<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Publishing DTD v1.3 20210610//EN" "JATS-journalpublishing1-3.dtd">
<article article-type="research-article" dtd-version="1.3" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xml:lang="ru"><front><journal-meta><journal-id journal-id-type="publisher-id">mateltech</journal-id><journal-title-group><journal-title xml:lang="ru">Известия высших учебных заведений. Материалы электронной техники</journal-title><trans-title-group xml:lang="en"><trans-title>Izvestiya Vysshikh Uchebnykh Zavedenii. Materialy Elektronnoi Tekhniki = Materials of Electronics Engineering</trans-title></trans-title-group></journal-title-group><issn pub-type="ppub">1609-3577</issn><issn pub-type="epub">2413-6387</issn><publisher><publisher-name>MISIS</publisher-name></publisher></journal-meta><article-meta><article-id pub-id-type="doi">10.17073/1609-3577-2019-4-272-278</article-id><article-id custom-type="elpub" pub-id-type="custom">mateltech-346</article-id><article-categories><subj-group subj-group-type="heading"><subject>Research Article</subject></subj-group><subj-group subj-group-type="section-heading" xml:lang="ru"><subject>Математическое моделирование в материаловедении электронных компонентов</subject></subj-group><subj-group subj-group-type="section-heading" xml:lang="en"><subject>MATHEMATICAL MODELING IN MATERIALS SCIENCE OF ELECTRONIC COMPONENTS</subject></subj-group></article-categories><title-group><article-title>Вопросы реализации нейросетевых алгоритмов на мемристорных кроссбарах</article-title><trans-title-group xml:lang="en"><trans-title>Issues of implementing neural network algorithms on memristor crossbars</trans-title></trans-title-group></title-group><contrib-group><contrib contrib-type="author" corresp="yes"><contrib-id contrib-id-type="orcid">https://orcid.org/0000-0003-0364-8665</contrib-id><name-alternatives><name name-style="eastern" xml:lang="ru"><surname>Морозов</surname><given-names>А. Ю.</given-names></name><name name-style="western" xml:lang="en"><surname>Morozov</surname><given-names>A. Yu.</given-names></name></name-alternatives><bio xml:lang="ru"/><bio xml:lang="en"/><email xlink:type="simple">morozov@infway.ru</email><xref ref-type="aff" rid="aff-1"/></contrib><contrib contrib-type="author" corresp="yes"><contrib-id contrib-id-type="orcid">https://orcid.org/0000-0003-0998-7975</contrib-id><name-alternatives><name name-style="eastern" xml:lang="ru"><surname>Ревизников</surname><given-names>Д. Л.</given-names></name><name name-style="western" xml:lang="en"><surname>Reviznikov</surname><given-names>D. L.</given-names></name></name-alternatives><bio xml:lang="ru"/><bio xml:lang="en"/><email xlink:type="simple">reviznikov@gmail.com</email><xref ref-type="aff" rid="aff-1"/></contrib><contrib contrib-type="author" corresp="yes"><contrib-id contrib-id-type="orcid">https://orcid.org/0000-0002-0059-0712</contrib-id><name-alternatives><name name-style="eastern" xml:lang="ru"><surname>Абгарян</surname><given-names>К. К.</given-names></name><name name-style="western" xml:lang="en"><surname>Abgaryan</surname><given-names>K. K.</given-names></name></name-alternatives><bio xml:lang="ru"/><bio xml:lang="en"/><email xlink:type="simple">kristal83@mail.ru</email><xref ref-type="aff" rid="aff-1"/></contrib></contrib-group><aff-alternatives id="aff-1"><aff xml:lang="ru"><institution>Федеральный исследовательский центр «Информатика и управление» &#13;
Российской академии наук, &#13;
ул. Вавилова, д. 44, корп. 2, Москва, 119333, Россия;&#13;
Московский авиационный институт (национальный исследовательский университет), &#13;
Волоколамское шоссе, д. 4, Москва, 125993, Россия</institution><country>Россия</country></aff><aff xml:lang="en"><institution>Federal Research Centre “Information and Control” of the Russian Academy of Sciences, &#13;
44 Vavilov Str., Moscow 119333, Russia;&#13;
Moscow Aviation Institute (National Research University), &#13;
4 Volokolamskoe shosse, 4, Moscow, 125993, Russia</institution><country>Russian Federation</country></aff></aff-alternatives><pub-date pub-type="collection"><year>2019</year></pub-date><pub-date pub-type="epub"><day>07</day><month>04</month><year>2020</year></pub-date><volume>22</volume><issue>4</issue><fpage>272</fpage><lpage>278</lpage><permissions><copyright-statement>Copyright &amp;#x00A9; Морозов А.Ю., Ревизников Д.Л., Абгарян К.К., 2020</copyright-statement><copyright-year>2020</copyright-year><copyright-holder xml:lang="ru">Морозов А.Ю., Ревизников Д.Л., Абгарян К.К.</copyright-holder><copyright-holder xml:lang="en">Morozov A.Y., Reviznikov D.L., Abgaryan K.K.</copyright-holder><license xml:lang="ru" license-type="creative-commons-attribution" xlink:href="https://creativecommons.org/licenses/by/4.0/" xlink:type="simple"><license-p>Данная работа распространяется под лицензией Creative Commons Attribution 4.0.</license-p></license><license xml:lang="en" license-type="creative-commons-attribution" xlink:href="https://creativecommons.org/licenses/by/4.0/" xlink:type="simple"><license-p>This work is licensed under a Creative Commons Attribution 4.0 License.</license-p></license></permissions><self-uri xlink:href="https://met.misis.ru/jour/article/view/346">https://met.misis.ru/jour/article/view/346</self-uri><abstract><p>Присущее мемристорным кроссбарам свойство естественной параллелизации матрично-векторных операций создает возможности для их эффективного использования в нейросетевых вычислениях. Аналоговые вычисления производятся на порядки быстрее по сравнению с вычислениями на центральном процессоре и на графических ускорителях. Кроме того, значительно ниже энергозатраты на проведение математических операций. При этом существенной особенностью аналоговых вычислений является небольшая точность. В связи с этим актуальным является исследование зависимости качества работы нейронной сети от точности задания ее весов. Рассмотрены две сверточные нейронные сети, обученные на наборах данных MNIST (рукописные цифры) и CIFAR_10 (самолеты, лодки, машины и т. д.). Первая состоит из двух сверточных слоев, одного слоя подвыборки и двух полносвязанных слоев, а вторая — из четырех сверточных слоев, двух слоев подвыборки и двух полносвязаных слоев. Вычисления в сверточных и полносвязных слоях выполняются через матрично-векторные операции, которые эффективно реализуются на мемристорных кроссбарах. Слои подвыборки подразумевают операцию нахождения максимального значения из нескольких, которая также может быть реализована на аналоговом уровне. Процесс обучения нейронной сети происходит отдельно от анализа данных. Как правило, на этапе обучения используются градиентные методы оптимизации, реализацию которых целесообразно выполнять на центральном процессоре. Показано, что для получения приемлемого качества распознавания в случае с сетью, обученной на MNIST, требуется 3—4 бита точности при задании ее весов, а в случае с сетью, обученной на CIFAR_10, — 6—8 бит.</p></abstract><trans-abstract xml:lang="en"><p>The property of natural parallelization of matrix-vector operations inherent in memristor crossbars creates opportunities for their effective use in neural network computing. Analog calculations are orders of magnitude faster in comparison to calculations on the central processor and on graphics accelerators. Besides, mathematical operations energy costs are significantly lower. The essential feature of analog computing is its low accuracy. In this regard, studying the dependence of neural network quality on the accuracy of setting its weights is relevant. The paper considers two convolutional neural networks trained on the MNIST (handwritten digits) and CIFAR_10 (airplanes, boats, cars, etc.) data sets. The first convolutional neural network consists of two convolutional layers, one subsample layer and two fully connected layers. The second one consists of four convolutional layers, two subsample layers and two fully connected layers. Calculations in convolutional and fully connected layers are performed through matrix-vector operations that are implemented on memristor crossbars. Sub-sampling layers imply the operation of finding the maximum value from several values. This operation can be implemented at the analog level. The process of training a neural network runs separately from data analysis. As a rule, gradient optimization methods are used at the training stage. It is advisable to perform calculations using these methods on CPU. When setting the weights, 3—4 precision bits are required to obtain an acceptable recognition quality in the case the network is trained on MNIST. 6-10 precision bits are required if the network is trained on CIFAR_10.</p></trans-abstract><kwd-group xml:lang="ru"><kwd>мемристор</kwd><kwd>кроссбар</kwd><kwd>точность</kwd><kwd>нейросеть</kwd><kwd>свертка</kwd><kwd>MNIST</kwd><kwd>CIFAR_10</kwd></kwd-group><kwd-group xml:lang="en"><kwd>memristor</kwd><kwd>crossbar</kwd><kwd>accuracy</kwd><kwd>neural network</kwd><kwd>convolution</kwd><kwd>MNIST</kwd><kwd>CIFAR_10</kwd></kwd-group><funding-group><funding-statement xml:lang="ru">Работа выполнена при поддержке гранта РФФИ № 19-29-03051 мк.</funding-statement><funding-statement xml:lang="en">This work was supported by the RFBR grant No. 19-29-03051 MK.</funding-statement></funding-group></article-meta></front><back><ref-list><title>References</title><ref id="cit1"><label>1</label><citation-alternatives><mixed-citation xml:lang="ru">Wong H.-S. P., Lee H.-Y., Yu S., Chen Y.-S., Wu Y., Chen P.-S., Lee B., Chen F. T., Tsai M.-J. Metal-oxide RRAM // Proceedings of the IEEE. 2012. V. 100, Iss. 6. P. 1951—1970. DOI: 10.1109/JPROC.2012.2190369</mixed-citation><mixed-citation xml:lang="en">Wong H.-S. P., Lee H.-Y., Yu S., Chen Y.-S., Wu Y., Chen P.-S., Lee B., Chen F. T., Tsai M.-J. Metal-oxide RRAM. Proceedings of the IEEE, 2012, vol. 100, no. 6, pp. 1951—1970. DOI: 10.1109/JPROC.2012.2190369</mixed-citation></citation-alternatives></ref><ref id="cit2"><label>2</label><citation-alternatives><mixed-citation xml:lang="ru">Yang J. J., Strukov D. B., Stewart D. R., Memristive devices for computing // Nature Nanotechnology. 2013. V. 8, N 1. P. 13—24. DOI: 10.1038/nnano.2012.240</mixed-citation><mixed-citation xml:lang="en">Yang J. J., Strukov D. B., Stewart D. R., Memristive devices for computing. Nature Nanotechnology, 2013, vol. 8, no. 1, pp. 13—24. DOI: 10.1038/nnano.2012.240</mixed-citation></citation-alternatives></ref><ref id="cit3"><label>3</label><citation-alternatives><mixed-citation xml:lang="ru">Li C., Hu M., Li Y., Jiang H., Ge N., Montgomery E., Zhang J., Song W., Dávila N., Graves C. E., Li Z., Strachan J. P., Lin P., Wang Z., Barnell M., Wu Q., Williams R. S., Yang J. J., Xia Q. Analogue signal and image processing with large memristor crossbars // Nature Electronics. 2018. V. 1, N 1. P. 52—59. DOI: 10.1038/s41928-017-0002-z</mixed-citation><mixed-citation xml:lang="en">Li C., Hu M., Li Y., Jiang H., Ge N., Montgomery E., Zhang J., Song W., Dávila N., Graves C. E., Li Z., Strachan J. P., Lin P., Wang Z., Barnell M., Wu Q., Williams R. S., Yang J. J., Xia Q. Analogue signal and image processing with large memristor crossbars. Nature Electronics, 2018, vol. 1, no. 1, pp. 52—59. DOI: 10.1038/s41928-017-0002-z</mixed-citation></citation-alternatives></ref><ref id="cit4"><label>4</label><citation-alternatives><mixed-citation xml:lang="ru">Hu M, Graves C. E., Li C., Li Y., Ge N., Montgomery E., Dávila N., Jiang H., Williams R. S., Yang J. J., Xia O., Strachan J. P. Memristor-based analog computation and neural network classification with a dot product engine // Advanced Materials. 2018. V. 30, Iss. 9. P. 1705914. DOI: 10.1002/adma.201705914</mixed-citation><mixed-citation xml:lang="en">Hu M, Graves C. E., Li C., Li Y., Ge N., Montgomery E., Dávila N., Jiang H., Williams R. S., Yang J. J., Xia O., Strachan J. P. Memristor-based analog computation and neural network classification with a dot product engine. Advanced Materials, 2018, vol. 30, no. 9, p. 1705914. DOI: 10.1002/adma.201705914</mixed-citation></citation-alternatives></ref><ref id="cit5"><label>5</label><citation-alternatives><mixed-citation xml:lang="ru">Тарков М. С. Реализация нейронной WTA-сети на мемристорном кроссбаре // ПДМ. Приложение, 2015, Вып. 8. С. 151—154. DOI: 10.17223/2226308X/8/59</mixed-citation><mixed-citation xml:lang="en">Tarkov M. S. Implementation of a neural WTA-network on the memristor crossbar. Prikl. Diskr. Mat. Suppl., 2015, no. 8, pp. 151—154. (In. Russ.). DOI: 10.17223/2226308X/8/59</mixed-citation></citation-alternatives></ref><ref id="cit6"><label>6</label><citation-alternatives><mixed-citation xml:lang="ru">Diehl P. U., Cook M. Unsupervised learning of digit recognition using spike-timing-dependent plasticity // Front. Comput. Neurosci. 2015. V. 9. Art. No. 99. P. 9. DOI: 10.3389/fncom.2015.00099</mixed-citation><mixed-citation xml:lang="en">Diehl P. U., Cook M. Unsupervised learning of digit recognition using spike-timing-dependent plasticity. Front. Comput. Neurosci., 2015, vol. 9, art. 99, p. 9. DOI: 10.3389/fncom.2015.00099</mixed-citation></citation-alternatives></ref><ref id="cit7"><label>7</label><citation-alternatives><mixed-citation xml:lang="ru">Ambrogio S., Balatti S., Milo V., Carboni R., Wang Z.-Q., Calderoni A., Ramaswamy N., Ielmini D. Neuromorphic learning and recognition with one-transistor-one-resistor synapses and bistable metal oxide RRAM // IEEE Transactions on Electron Devices. 2016. V. 63, N 4. P. 1508—1515. DOI: 10.1109/TED.2016.2526647</mixed-citation><mixed-citation xml:lang="en">Ambrogio S., Balatti S., Milo V., Carboni R., Wang Z.-Q., Calderoni A., Ramaswamy N., Ielmini D. Neuromorphic learning and recognition with one-transistor-one-resistor synapses and bistable metal oxide RRAM. IEEE Transactions on Electron Devices, 2016, vol. 63, no. 4, pp. 1508—1515. DOI: 10.1109/TED.2016.2526647</mixed-citation></citation-alternatives></ref><ref id="cit8"><label>8</label><citation-alternatives><mixed-citation xml:lang="ru">Guo Y., Wu H., Gao B., Qian H. Unsupervised learning on resistive memory array based spiking neural networks // Front. Neurosci. 2019. V. 13, Art. No. 812. DOI: 10.3389/fnins.2019.00812</mixed-citation><mixed-citation xml:lang="en">Guo Y., Wu H., Gao B., Qian H. Unsupervised learning on resistive memory array based spiking neural networks. Front. Neurosci., 2019, vol. 13, art. 812. DOI: 10.3389/fnins.2019.00812</mixed-citation></citation-alternatives></ref><ref id="cit9"><label>9</label><citation-alternatives><mixed-citation xml:lang="ru">Milo V., Pedretti G., Laudato M., Bricalli A., Ambrosi E., Bianchi S., Chicca E., Ielmini D. Resistive switching synapses for unsupervised learning in feed-forward and recurrent neural networks // IEEE International Symposium on Circuits and Systems (ISCAS). Florence (Italy): IEEE, 2018. P. 1—5. DOI: 10.1109/ISCAS.2018.8351824</mixed-citation><mixed-citation xml:lang="en">Milo V., Pedretti G., Laudato M., Bricalli A., Ambrosi E., Bianchi S., Chicca E., Ielmini D. Resistive switching synapses for unsupervised learning in feed-forward and recurrent neural networks. In: IEEE International Symposium on Circuits and Systems (ISCAS). Florence (Italy): IEEE, 2018, pp. 1—5. DOI: 10.1109/ISCAS.2018.8351824</mixed-citation></citation-alternatives></ref><ref id="cit10"><label>10</label><citation-alternatives><mixed-citation xml:lang="ru">Pedretti G., Bianchi S., Milo V., Calderoni A., Ramaswamy N., Ielmini D. Modeling-based design of brain-inspired spiking neural networks with RRAM learning synapses // IEEE International Electron Devices Meeting (IEDM). San Francisco (CA, USA): IEEE, 2017. P. 28.1.1—28.1.4. DOI: 10.1109/IEDM.2017.8268467</mixed-citation><mixed-citation xml:lang="en">Pedretti G., Bianchi S., Milo V., Calderoni A., Ramaswamy N., Ielmini D. Modeling-based design of brain-inspired spiking neural networks with RRAM learning synapses. In: IEEE International Electron Devices Meeting (IEDM). San Francisco (CA, USA): IEEE, 2017, pp. 28.1.1—28.1.4. DOI: 10.1109/IEDM.2017.8268467</mixed-citation></citation-alternatives></ref><ref id="cit11"><label>11</label><citation-alternatives><mixed-citation xml:lang="ru">Milo V., Ielmini D., Chicca E. Attractor networks and associative memories with STDP learning in RRAM synapses // IEEE International Electron Devices Meeting (IEDM). San Francisco (CA, USA): IEEE, 2017. P. 11.2.1—11.2.4. DOI: 10.1109/IEDM.2017.8268369</mixed-citation><mixed-citation xml:lang="en">Milo V., Ielmini D., Chicca E. Attractor networks and associative memories with STDP learning in RRAM synapses. In: IEEE International Electron Devices Meeting (IEDM). San Francisco (CA, USA): IEEE, 2017, pp. 11.2.1—11.2.4. DOI: 10.1109/IEDM.2017.8268369</mixed-citation></citation-alternatives></ref><ref id="cit12"><label>12</label><citation-alternatives><mixed-citation xml:lang="ru">Li B., Shan Y., Hu M., Wang Y., Chen Y., Yang H., Memristor-based approximated computation // International Symposium on Low Power Electronics and Design (ISLPED). Beijing (China): IEEE, 2013. P. 242—247. DOI: 10.1109/ISLPED.2013.6629302</mixed-citation><mixed-citation xml:lang="en">Li B., Shan Y., Hu M., Wang Y., Chen Y., Yang H., Memristor-based approximated computation. In: International Symposium on Low Power Electronics and Design (ISLPED). Beijing (China): IEEE, 2013, pp. 242—247. DOI: 10.1109/ISLPED.2013.6629302</mixed-citation></citation-alternatives></ref><ref id="cit13"><label>13</label><citation-alternatives><mixed-citation xml:lang="ru">Теплов Г. С., Горнев Е. С. Модель на языке Verilog-A многоуровневого биполярного мемристора с учетом девиаций параметров переключения // Микроэлектроника. 2019. Т. 48, № 3. С. 163—175. DOI: 10.1134/S0544126919030104</mixed-citation><mixed-citation xml:lang="en">Teplov G. S., Gornev E. S. Multilevel bipolar memristor model considering deviations of switching parameters in the Verilog-A language. Russ. Microelectron., 2019, vol. 48, no. 3, pp. 131—142. DOI: 10.1134/S1063739719030107</mixed-citation></citation-alternatives></ref><ref id="cit14"><label>14</label><citation-alternatives><mixed-citation xml:lang="ru">Morozov A. Yu., Reviznikov D. L. Adaptive interpolation algorithm based on a kd-tree for numerical integration of systems of ordinary differential equations with interval initial conditions // Differential Equations. 2018. V. 54, N 7. P. 945—956. DOI: 10.1134/S0012266118070121</mixed-citation><mixed-citation xml:lang="en">Morozov A. Yu., Reviznikov D. L. Adaptive interpolation algorithm based on a kd-tree for numerical integration of systems of ordinary differential equations with interval initial conditions. Differential Equations., 2018, vol. 54, no. 7, pp. 945—956. DOI: 10.1134/S0012266118070121</mixed-citation></citation-alternatives></ref><ref id="cit15"><label>15</label><citation-alternatives><mixed-citation xml:lang="ru">Morozov A. Yu., Reviznikov D. L., Gidaspov V. Yu. Adaptive interpolation algorithm based on a kd-tree for the problems of chemical kinetics with interval parameters // Mathematical Models and Computer Simulations. 2019. V. 11, N 4. P. 622—633. DOI: 10.1134/S2070048219040100</mixed-citation><mixed-citation xml:lang="en">Morozov A. Yu., Reviznikov D. L., Gidaspov V. Yu. Adaptive interpolation algorithm based on a kd-tree for the problems of chemical kinetics with interval parameters. Mathematical Models and Computer Simulations, 2019, vol. 11, no. 4, pp. 622—633. DOI: 10.1134/S2070048219040100</mixed-citation></citation-alternatives></ref><ref id="cit16"><label>16</label><citation-alternatives><mixed-citation xml:lang="ru">Morozov A. Yu., Reviznikov D. L. Modelling of dynamic systems with interval parameters on graphic processors // Программная инженерия. 2019. Т. 10, № 2. С. 69—76. DOI: 10.17587/prin.10.69-76</mixed-citation><mixed-citation xml:lang="en">Morozov A. Yu., Reviznikov D. L. Modelling of dynamic systems with interval parameters on graphic processors. Programmnaya Ingeneria = Software Engineering, 2019, vol. 10, no. 2, pp. 69—76. DOI: 10.17587/prin.10.69-76</mixed-citation></citation-alternatives></ref><ref id="cit17"><label>17</label><citation-alternatives><mixed-citation xml:lang="ru">Морозов А. Ю., Ревизников Д. Л. Методы компьютерного моделирования динамических систем с интервальными параметрами. М.: Изд-во МАИ, 2019. 160 с.</mixed-citation><mixed-citation xml:lang="en">Morozov A. Yu., Reviznikov D. L. Metody komp’yuternogo modelirovaniya dinamicheskikh sistem s interval’nymi parametrami [Methods for computer modeling of dynamic systems with interval parameters ]. Moscow: Izd-vo Moskovskogo aviatsionnogo instituta, 2019, 160 p. (In Russ.)</mixed-citation></citation-alternatives></ref><ref id="cit18"><label>18</label><citation-alternatives><mixed-citation xml:lang="ru">Gulli A., Pal S. Deep learning with Keras: implement neural networks with Keras on Theano and TensorFlow. Birmingham; Mumbai: Packt Publishing Ltd., 2017. 490 p.</mixed-citation><mixed-citation xml:lang="en">Gulli A., Pal S. Deep learning with Keras: implement neural networks with Keras on Theano and TensorFlow. Birmingham; Mumbai: Packt Publishing Ltd., 2017, 490 p.</mixed-citation></citation-alternatives></ref><ref id="cit19"><label>19</label><citation-alternatives><mixed-citation xml:lang="ru">MNIST CNN. URL: https://keras.io/examples/mnist_cnn/ (дата обращения: 01.10.2019).</mixed-citation><mixed-citation xml:lang="en">MNIST CNN. URL: https://keras.io/examples/mnist_cnn/ (accessed: 01.10.2019).</mixed-citation></citation-alternatives></ref><ref id="cit20"><label>20</label><citation-alternatives><mixed-citation xml:lang="ru">Train a simple deep CNN on the CIFAR10 small images dataset. URL: https://keras.io/examples/cifar10_cnn/ (дата обращение: 01.10.2019).</mixed-citation><mixed-citation xml:lang="en">Train a simple deep CNN on the CIFAR10 small images dataset. URL: https://keras.io/examples/cifar10_cnn/ (accessed: 01.10.2019).</mixed-citation></citation-alternatives></ref></ref-list><fn-group><fn fn-type="conflict"><p>The authors declare that there are no conflicts of interest present.</p></fn></fn-group></back></article>
