Pith. sign in

REVIEW 3 major objections 5 minor 53 references

Digital Biologically Plausible Implementation of Binarized Neural Networks with Differential Hafnium Oxide Resistive Memory Arrays

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A differential hafnium-oxide resistive memory array folds the XNOR multiply into the sense amplifier, cutting synaptic bit errors by up to four orders of magnitude and letting binarized neural networks run without error-correcting codes.

desk verdict Solid 2T2R array measurements; the XNOR-in-sense-amplifier in-memory claim is unverified. read the letter →

arxiv 1908.04066 v2 pith:JTLGGILK submitted 2019-08-12 cs.ET

Tifenn Hirtzlin , Marc Bocquet , Bogdan Penkovsky , Jacques-Olivier Klein , Etienne Nowak , Elisa Vianello , Jean-Michel Portal , Damien Querlioz This is my paper · ORCID
classification cs.ET
keywords binarizedneuralnetworksresistiverandomaccessmemoryOxRAMin-memorycomputingXNORsenseamplifier2T2Rdifferentialprechargebiterrortolerance
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper aims to establish that a binarized neural network can be implemented in digital hardware with hafnium-oxide resistive memory and no error-correcting codes, despite the intrinsic variability of the memory devices. The design stores each synaptic weight as a complementary pair of resistances in a two-transistor, two-resistor (2T2R) cell, and performs the network's XNOR multiplication directly inside the precharge sense amplifier during readout. Based on electrical measurements of a fabricated 1-kilobit array and simulations of trained networks on MNIST, CIFAR-10, ImageNet, and an ECG task, the paper argues that 2T2R readout cuts bit errors by up to four orders of magnitude relative to single-device 1T1R cells, and that binarized networks lose no accuracy at weight error rates up to $10^{-4}$ and only minimal accuracy at $10^{-3}$. A sympathetic reader would care because removing ECC, plus tolerating weak programming regimes, points toward low-energy inference at the edge.

What carries the argument

The load-bearing object is the 2T2R differential bit cell paired with a precharge sense amplifier (PCSA). Each bit is stored as a complementary pair of OxRAM devices, one in a low resistance state and one in a high resistance state, so a bit error occurs only if the low-resistance device reads more resistive than its high-resistance partner. The PCSA compares the two resistances through a precharge and discharge cycle, and four additional transistors in its discharge branches gate the discharge on the neuron input voltage; the sense operation itself therefore computes the XNOR of the stored weight and the input. This feeds a digital POPCOUNT circuit, which counts the ones in the XNOR results, and a threshold subtraction, implementing the binarized neuron equation $A_j = \mathrm{sign}(\mathrm{POPCOUNT}_i(\mathrm{XNOR}(W_{ji},X_i)) - T_j)$.

What would settle it

Program a fabricated 2T2R array with known weight patterns, drive the XNOR-enriched sense amplifiers with real neuron input voltages, and compare each output against the expected XNOR truth table; any error rate above the level explained by resistance variability would show that the in-memory XNOR operation is not working as simulated.

Watch

Extended reading notes

Core claim

The central discovery is that the differential 2T2R memory structure, combined with the inherent error tolerance of binarized neural networks, makes formal error correction unnecessary for RRAM-based in-memory inference. In a fabricated 1-kilobit array of HfO2 OxRAM devices integrated in the back end of line of a 130 nm CMOS process, 2T2R readout through precharge sense amplifiers holds bit error rates below $10^{-3}$ over 700 million program/erase cycles under strong programming conditions, and below $10^{-4}$ over more than $10^{10}$ cycles under endurance-optimized conditions, while the same devices used as 1T1R cells exceed $10^{-3}$ after a few million cycles. The paper further shows that trained binarized networks keep full accuracy at weight error rates up to $10^{-4}$ and suffer only minimal degradation at $10^{-3}$ across four tasks, and that the read-plus-XNOR-add operation can be integrated into the sense amplifier, estimated at 14 fJ per operation in an advanced node.

Load-bearing premise

The fabricated array was only measured for differential readout of stored weights, not for the XNOR operation with applied neuron inputs; if the added discharge transistors introduce timing asymmetry, voltage-margin loss, or reliability effects, the in-memory computing claim and the 14 fJ per read-plus-add estimate would not transfer from simulation to silicon.

Editorial extensions

If this is right

  • RRAM-based inference hardware can drop formal ECC: at weight error rates up to $10^{-3}$, binarized networks show essentially no accuracy loss, removing the area, energy, and latency of syndrome computation.
  • Memory devices can be operated in low-energy programming regimes (as low as 20–30 pJ per bit) and high-endurance regimes (beyond $10^{10}$ cycles) while keeping the bit error rate below $10^{-3}$.
  • A full MNIST inference with the proposed system is estimated at 25 nJ, compared with millijoules for CPU and GPU implementations.
  • At equal redundancy, the 2T2R differential readout provides bit-error reduction comparable to a SECDED(8,4) Hamming code, but without any decoding logic.
  • The energy advantage of binarization over 8-bit fixed-point arithmetic grows as the target accuracy decreases, reaching nearly a factor of ten at lower precision.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the XNOR-in-sense-amplifier works in silicon, the same differential readout trick could be carried over to other emerging memories such as phase-change or magnetic RAM, since the argument rests on the readout circuit, not on hafnium oxide specifically.
  • The measured tolerance to $10^{-3}$ weight errors suggests that write-verify loops, which dominate programming energy in analog RRAM synapses, could be omitted for binary digital weights; a direct energy comparison with and without verify would test this.
  • A silicon demonstration of the XNOR-enabled sense amplifier would settle whether the 14 fJ per read-plus-add estimate survives real timing and voltage margins; that figure currently comes from an advanced-node estimate, not from the fabricated 130 nm array.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. This manuscript proposes a digital, biologically inspired implementation of binarized neural network inference using a differential 2T2R HfO2 OxRAM array with precharge sense amplifiers (PCSA) that are designed to perform XNOR operations during readout. The authors fabricated a 1-kb array in a 130 nm CMOS process and characterized its programming, bit-error-rate, and endurance behavior. They report that the 2T2R differential readout reduces synaptic bit error rates by up to four orders of magnitude relative to single-device 1T1R readout in the same programming conditions, and that the benefit is comparable to SEC/SECDED ECC at equivalent redundancy without decoding circuitry. Error-injection simulations on MNIST, CIFAR-10, ImageNet, and ECG tasks show that accuracy is maintained up to weight error rates around 10^-3. A synthesizable SystemVerilog system is described, with energy estimates of 25 nJ per MNIST digit and 14 fJ per read-and-add operation in an advanced node, enabling the use of low-programming-energy (20-30 pJ) and high-endurance (>10^10 cycles) regimes. The paper concludes that the differential architecture allows error-tolerant neural network hardware without formal ECC.

Significance. If fully validated, this would be a valuable demonstration of error-tolerant in-memory computing with emerging nonvolatile memories. The paper's main measured strengths are the direct experimental comparison of 2T2R against 1T1R bit error rates obtained from the same programmed distributions, the extensive endurance measurements over hundreds of millions and tens of billions of cycles, and the error-injection study on four tasks with five repetitions and error bars. These results give a quantitative, falsifiable basis for the claim that differential readout can replace formal ECC in binarized neural network hardware. The significance is partly bounded, however, because the XNOR-in-sense-amplifier operation is presented as a schematic and as part of a system-level simulation, not as a measured circuit function; the silicon data validate only the differential readout. The energy figures are projections, not measured silicon results. With those caveats addressed or the claims reframed, the work would be a solid contribution to the cs.ET / neuromorphic-hardware literature.

major comments (3)
  1. [Section 2.2, Fig. 2(b); Section 3.1] The manuscript never reports a measurement of the XNOR-augmented PCSA. All silicon data in Section 3.1 (Figs. 4-8) are obtained with the standard differential readout; no experiment applies a neuron input X during sensing and compares the sense-amplifier output against the XNOR truth table for the four (W,X) combinations. Since the four additional transistors lie in the discharge branches that determine the race, their on-resistance and timing can alter the voltage margin characterized in Fig. 4(d). The abstract's claim that the circuit 'allows performing the exclusive NOR operations ... directly within the sense amplifiers' is therefore supported only by schematic and system-level simulation. Please add a measured XNOR-mode characterization, or at minimum a post-layout simulation with the measured RRAM distributions, and adjust the claims accordingly.
  2. [Section 4.1.3] The 14 fJ per read-plus-addition estimate is an advanced-node projection, not a silicon result, and it assumes the XNOR-augmented PCSA works as drawn in Fig. 2(b). The text states 'in our estimates,' but the assumptions behind this number (technology node, supply voltage, activity factor, inclusion of the XNOR discharge transistors, and whether the RRAM distributions are modeled) are not fully specified. As a projection the figure is acceptable, but it should be clearly labeled as a design estimate for a future implementation rather than as a validated property of the fabricated array.
  3. [Section 3.2, Fig. 9] The error-tolerance simulations inject bit errors independently and uniformly with probability p, but Section 3.1 shows that hardware errors are not independent: Figs. 4(b)-4(c) and Fig. 5 report device-to-device and cycle-to-cycle variability, and Figs. 6-7 show aging trends. To support the 'without ECC' claim, the tolerance evaluation should also be performed with correlated error maps derived from the measured array (for example, using the per-device error rates of Figs. 4(e-f)) rather than only i.i.d. flips. This would directly test whether the system remains robust to the actual error structure of the fabricated memory.
minor comments (5)
  1. [Section 2.2] The sentence containing 'while perforning the XNOR operation' contains a typo: 'perforning' should be 'performing'.
  2. [Introduction, last paragraph] The word 'benckmarking' should be 'benchmarking'.
  3. [Fig. 5 caption] The caption says 'Number of errors' while the y-axis label appears to be 'Error rate (%)'; please unify these terms and state whether the plotted quantity is a count or a rate.
  4. [Table 1] The programming-energy ranges are reported without a methodological footnote; please state whether these values are measured from current/voltage waveforms, calculated from programming conditions, or obtained from the system-level simulator.
  5. [Section 4.1.1] The 25 nJ per digit figure should be explicitly tied to the synthesized system at the considered process node and to the fully connected MNIST network used; currently the sentence reads as a more general statement than the supporting method provides.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the 2T2R BER reduction is a direct measurement, BNN error tolerance comes from independent error injection, and the energy projection is from synthesis, not from the target results.

full rationale

The paper's derivation chain is not circular. The 2T2R BER reduction (Figs. 4-8, Section 3.1) is a direct experimental measurement of the fabricated array: the 1T1R BER is obtained from the same programmed resistance distributions via an ideal-threshold calculation, while the 2T2R BER is read by the on-chip precharge sense amplifier, so the comparison is an empirical benchmark rather than a fit. The BNN error-tolerance curves (Section 3.2, Fig. 9) are produced by randomly flipping trained network weights, independently of the measured device BER; they are not fitted to the array data. The system energy estimates (Section 4.1) come from a standard synthesizable SystemVerilog flow with VCD-based power evaluation, and the 14 fJ figure is explicitly stated as an estimate in an advanced node, not a back-fit. The XNOR-in-PCSA building block is imported from prior published work, including co-authored work ('we follow the pioneering works of (Zhao et al., 2014), which shows that precharge sense amplifier can be enriched with any logic operation'), but that prior work is an external circuit-design result and the paper's measured claims do not reduce to it; the absence of a direct measurement of the XNOR operation is a validation gap, not a circular definition. No fitted parameter is renamed as a prediction, and no result is equivalent to its own input by construction.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central claim rests on measured device statistics, a schematic-level XNOR operation, and simulation-based error and energy models. The main unverified premise is the in-sensing XNOR; the main modeling risk is independent random bit flips.

assumptions (4)
  • domain assumption Random independent bit flips in all weights represent OxRAM programming and read errors.
    Section 3.2 injects errors uniformly and independently into trained weights; real arrays show device-to-device and cycle-to-cycle correlations and aging (Section 3.1), which could make actual resilience lower for some layers or devices.
  • domain assumption The XNOR-enriched PCSA behaves as the schematic in Fig. 2(b) without parasitic imbalance or reliability issues.
    No measurement of the XNOR function is reported; measured PCSA validation in Fig. 4(d) covers only differential resistance comparison.
  • domain assumption Cadence synthesis with VCD activity files gives representative energy for the full system.
    Energy estimates use the 130 nm design kit and a 28 nm projection, not a fabricated full chip; the memory array is the only measured block.
  • domain assumption Programming conditions in Table 1 satisfy BER below 10^-3 for all tasks because the networks tolerate that rate.
    Section 4.1.4 selects conditions with measured BER below 10^-3 based on the tolerance curves of Section 3.2; this selection assumes the tolerance simulations transfer to the real system.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Digital Biologically Plausible Implementation of Binarized Neural Networks with Differential Hafnium Oxide Resistive Memory Arrays." pith.science (2026). https://pith.science/paper/JTLGGILK

@misc{pith2026190804066,
  author       = {Pith},
  title        = {Pith review of: Digital Biologically Plausible Implementation of Binarized Neural Networks with Differential Hafnium Oxide Resistive Memory Arrays},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JTLGGILK}},
  note         = {Machine review of arXiv:1908.04066}
}
read the original abstract

The brain performs intelligent tasks with extremely low energy consumption. This work takes inspiration from two strategies used by the brain to achieve this energy efficiency: the absence of separation between computing and memory functions, and the reliance on low precision computation. The emergence of resistive memory technologies indeed provides an opportunity to co-integrate tightly logic and memory in hardware. In parallel, the recently proposed concept of Binarized Neural Network, where multiplications are replaced by exclusive NOR (XNOR) logic gates, offers a way to implement artificial intelligence using very low precision computation. In this work, we therefore propose a strategy to implement low energy Binarized Neural Networks, which employs brain-inspired concepts, while retaining energy benefits from digital electronics. We design, fabricate and test a memory array, including periphery and sensing circuits, optimized for this in-memory computing scheme. Our circuit employs hafnium oxide resistive memory integrated in the back end of line of a 130 nanometer CMOS process, in a two transistors - two resistors cell, which allows performing the exclusive NOR operations of the neural network directly within the sense amplifiers. We show, based on extensive electrical measurements, that our design allows reducing the amount of bit errors on the synaptic weights, without the use of formal error correcting codes. We design a whole system using this memory array. We show on standard machine learning tasks (MNIST, CIFAR-10, ImageNet and an ECG task) that the system has an inherent resilience to bit errors. We evidence that its energy consumption is attractive compared to more standard approaches, and that it can use the memory devices in regimes where they exhibit particularly low programming energy and high endurance.

Figures

Figures reproduced from arXiv: 1908.04066 by the authors.

Figure 1
Figure 1. (a) Scanning Electron Microscopy image of the back-end-of-line of the CMOS process integrating an OxRAM device. (b) Photograph and (c) simplified schematic of the one kilobit in-memory computing-targeted memory array characterized in this work. We chose hafnium oxide OxRAMs as they are known to provide non-volatile memories compatible with modern CMOS process, and only involve foundry-friendly materials and process … view at source ↗
Figure 2
Figure 2. (a) Schematic of the precharge sense amplifier used in this work to read 2T2R memory cells. (b) Schematic of the precharge sense amplifier augmented with a XNOR logic operation. The programming of devices in our array is made sequentially, i.e. on a device-by-device basis. The first time that the memory array is used, all devices are “formed”. To form the device of row i and column j, the bit line BLj , connected to… view at source ↗
Figure 3
Figure 3. Schematization of the full architecture to implement Binarized Neural Network, in the “parallel to sequential” configuration. The system assembles memory block surrounded by logic circuits, and moves minimal data between the blocks. The architecture is presented with three rows and three columns (i.e. N = M = 3) of kilobit memory blocks (i.e. n = 32). Based on the basic memory array with PCSAs enriched with XNOR, we… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: (a) Distribution of the LRS and the HRS of the OxRAM devices in an array programmed with a checkerboard pattern. RESET voltage of 2.5V , SET current of 55µA and programming time of 1µs. (b-c) Proportion of 1 values read by the onchip precharge sense amplifier, over 100…
Figure 5
Figure 5. Figure 5: Number of errors for different programming conditions, as measured by the precharge sense amplifier, for 2T2R configuration on a kilobit memory array. The “< 1” label means that no errors were detected. The error bars present the minimum and maximum number of detected …
Figure 6
Figure 6. Figure 6: (a-b) Distribution of the resistance values, (c-d) mean resistance value and (e) mean bit error rate over 10 million cycles measured by the precharge sense amplifier, in the 2T2R configuration, as function of the number of cycles that a device has been programmed. RESE…
Figure 7
Figure 7. Figure 7: (a-b) Mean resistance value of the BL and BLb device over 10 thousand cycles for measurements of a device pair over 5 × 1010 cycles. RESET voltage of 1.5V , SET current of 200µA and programming time of 1µs. In [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Experimental bit error rate of the 2T2R array, measured by the precharge sense amplifiers, as a function of the bit error rate obtained individual (1T1R) RRAM devices in the same programming conditions. The detailed methodology for obtaining this graph is presented in …
Figure 9
Figure 9. Figure 9: Recognition rate on the validation dataset of the fully connected neural network for MNIST, the convolutional neural network for CIFAR10, and AlexNet for ImageNet (Top-5 and Top-1) accuracies and the ECG analysis task, as a function of the bit error rate over the weigh…
Figure 10
Figure 10. Figure 10: Dark blue circles: MNIST validation accuracy as a function of the inference energy of our Binarized Neural Network hardware design. Light blue square: same, as function of the energy used for arithmetic operation in a real valued neural networks employing eight bits f…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 44 canonical work pages

  1. [1]

    M., Boybat, I., Nolfo, C., et al

    Ambrogio, S., Narayanan, P., Tsai, H., Shelby, R. M., Boybat, I., Nolfo, C., et al. (2018). Equivalent-accuracy accelerated neural-network training using analogue memory. Nature 558, 60 ambrogio2018equivalent

    work page 2018

  2. [2]

    Ando, K., Ueyoshi, K., Orimo, K., Yonekawa, H., Sato, S., Nakahara, H., et al. (2017). Brein memory: A 13-layer 4.2 k neuron/0.8 m synapse binary/ternary reconfigurable in-memory deep neural network accelerator in 65 nm cmos. In Proc. VLSI Symp. on Circuits (IEEE), C24--C25 ando2017brein

    work page 2017

  3. [3]

    Bankman, D., Yang, L., Moons, B., Verhelst, M., and Murmann, B. (2018). An always-on 3.8muj/86 \ with all memory on chip in 28-nm cmos. IEEE Journal of Solid-State Circuits 54, 158--172 bankman2018always

    work page 2018

  4. [4]

    Bocquet, M., Hirztlin, T., Klein, J.-O., Nowak, E., Vianello, E., Portal, J.-M., et al. (2018). In-memory and error-immune differential rram implementation of binarized deep neural networks. In IEDM Tech. Dig. (IEEE), 20.6.1 bocquet2018

    work page 2018

  5. [5]

    Chen, A. (2016). A review of emerging non-volatile memory (nvm) technologies and applications. Solid-State Electronics 125, 25--38 chen2016review

    work page 2016

  6. [6]

    H., Li, K

    Chen, W. H., Li, K. X., Lin, W. Y., Hsu, K. H., Li, P. Y., Yang, C. H., et al. (2018). A 65nm 1mb nonvolatile computing-in-memory ReRAM macro with sub-16ns multiply-and-accumulate for binary DNN AI edge processors. In Proc. ISSCC . 494--496. doi:10.1109/ISSCC.2018.8310400 chen_65nm_2018

    arXiv 2018

  7. [7]

    H., Lin, W

    Chen, W. H., Lin, W. J., Lai, L. Y., Li, S., Hsu, C. H., Lin, H. T., et al. (2017). A 16mb dual-mode ReRAM macro with sub-14ns computing-in-memory and memory functions enabled by self-write termination scheme. In IEDM Tech . Dig . 28.2.1--28.2.4. doi:10.1109/IEDM.2017.8268468 chen_16mb_2017

    work page arXiv 2017

  8. [8]

    Courbariaux, M., Hubara, I., Soudry, D., El-Yaniv, R., and Bengio, Y. (2016). Binarized neural networks: Training deep neural networks with weights and activations constrained to+ 1 or-1. arXiv preprint arXiv:1602.02830 courbariaux2016binarized

    arXiv 2016

Show all 53 references
  1. [9]

    Covi, E., Brivio, S., Serb, A., Prodromakis, T., Fanciulli, M., and Spiga, S. (2016). Analog memristive synapse in spiking networks implementing unsupervised learning. Frontiers in neuroscience 10, 482 covi2016analog

    2016

  2. [10]

    Dong, Q., Kim, Y., Lee, I., Choi, M., Li, Z., Wang, J., et al. (2017). 11.2 a 1mb embedded nor flash memory with 39 w program power for mm-scale high-temperature sensor nodes. In 2017 IEEE International Solid-State Circuits Conference (ISSCC) (IEEE), 198--199 dong201711

    2017

  3. [11]

    Big data needs a hardware revolution

    Editorial (2018). Big data needs a hardware revolution. Nature 554, 145. doi:10.1038/d41586-018-01683-1 editorial_big_2018

    2018 doi

  4. [12]

    A., Selen, L

    Faisal, A. A., Selen, L. P., and Wolpert, D. M. (2008). Noise in the nervous system. Nature reviews neuroscience 9, 292 faisal2008noise

    2008

  5. [13]

    Giacomin, E., Greenberg-Toledo, T., Kvatinsky, S., and Gaillardon, P.-E. (2019). A robust digital rram-based convolutional block for low-power image processing and learning applications. IEEE Transactions on Circuits and Systems I: Regular Papers 66, 643--654 giacomin2019robust

    2019

  6. [14]

    Gregori, S., Cabrini, A., Khouri, O., and Torelli, G. (2003). On-chip error correcting techniques for new-generation flash memories. Proc. IEEE 91, 602--616 gregori2003chip

    2003

  7. [15]

    Grossi, A., Nowak, E., Zambelli, C., Pellissier, C., Bernasconi, S., Cibrario, G., et al. (2016). Fundamental variability limits of filament-based rram. In IEDM Tech. Dig. (IEEE), 4--7 grossi2016fundamental

    2016

  8. [16]

    Grossi, A., Vianello, E., Zambelli, C., Royer, P., Noel, J.-P., Giraud, B., et al. (2018). Experimental investigation of 4-kb rram arrays programming conditions suitable for tcam. IEEE Transactions on Very Large Scale Integration (VLSI) Systems 26, 2599--2607 grossi2018experimental

    2018

  9. [17]

    Hsieh, W.-T., Chih, Y.-D., Chang, J., Lin, C.-J., and King, Y.-C. (2017). Differential Contact RRAM Pair for Advanced CMOS Logic NVM applications. Proc. SSDM , 171 hsieh_differential_2017

    2017

  10. [18]

    and Wong, H.-S

    Ielmini, D. and Wong, H.-S. P. (2018). In-memory computing with resistive switching devices. Nature Electronics 1, 333 ielmini2018memory

    2018

  11. [19]

    and Liu, S.-C

    Indiveri, G. and Liu, S.-C. (2015). Memory and information processing in neuromorphic systems. Proc. IEEE 103, 1379--1397 indiveri2015memory

    2015

  12. [20]

    and Szegedy, C

    Ioffe, S. and Szegedy, C. (2015). Batch normalization: Accelerating deep network training by reducing internal covariate shift. arXiv preprint arXiv:1502.03167 ioffe2015batch

    2015 arXiv

  13. [21]

    P., Young, C., Patil, N., Patterson, D., Agrawal, G., Bajwa, R., et al

    Jouppi, N. P., Young, C., Patil, N., Patterson, D., Agrawal, G., Bajwa, R., et al. (2017). In-datacenter performance analysis of a tensor processing unit. In Proc. ISCA (IEEE), 1--12 jouppi2017datacenter

    2017

  14. [22]

    Kingma, D. P. and Ba, J. (2014). Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 kingma2014adam

    2014 arXiv

  15. [23]

    and Bornholdt, S

    Klemm, K. and Bornholdt, S. (2005). Topology of biological networks and reliability of information processing. Proceedings of the National Academy of Sciences 102, 18414--18419 klemm2005topology

    2005

  16. [24]

    and Hinton, G

    Krizhevsky, A. and Hinton, G. (2009). Learning multiple layers of features from tiny images. Tech. rep., Citeseer krizhevsky2009learning

    2009

  17. [25]

    Krizhevsky, A., Sutskever, I., and Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems. 1097--1105 krizhevsky2012imagenet

    2012

  18. [26]

    D., Bhattacharya, S., Georgiev, P., Forlivesi, C., Jiao, L., Qendro, L., et al

    Lane, N. D., Bhattacharya, S., Georgiev, P., Forlivesi, C., Jiao, L., Qendro, L., et al. (2016). Deepx: A software accelerator for low-power deep learning inference on mobile devices. In Proceedings of the 15th International Conference on Information Processing in Sensor Netwo...

    2016

  19. [27]

    LeCun, Y., Bottou, L., Bengio, Y., and Haffner, P. (1998). Gradient-based learning applied to document recognition. Proc. IEEE 86, 2278--2324 lecun1998gradient

    1998

  20. [28]

    Li, C., Belkin, D., Li, Y., Yan, P., Hu, M., Ge, N., et al. (2018). Efficient and self-adaptive in-situ learning in multilayer memristor neural networks. Nature communications 9, 2385 li2018efficient

    2018

  21. [29]

    Lin, X., Zhao, C., and Pan, W. (2017). Towards accurate binary convolutional neural network. In Advances in Neural Information Processing Systems. 345--353 lin2017towards

    2017

  22. [30]

    Ly, D. R. B., Grossi, A., Fenouillet-Beranger, C., Nowak, E., Querlioz, D., and Vianello, E. (2018). Role of synaptic variability in resistive memory-based spiking neural networks with unsupervised learning. J. Phys. D: Applied Physics ly2018role

    2018

  23. [31]

    K., and Strukov, D

    Merrikh-Bayat, F., Guo, X., Klachko, M., Prezioso, M., Likharev, K. K., and Strukov, D. B. (2017). High-performance mixed-signal neurocomputing with nanoscale floating-gate memory cell arrays. IEEE transactions on neural networks and learning systems 29, 4782--4790 merrikh2017high

    2017

  24. [32]

    Natsui, M., Chiba, T., and Hanyu, T. (2018). Design of mtj-based nonvolatile logic gates for quantized neural networks. Microelectronics journal 82, 13--21 natsui2018design

    2018

  25. [33]

    Pedram, A., Richardson, S., Horowitz, M., Galal, S., and Kvatinsky, S. (2017). Dark memory and accelerator-rich system optimization in the dark silicon era. IEEE Design & Test 34, 39--50 pedram2017dark

    2017

  26. [34]

    C., Likharev, K

    Prezioso, M., Merrikh-Bayat, F., Hoskins, B., Adam, G. C., Likharev, K. K., and Strukov, D. B. (2015). Training and operation of an integrated neuromorphic network based on metal-oxide memristors. Nature 521, 61 prezioso2015training

    2015

  27. [35]

    F., and Gamrat, C

    Querlioz, D., Bichler, O., Vincent, A. F., and Gamrat, C. (2015). Bioinspired programming of memory devices for implementing an inference engine. Proc. IEEE 103, 1398--1416 querlioz2015bioinspired

    2015

  28. [36]

    Rastegari, M., Ordonez, V., Redmon, J., and Farhadi, A. (2016). Xnor-net: Imagenet classification using binary convolutional neural networks. In Proc. ECCV (Springer), 525--542 rastegari2016xnor

    2016

  29. [37]

    G., Serrano-Gotarredona, T., Schmidt, H., Lecerf, G., Tomas, J., et al

    Sa \" ghi, S., Mayr, C. G., Serrano-Gotarredona, T., Schmidt, H., Lecerf, G., Tomas, J., et al. (2015). Plasticity in memristive devices for spiking neural networks. Frontiers in neuroscience 9, 51 saighi2015plasticity

    2015

  30. [38]

    Serb, A., Bill, J., Khiat, A., Berdan, R., Legenstein, R., and Prodromakis, T. (2016). Unsupervised learning in probabilistic neural networks with multi-state metal-oxide memristive synapses. Nature communications 7, 12611 serb2016unsupervised

    2016

  31. [39]

    P., Hu, M., et al

    Shafiee, A., Nag, A., Muralimanohar, N., Balasubramonian, R., Strachan, J. P., Hu, M., et al. (2016). Isaac: A convolutional neural network accelerator with in-situ analog arithmetic in crossbars. ACM SIGARCH Computer Architecture News 44, 14--26 shafiee2016isaac

    2016

  32. [40]

    J., and King, Y.-C

    Shih, Y.-H., Hsu, M.-Y., Lin, C. J., and King, Y.-C. (2017). Twin-bit Via RRAM in 16nm FinFET Logic Technologies . Proc. SSDM , 137 shih_twin-bit_2017

    2017

  33. [41]

    (2018 a )

    Sun, X., Peng, X., Chen, P.-Y., Liu, R., Seo, J.-s., and Yu, S. (2018 a ). Fully parallel rram synaptic array for implementing binary neural network with (+ 1,- 1) weights and (+ 1, 0) neurons. In Proc. ASP-DAC (IEEE Press), 574--579 sun2018fully

    2018

  34. [42]

    (2018 b )

    Sun, X., Yin, S., Peng, X., Liu, R., Seo, J.-s., and Yu, S. (2018 b ). Xnor-rram: A scalable and parallel resistive synaptic architecture for binary neural networks. algorithms 2, 3 sun2018xnor

    2018

  35. [43]

    Tang, T., Xia, L., Li, B., Wang, Y., and Yang, H. (2017). Binary convolutional neural network on rram. In Proc. ASP-DAC (IEEE), 782--787 tang2017binary

    2017

  36. [44]

    Wang, Z., Ambrogio, S., Balatti, S., and Ielmini, D. (2015). A 2-transistor/1-resistor artificial synapse capable of communication and stochastic learning in neuromorphic systems. Frontiers in neuroscience 8, 438 wang20152

    2015

  37. [45]

    Wang, Z., Joshi, S., Savel’ev, S., Song, W., Midya, R., Li, Y., et al. (2018). Fully memristive neural networks for pattern classification with unsupervised learning. Nature Electronics 1, 137 wang2018fully

    2018

  38. [46]

    Yu, S. (2018). Neuro-inspired computing with emerging nonvolatile memorys. Proc. IEEE 106, 260--285 yu2018neuro

    2018

  39. [47]

    Yu, S., Li, Z., Chen, P.-Y., Wu, H., Gao, B., Wang, D., et al. (2016). Binary neural network with 16 mb rram macro chip for classification and online training. In IEDM Tech. Dig. (IEEE), 16--2 yu2016binary

    2016

  40. [48]

    Zhao, W., Chappert, C., Javerliac, V., and Noziere, J.-P. (2009). High speed, high stability and low power sensing amplifier for mtj/cmos hybrid logic circuits. IEEE Transactions on Magnetics 45, 3784--3787 zhao2009high

    2009

  41. [49]

    Zhao, W., Moreau, M., Deng, E., Zhang, Y., Portal, J.-M., Klein, J.-O., et al. (2014). Synchronous non-volatile logic gate design based on resistive switching memories. IEEE Transactions on Circuits and Systems I: Regular Papers 61, 443--454 zhao2014synchronous

    2014

  42. [50]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key language month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence...

  43. [51]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

  44. [52]

    , " * write output.state after.block = add.period write newline

    ENTRY address annote author booktitle chapter doi edition editor eid howpublished institution journal key language month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.s...

  45. [53]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.