed to the continuous analogue conductance of RRAMs without quantization as shown by the continuous diagonal band at the bottom. d, Chip-in-the-loop progressive fine-tuning technique: weights are progressively mapped onto the chip one layer at a time. The hardware-measured outputs from layer n are used as inputs to fine-tune the remaining layers n + 1 to N.CC BY · PubMed Central
Fig. 5 Measured results showing the efficacy of the hardware-algorithm co-optimization techniques. a, Simulated (blue) and measured (red) CIFAR-10 test-set classification accuracies. b, CIFAR-10 classification accuracy at various time steps of chip-in-the-loop fine-tuning. From left to right, each data point represents a new layer (Conv0 to Dense) programmed onto the chip. The accuracy at a layer is evaluated by using the hardware-measured outputs from that layer as inputs to the remaining layers that are simulated in software. Two curves compare the test-set inference accuracy with and without applying fine-tuning during training. c, RBM-based image recovery on noisy images (top) and partially occluded images measured on NeuRRAM (bottom).
Fig. 5 Measured results showing the efficacy of the hardware-algorithm co-optimization techniques. a, Simulated (blue) and measured (red) CIFAR-10 test-set classification accuracies. b, CIFAR-10 classification accuracy at various time steps of chip-in-the-loop fine-tuning. From left to right, each data point represents a new layer (Conv0 to Dense) programmed onto the chip. The accuracy at a layer is evaluated by using the hardware-measured outputs from that layer as inputs to the remaining layers that are simulated in software. Two curves compare the test-set inference accuracy with and without applying fine-tuning during training. c, RBM-based image recovery on noisy images (top) and partially occluded images measured on NeuRRAM (bottom).CC BY · PubMed Central
Extended Data Fig. 1 Peripheral driver circuits for TNSA and chip operating modes. a, driver circuits’ configuration under the weight-programming mode. b, under the neuron-testing mode. c, under the MVM mode. d, circuit diagram of the two counter-propagating LFSR chains XORed to generate pseudo-random sequences for probabilistic sampling.
Extended Data Fig. 1 Peripheral driver circuits for TNSA and chip operating modes. a, driver circuits’ configuration under the weight-programming mode. b, under the neuron-testing mode. c, under the MVM mode. d, circuit diagram of the two counter-propagating LFSR chains XORed to generate pseudo-random sequences for probabilistic sampling.CC BY · PubMed Central
Extended Data Fig. 2 Various MVM dataflow directions and their CIM implementations. Left, various MVM dataflow directions commonly seen in different AI models. Middle, conventional CIM implementation of various dataflow directions. Conventional designs typically locate all peripheral circuits such as ADCs outside of RRAM array. The resulting implementations of bidirectional and recurrent MVMs incur overheads in area, latency, and energy. Right, the Transposable Neurosynaptic Array (TNSA) interleaves RRAM weights and CMOS neurons across the array and supports diverse MVM directions with minimal overhead.
Extended Data Fig. 2 Various MVM dataflow directions and their CIM implementations. Left, various MVM dataflow directions commonly seen in different AI models. Middle, conventional CIM implementation of various dataflow directions. Conventional designs typically locate all peripheral circuits such as ADCs outside of RRAM array. The resulting implementations of bidirectional and recurrent MVMs incur overheads in area, latency, and energy. Right, the Transposable Neurosynaptic Array (TNSA) interleaves RRAM weights and CMOS neurons across the array and supports diverse MVM directions with minimal overhead.CC BY · PubMed Central
Extended Data Fig. 3 Iterative write–verify RRAM programming. a, Flowchart of the incremental-pulse write–verify technique to program RRAMs into target analogue conductance range. b, An example sequence of the write–verify programming. c, RRAM conductance distribution measured during and after the write–verify programming. Each blue dot represents one RRAM cell measured during write–verify. The grey shades show that the RRAM conductance relaxation cause the distribution to spread out from the target values. The darker shade shows that the iterative programming helps narrow the distribution. d, Standard deviation of conductance change measured at different initial conductance states and different time duration after the initial programming. The initial conductance relaxation happens at a faster rate than longer term retention degradation. e, Standard deviation of conductance relaxation decreases with increasing iterative programming cycles. f, Distribution of the number of SET/RESET pulses needed to reach conductance acceptance range.
Extended Data Fig. 3 Iterative write–verify RRAM programming. a, Flowchart of the incremental-pulse write–verify technique to program RRAMs into target analogue conductance range. b, An example sequence of the write–verify programming. c, RRAM conductance distribution measured during and after the write–verify programming. Each blue dot represents one RRAM cell measured during write–verify. The grey shades show that the RRAM conductance relaxation cause the distribution to spread out from the target values. The darker shade shows that the iterative programming helps narrow the distribution. d, Standard deviation of conductance change measured at different initial conductance states and different time duration after the initial programming. The initial conductance relaxation happens at a faster rate than longer term retention degradation. e, Standard deviation of conductance relaxation decreases with increasing iterative programming cycles. f, Distribution of the number of SET/RESET pulses needed to reach conductance acceptance range.CC BY · PubMed Central
Extended Data Fig. 4 4 basic neuron operations that enable MVM with multi-bit inputs and outputs. a, Initialization, precharge sampling capacitor Csample and output wires (SLs), and discharge integration capacitor Cinteg. b, Sampling and integration, sample SL voltage onto Csample, followed by integrating the charge onto Cinteg. c, Comparison and readout. The amplifier is turned into comparator mode to determine the polarity of the integrated voltage. Comparator outputs are written out of the neuron through the outer feedback loop. d, Charge decrement, charge is added or subtracted on Cinteg through the outer feedback loop, depending on value stored in the latch.
Extended Data Fig. 4 4 basic neuron operations that enable MVM with multi-bit inputs and outputs. a, Initialization, precharge sampling capacitor Csample and output wires (SLs), and discharge integration capacitor Cinteg. b, Sampling and integration, sample SL voltage onto Csample, followed by integrating the charge onto Cinteg. c, Comparison and readout. The amplifier is turned into comparator mode to determine the polarity of the integrated voltage. Comparator outputs are written out of the neuron through the outer feedback loop. d, Charge decrement, charge is added or subtracted on Cinteg through the outer feedback loop, depending on value stored in the latch.CC BY · PubMed Central
Extended Data Fig. 5 Scatter plots of measured MVMs vs. ideal MVMs. Results in a-d are generated using the same 64×64 normally distributed random matrix and 1000 uniformed distributed floating-point vectors ϵ [-1, 1]. a, Forward MVM using differential input scheme with inputs quantized to 4-bit and outputs 6-bit. b, Backward MVM using differential output scheme. The higher RMSE is caused by more voltage drop on each SL driver that needs to drive 128 RRAM cells, compared to 64 cells driven by each BL driver during forward MVM. c, MVM root-mean-square error (RMSE) does not reduce when increasing input from 4-bit (a) to 6-bit. This is caused by using a lower input voltage that leads to worse signal-to-noise-ratio. d, 2-phase operation reduces MVM RMSE with 6-bit input by breaking inputs into 2 segments and performing MVMs separately, such that input voltage does not need to be reduced. e–f, Outputs from conv15 layer of ResNet-20. Weights of conv15 are divided to 3 CIM cores. Layer outputs show a higher RMSE when performing MVM in parallel on the 3 cores (f) than sequentially on the 3 cores (e).
Extended Data Fig. 5 Scatter plots of measured MVMs vs. ideal MVMs. Results in a-d are generated using the same 64×64 normally distributed random matrix and 1000 uniformed distributed floating-point vectors ϵ [-1, 1]. a, Forward MVM using differential input scheme with inputs quantized to 4-bit and outputs 6-bit. b, Backward MVM using differential output scheme. The higher RMSE is caused by more voltage drop on each SL driver that needs to drive 128 RRAM cells, compared to 64 cells driven by each BL driver during forward MVM. c, MVM root-mean-square error (RMSE) does not reduce when increasing input from 4-bit (a) to 6-bit. This is caused by using a lower input voltage that leads to worse signal-to-noise-ratio. d, 2-phase operation reduces MVM RMSE with 6-bit input by breaking inputs into 2 segments and performing MVMs separately, such that input voltage does not need to be reduced. e–f, Outputs from conv15 layer of ResNet-20. Weights of conv15 are divided to 3 CIM cores. Layer outputs show a higher RMSE when performing MVM in parallel on the 3 cores (f) than sequentially on the 3 cores (e).CC BY · PubMed Central
Extended Data Fig. 6 Data distribution with and without model-driven chip calibration. Left, Distribution of inputs to the final fully connected layer of ResNet-20 when the inputs are generated from (top-to-bottom) CIFAR-10 test-set data, training-set data, and random uniform data. Right, Distribution of outputs from the final fully connected layer of ResNet-20. The test-set and training-set have similar distributions while random uniform data produces a markedly different output distribution. To ensure that the MVM output voltage dynamic range during testing is calibrated to occupy the full ADC input swing, the calibration data should come from training-set data that closely resembles the test-set data.
Extended Data Fig. 6 Data distribution with and without model-driven chip calibration. Left, Distribution of inputs to the final fully connected layer of ResNet-20 when the inputs are generated from (top-to-bottom) CIFAR-10 test-set data, training-set data, and random uniform data. Right, Distribution of outputs from the final fully connected layer of ResNet-20. The test-set and training-set have similar distributions while random uniform data produces a markedly different output distribution. To ensure that the MVM output voltage dynamic range during testing is calibrated to occupy the full ADC input swing, the calibration data should come from training-set data that closely resembles the test-set data.CC BY · PubMed Central
Extended Data Fig. 7 Noise-resilient training of CNNs, LSTMs and RBMs. a, Change in CIFAR-10 test-set classification accuracy under different weight noise levels during inference. Noise is represented as fraction of the maximum absolute value of weights. Different curves represent models trained at different levels of noise injection. b, Change in voice command recognition accuracy with weight noise levels. c, Change in MNIST image-reconstruction error with weight noise levels. d, Decreasing of image-reconstruction error with Gibbs sampling steps during RBM inference. e, Differences in weight distributions when trained without and with noise injection.
Extended Data Fig. 7 Noise-resilient training of CNNs, LSTMs and RBMs. a, Change in CIFAR-10 test-set classification accuracy under different weight noise levels during inference. Noise is represented as fraction of the maximum absolute value of weights. Different curves represent models trained at different levels of noise injection. b, Change in voice command recognition accuracy with weight noise levels. c, Change in MNIST image-reconstruction error with weight noise levels. d, Decreasing of image-reconstruction error with Gibbs sampling steps during RBM inference. e, Differences in weight distributions when trained without and with noise injection.CC BY · PubMed Central
Extended Data Fig. 8 Measured chip inference performance. a, CIFAR-10 training-set accuracy loss due to hardware non-idealities, and accuracy recovery at each step of the chip-in-the-loop progressive fine-tuning. From left to right, each data point represents a new layer programmed onto the chip. The blue solid lines represent the accuracy loss measured when performing inference of that layer on-chip. The red dotted lines represent the measured recovery in accuracy by fine-tuning subsequent layers. b, Ablation study showing the impacts of input, activation, and weight quantizations, and weight noise injection on inference errors.
Extended Data Fig. 8 Measured chip inference performance. a, CIFAR-10 training-set accuracy loss due to hardware non-idealities, and accuracy recovery at each step of the chip-in-the-loop progressive fine-tuning. From left to right, each data point represents a new layer programmed onto the chip. The blue solid lines represent the accuracy loss measured when performing inference of that layer on-chip. The red dotted lines represent the measured recovery in accuracy by fine-tuning subsequent layers. b, Ablation study showing the impacts of input, activation, and weight quantizations, and weight noise injection on inference errors.CC BY · PubMed Central
Extended Data Fig. 9 Implementation of various AI models. a, Architecture of ResNet-20 for CIFAR-10 classification. b, The batch normalization parameters are merged into convolutional weights and biases before mapping on-chip. c, Illustration of the process to map 4-dimensional weights of a convolutional layer to NeuRRAM CIM cores. d, Architecture of the LSTM model used for Google speech command recognition. The model contains 4 parallel LSTM cells and makes predictions based on the sum of outputs from the 4 cells. e, Architecture of the RBM model used for MNIST image recovery. During inference, MVMs and Gibbs sampling are performed back and-forth between visible and hidden neurons. f, Process to map RBM on NeuRRAM CIM cores. Adjacent pixels are assigned to different cores to equalize the MVM output dynamic range at different cores.
Extended Data Fig. 9 Implementation of various AI models. a, Architecture of ResNet-20 for CIFAR-10 classification. b, The batch normalization parameters are merged into convolutional weights and biases before mapping on-chip. c, Illustration of the process to map 4-dimensional weights of a convolutional layer to NeuRRAM CIM cores. d, Architecture of the LSTM model used for Google speech command recognition. The model contains 4 parallel LSTM cells and makes predictions based on the sum of outputs from the 4 cells. e, Architecture of the RBM model used for MNIST image recovery. During inference, MVMs and Gibbs sampling are performed back and-forth between visible and hidden neurons. f, Process to map RBM on NeuRRAM CIM cores. Adjacent pixels are assigned to different cores to equalize the MVM output dynamic range at different cores.CC BY · PubMed Central
Extended Data Fig. 10 Chip-measured image recovery using RBM. Top: Recovery of MNIST test-set images with randomly selected 20% of pixels flipped to complementary intensity. Bottom: Recovery of MNIST test-set images with bottom 1/3 of pixels occluded.
Extended Data Fig. 10 Chip-measured image recovery using RBM. Top: Recovery of MNIST test-set images with randomly selected 20% of pixels flipped to complementary intensity. Bottom: Recovery of MNIST test-set images with bottom 1/3 of pixels occluded.CC BY · PubMed Central
Extended Data Fig. 11 NeuRRAM test system and chip micrographs at various scales. a, A NeuRRAM chip wire-bonded to a package. b, Measurement board that connects a packaged NeuRRAM chip (left) to a field-programmable gate array (FPGA, right). The board houses all the components necessary to power, operate and measure the chip. No external lab equipment is needed for the chip operations. c, Micrograph of a 48-core NeuRRAM chip. d, Zoomed-in micrograph of a single CIM core. e, Zoomed-in micrograph of 2×2 corelets within the TNSA. One neuron circuit occupies 1270 μm2, which is >100× smaller than most ADC designs in 130-nm summarized in an ADC survey62. f, Chip area breakdown.
Extended Data Fig. 11 NeuRRAM test system and chip micrographs at various scales. a, A NeuRRAM chip wire-bonded to a package. b, Measurement board that connects a packaged NeuRRAM chip (left) to a field-programmable gate array (FPGA, right). The board houses all the components necessary to power, operate and measure the chip. No external lab equipment is needed for the chip operations. c, Micrograph of a 48-core NeuRRAM chip. d, Zoomed-in micrograph of a single CIM core. e, Zoomed-in micrograph of 2×2 corelets within the TNSA. One neuron circuit occupies 1270 μm2, which is >100× smaller than most ADC designs in 130-nm summarized in an ADC survey62. f, Chip area breakdown.CC BY · PubMed Central
Extended Data Fig. 12 Energy consumption, latency, and throughput measurement results. a, Measured energy consumption per operation during the MVM input stage (without 2-phase operation) and output stage, where one multiply–accumulate (MAC) counts as two operations. b, Energy consumption breakdown at various MVM input and output bit-precisions. Outputs are 2-bit higher than inputs during a MVM to account for additional precision requirements from partial-sum accumulation. c, Latency for performing one MVM with 256×256 weight matrix. d, Peak computational throughput (in giga-operations per second). e, Throughput-power efficiency (in tera-operations per watt).
Extended Data Fig. 12 Energy consumption, latency, and throughput measurement results. a, Measured energy consumption per operation during the MVM input stage (without 2-phase operation) and output stage, where one multiply–accumulate (MAC) counts as two operations. b, Energy consumption breakdown at various MVM input and output bit-precisions. Outputs are 2-bit higher than inputs during a MVM to account for additional precision requirements from partial-sum accumulation. c, Latency for performing one MVM with 256×256 weight matrix. d, Peak computational throughput (in giga-operations per second). e, Throughput-power efficiency (in tera-operations per watt).CC BY · PubMed Central

Verification

Claims the paper makes

7 Valid · 1 Not valid · 7 Not checked · 130 Own results and interpretations

72 more claims

Create a free account to see all 145 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 37
  • It may state two separate claims as one. 41

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

2 more claims about this paper

Create a free account to see every claim other papers make about this one.

See all claims — free account

Citations

Cites

Show all 65 references

Research works printed but not linked: 1

  1. Koller, D. & Friedman, N. Probabilistic Graphical Models: Principles and Techniques (Adaptive Computation and Machine Learning series) (MIT Press, 2009).

Cited by

See all 943

Metadata

Authors
Wan, W., Kubendran, R., Schaefer, C., Eryilmaz, S. B., Zhang, W., Wu, D., Deiss, S., Raina, P., Qian, H., Gao, B., Joshi, S., Wu, H., Wong, H.-S. P., & Cauwenberghs, G.
Affiliations
  • University of California San Diego (US)
  • University of Pittsburgh (US)
  • University of Notre Dame (US)
  • Stanford University (US)
  • Tsinghua University (CN)
Corresponding author
  • Wan, Weier
  • Gao, Bin
  • Joshi, Siddharth
  • Wu, Huaqiang
  • Wong, H.-S. Philip
  • Cauwenberghs, Gert
Venue
Nature
Published
2022-08-17
Volume
608
Issue
7923
Pages
504-512
ISSN
  • 0028-0836 (ISSN-L)
  • 1476-4687
OpenAlex
W4292121737
Cited by
943
References
65
Citations by year
2026: 235 · 2025: 273 · 2024: 246 · 2023: 167 · 2022: 22
FWCI
48.45
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • eScholarship (California Digital Library) (submittedVersion, cc-by)
Funders
  • National Science Foundation (1317470)
  • Western Digital
  • University of Pennsylvania
  • Pennsylvania State University
  • Stanford SystemX Alliance
  • National Natural Science Foundation of China (61851404)
  • Office of Naval Research
Topics
  • Physical Sciences › Engineering › Electrical and Electronic Engineering › Advanced Memory and Neural Computing
  • Physical Sciences › Engineering › Electrical and Electronic Engineering › Ferroelectric and Negative Capacitance Devices
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
Keywords
in-memory computing, resistive random access memory, neuromorphic hardware, edge AI, energy efficiency, weight quantization, hardware-software co-optimization, non-volatile memory, image classification, voice command recognition

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: A compute-in-memory chip based on resistive random-access memory. https://app.cochat.ai/papers/10.1038/s41586-022-04992-8
The paper
Wan, W., Kubendran, R., Schaefer, C., Eryilmaz, S. B., Zhang, W., Wu, D., Deiss, S., Raina, P., Qian, H., Gao, B., Joshi, S., Wu, H., Wong, H.-S. P., & Cauwenberghs, G. (2022). A compute-in-memory chip based on resistive random-access memory. Nature. https://doi.org/10.1038/s41586-022-04992-8

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.