CoChat

In‐Memory Vector‐Matrix Multiplication in Monolithic Complementary Metal–Oxide–Semiconductor‐Memristor Integrated Circuits: Design Choices, Challenges, and Perspectives

Open access

The DOI matches its Crossref record; 93 of the 93 references it cites were found in a registry or an index.

Abstract

The low communication bandwidth between memory and processing units in conventional von Neumann machines does not support the requirements of emerging applications that rely extensively on large sets of data. More recent computing paradigms, such as high parallelization and near‐memory computing, help alleviate the data communication bottleneck to some extent, but paradigm‐shifting concepts are required. In‐memory computing has emerged as a prime candidate to eliminate this bottleneck by colocating memory and processing. In this context, resistive switching (RS) memory devices is a key promising choice, due to their unique intrinsic device‐level properties, enabling both storing and computing with a small, massively‐parallel footprint at low power. Theoretically, this directly translates to a major boost in energy efficiency and computational throughput, but various practical challenges remain. A qualitative and quantitative analysis of several key existing challenges in implementing high‐capacity, high‐volume RS memories for accelerating the most computationally demanding computation in machine learning (ML) inference, that of vector‐matrix multiplication (VMM), is presented. The monolithic integration of RS memories with complementary metal–oxide–semiconductor (CMOS) integrated circuits is presented as the core underlying technology. The key existing design choices in terms of device‐level physical implementation, circuit‐level design, and system‐level considerations is reviewed and an outlook for future directions is provided.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1002/aisy.202000115

Cites

93 of 93 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Various computing hardware performance overview as well as challenges and limitations hindering the path for monolithic CMOS-memristor VMM integrated circuits to become mainstream AI hardware. (a) A simple view of application processing unit (APU) platform's energy efficiency performance and its flexibility in terms of the application versatility is compared with conventional platforms like CPUs and GPUs. Different RS-based APU classes with low to high resolution weight networks are depicted in terms of energy efficiency and application spectrum flexibility. At the opposite of the trade-off between flexibility and energy that existing hardware exhibit, Spiking Neural Networks (SNN), inspired by biology combine both flexibility and low energy consumption. Finding the keys for this implementation may be a disruptive direction for future hardware design. (b) The challenges have been divided into three different categories: physical constraints, circuit-level challenges, and system-level challenges. (c) Manuscript organization.
Figure 1 Various computing hardware performance overview as well as challenges and limitations hindering the path for monolithic CMOS-memristor VMM integrated circuits to become mainstream AI hardware. (a) A simple view of application processing unit (APU) platform's energy efficiency performance and its flexibility in terms of the application versatility is compared with conventional platforms like CPUs and GPUs. Different RS-based APU classes with low to high resolution weight networks are depicted in terms of energy efficiency and application spectrum flexibility. At the opposite of the trade-off between flexibility and energy that existing hardware exhibit, Spiking Neural Networks (SNN), inspired by biology combine both flexibility and low energy consumption. Finding the keys for this implementation may be a disruptive direction for future hardware design. (b) The challenges have been divided into three different categories: physical constraints, circuit-level challenges, and system-level challenges. (c) Manuscript organization.CC-BY · From the PDF
Figure 2 (a) A basic neural network structure is shown including the input vector, weight matrix and output vector. (b) Schematic of the digital and analog vector-matrix multiplication and their implementations. VMM digital implementation is realized by pipelining multiple adders and multiplier digital blocks. Analog VMM on M × N RS-based crossbar is realized by summing currents from M lines in N columns. (c) The physical implementation of RS-based VMM engine shows the input vector is applied as a voltage vector into the word-lines of the array (bottom electrode (BE)), the weight matrix is stored as the RS devices conductances and the output is sensed as accumulated current in the bit-line (top electrode (TE)). (d) 3D illustration of an RS-based crossbar monolithically integrated on top of the CMOS substrate using a back-end-of-line (BEOL) process.
Figure 2 (a) A basic neural network structure is shown including the input vector, weight matrix and output vector. (b) Schematic of the digital and analog vector-matrix multiplication and their implementations. VMM digital implementation is realized by pipelining multiple adders and multiplier digital blocks. Analog VMM on M × N RS-based crossbar is realized by summing currents from M lines in N columns. (c) The physical implementation of RS-based VMM engine shows the input vector is applied as a voltage vector into the word-lines of the array (bottom electrode (BE)), the weight matrix is stored as the RS devices conductances and the output is sensed as accumulated current in the bit-line (top electrode (TE)). (d) 3D illustration of an RS-based crossbar monolithically integrated on top of the CMOS substrate using a back-end-of-line (BEOL) process.CC-BY · From the PDF
Figure 3 (a) In this spider diagram, resistive switching memories, DRAM, SRAM, and Flash memories are compared in terms of the cost, read time, write time, energy consumption, endurance and retention. (b) In this diagram, the same memories are compared in terms of other criteria: flexibility, footprint size, maturity, density, variability, and potential of the scalability. (c) The i -v curve of the prototypical digital HfOx device with its sharp switching behavior in SET and RESET regions is depicted. (d) The switching behavior for the prototypical memristive TiOx analog device 58 is shown. Both RRAM and memristive devices belong to the resistive switching memories family and described in the spider diagrams in sub-figures (a) and (b) .
Figure 3 (a) In this spider diagram, resistive switching memories, DRAM, SRAM, and Flash memories are compared in terms of the cost, read time, write time, energy consumption, endurance and retention. (b) In this diagram, the same memories are compared in terms of other criteria: flexibility, footprint size, maturity, density, variability, and potential of the scalability. (c) The i -v curve of the prototypical digital HfOx device with its sharp switching behavior in SET and RESET regions is depicted. (d) The switching behavior for the prototypical memristive TiOx analog device 58 is shown. Both RRAM and memristive devices belong to the resistive switching memories family and described in the spider diagrams in sub-figures (a) and (b) .CC-BY · From the PDF
Box 1: Analysis of non-ideal parameters of RS memories that impact neural network accuracy
Box 1: Analysis of non-ideal parameters of RS memories that impact neural network accuracyCC-BY · From the PDF
Figure 5 Scalability challenges and RC network Elmore delay model for RS crossbar array. (a) The scalability challenges: Both bit-line current and word/bit-line resistance increase with the size of the RS crossbar array. (b) RS crossbar RC Elmore delay model is shown by dividing the array delay into four regions corresponding to t 1 , t 2 , t 3 , and t 4 which are the delays from a row, RS cell, column and output circuit, respectively. (c) Increasing the number of rows increases the accumulated current in the column and can become a major challenge for output circuits design. The same limitation applies for the large number of columns required to inject large current into the row and affecting input circuits design. (d) The line resistance is another challenge for scalability of RS-based arrays due to the voltage degradation in the rows ( D V ROW ) and columns ( D V COL ). This issue can be leveraged by engineering optimization and/or compensated by input/output circuits strategies.
Figure 5 Scalability challenges and RC network Elmore delay model for RS crossbar array. (a) The scalability challenges: Both bit-line current and word/bit-line resistance increase with the size of the RS crossbar array. (b) RS crossbar RC Elmore delay model is shown by dividing the array delay into four regions corresponding to t 1 , t 2 , t 3 , and t 4 which are the delays from a row, RS cell, column and output circuit, respectively. (c) Increasing the number of rows increases the accumulated current in the column and can become a major challenge for output circuits design. The same limitation applies for the large number of columns required to inject large current into the row and affecting input circuits design. (d) The line resistance is another challenge for scalability of RS-based arrays due to the voltage degradation in the rows ( D V ROW ) and columns ( D V COL ). This issue can be leveraged by engineering optimization and/or compensated by input/output circuits strategies.CC-BY · From the PDF
Figure 6 (a) General RS-based current-mode VMM architecture with individually allocated input circuit for each word-line. (b) Shared input circuit architecture is displayed with a sample schematic of resistive DAC with OP-AMP follower.
Figure 6 (a) General RS-based current-mode VMM architecture with individually allocated input circuit for each word-line. (b) Shared input circuit architecture is displayed with a sample schematic of resistive DAC with OP-AMP follower.CC-BY · From the PDF
Figure 7 (a) The DAC area usages versus the sample frequency and effective number of bits. In the following two specific cases has been described by sub figures. (i) Area usage versus sample frequency at effective number of bit (ENOB) equals 12. (ii) Area usage versus ENOB at sample frequency of 100 MS/s. (b) The DAC power consumption versus sample frequency and device count (the number of memristor devices driven by one DAC), assuming the total parasitic capacitance is 10 pF. In the following two specific cases has been described by sub figures. (iii) The power dissipation versus sample frequency when the device count is 500. (iv) The device count versus power consumption at sample frequency of 100 MS/s.
Figure 7 (a) The DAC area usages versus the sample frequency and effective number of bits. In the following two specific cases has been described by sub figures. (i) Area usage versus sample frequency at effective number of bit (ENOB) equals 12. (ii) Area usage versus ENOB at sample frequency of 100 MS/s. (b) The DAC power consumption versus sample frequency and device count (the number of memristor devices driven by one DAC), assuming the total parasitic capacitance is 10 pF. In the following two specific cases has been described by sub figures. (iii) The power dissipation versus sample frequency when the device count is 500. (iv) The device count versus power consumption at sample frequency of 100 MS/s.CC-BY · From the PDF
Figure 8 Main aspects trade-off of the ADC published in the International Solid-State Circuits Conference (ISSCC) from 1997 to 2020.(a) ADC area increases with SNDR and is classified based on different ADC architectures. (b) ADC power consumption increments with the increase of Nyquist sampling frequency and marked based on different resolutions. (c) ADC power consumption versus the SNDR has been illustrated for different design operating frequencies. (d) ADC sampling frequency versus the SNDR has been shown for different ADC architectures. (e) ADC FOM versus sampling frequency is depicted and classified based on CMOS technology node (in m m). (f) ADC area versus FOM is shown and categorized based on different CMOS technology node.
Figure 8 Main aspects trade-off of the ADC published in the International Solid-State Circuits Conference (ISSCC) from 1997 to 2020.(a) ADC area increases with SNDR and is classified based on different ADC architectures. (b) ADC power consumption increments with the increase of Nyquist sampling frequency and marked based on different resolutions. (c) ADC power consumption versus the SNDR has been illustrated for different design operating frequencies. (d) ADC sampling frequency versus the SNDR has been shown for different ADC architectures. (e) ADC FOM versus sampling frequency is depicted and classified based on CMOS technology node (in m m). (f) ADC area versus FOM is shown and categorized based on different CMOS technology node.CC-BY · From the PDF
Figure 9 Design choices for RS-based VMM engines are defined based on the combination of the input, weight cell, and output precision targeted for specific applications. Here, a combination lock is shown as VMM design choices, which may be unlocked with different combination of the input, weight, and output. The input, weight and output choices are binary, ternary, multi-bit, and analog input. Here, we have depicted four examples of different design choices from the recent fully integrated CMOS/RS-based chips 76,77,89,90 .
Figure 9 Design choices for RS-based VMM engines are defined based on the combination of the input, weight cell, and output precision targeted for specific applications. Here, a combination lock is shown as VMM design choices, which may be unlocked with different combination of the input, weight, and output. The input, weight and output choices are binary, ternary, multi-bit, and analog input. Here, we have depicted four examples of different design choices from the recent fully integrated CMOS/RS-based chips 76,77,89,90 .CC-BY · From the PDF
Figure 10 Three different computing architectures are shown with their corresponding data movement (input, weight and output) to carry out vector matrix multiplication operation. (a) Conventional von Neumann computing architecture is depicted comprising of processing unit and conventional memory. It has been shown that a high data movement for both inputs and weights as data needed to be fetched from or stored in the memory at different stages of the operation. Also, the digital MAC increases the computation time as several consecutive digital operations will be needed to perform large VMM. (b) Near-memory computing architecture (NMC) is shown in this part and, in addition to the main processing unit, near memory processing units (NMPU) have been placed in vicinity of DRAM and NVM blocks. This reduces the data movement cost significantly as the commute distance of data is reduced by placing the processing unit close to the memory. Although in NMC the distance of memory to processing unit is decreased, there is still a significant amount of data commute in between for input, weight and output data. Also, the problem of high computation time due to the digital MAC exists. (c) In-memory computing architectures (IMC) implement computing within the memory. Specifically, in RS-based IMC, the RS-array can implement highly parallel VMM operation in one step and it also stored the weight matrix which will completely omit the weight movement during the operations. The only data movement in the IMC corresponds to input data. In-memory VMM (iVMM) is implemented over RS-based array in fully parallel manner by implementing several parallel in-memory MAC (iMAC) operations.
Figure 10 Three different computing architectures are shown with their corresponding data movement (input, weight and output) to carry out vector matrix multiplication operation. (a) Conventional von Neumann computing architecture is depicted comprising of processing unit and conventional memory. It has been shown that a high data movement for both inputs and weights as data needed to be fetched from or stored in the memory at different stages of the operation. Also, the digital MAC increases the computation time as several consecutive digital operations will be needed to perform large VMM. (b) Near-memory computing architecture (NMC) is shown in this part and, in addition to the main processing unit, near memory processing units (NMPU) have been placed in vicinity of DRAM and NVM blocks. This reduces the data movement cost significantly as the commute distance of data is reduced by placing the processing unit close to the memory. Although in NMC the distance of memory to processing unit is decreased, there is still a significant amount of data commute in between for input, weight and output data. Also, the problem of high computation time due to the digital MAC exists. (c) In-memory computing architectures (IMC) implement computing within the memory. Specifically, in RS-based IMC, the RS-array can implement highly parallel VMM operation in one step and it also stored the weight matrix which will completely omit the weight movement during the operations. The only data movement in the IMC corresponds to input data. In-memory VMM (iVMM) is implemented over RS-based array in fully parallel manner by implementing several parallel in-memory MAC (iMAC) operations.CC-BY · From the PDF
Figure 11 (a) Different RS-based APUs have been compared in terms of energy efficiency, storage efficiency, and flexibility for the particular case of the VMM operation. Each implementation presents a balance between memory functionalities (from binary to analog) and CMOS circuits overhead complexity and cost. (b) VMM engine design 88 with 0T1R analog weight network is illustrated based on input amplitude encoding and its corresponding sensing circuit with a feedback resistor in an op-amp follower block in the bit-line. (c) VMM engine design 95 with 0T1R analog weight network by utilizing input pulse duration encoding scheme and its corresponding sensing circuit for amplitude encoded analog output is depicted. (d) VMM engine design concept 96 for 1T1R weight network by using digital input pulse duration encoding scheme and its corresponding sensing circuit for pulse duration encoded digital output.
Figure 11 (a) Different RS-based APUs have been compared in terms of energy efficiency, storage efficiency, and flexibility for the particular case of the VMM operation. Each implementation presents a balance between memory functionalities (from binary to analog) and CMOS circuits overhead complexity and cost. (b) VMM engine design 88 with 0T1R analog weight network is illustrated based on input amplitude encoding and its corresponding sensing circuit with a feedback resistor in an op-amp follower block in the bit-line. (c) VMM engine design 95 with 0T1R analog weight network by utilizing input pulse duration encoding scheme and its corresponding sensing circuit for amplitude encoded analog output is depicted. (d) VMM engine design concept 96 for 1T1R weight network by using digital input pulse duration encoding scheme and its corresponding sensing circuit for pulse duration encoded digital output.CC-BY · From the PDF
Image 12
CC-BY · From the PDF

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

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?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 93 references

Research works printed but not linked: 41

  1. 2. Horowitz, M. 1.1 computing's energy problem (and what we can do about it). In 2014 IEEE International Solid-State Circuits Conference Digest of Technical Papers (ISSCC) , 10-14 (IEEE, 2014).
  2. 8. Sze, V., Chen, Y.-H., Emer, J., Suleiman, A. & Zhang, Z. Hardware for machine learning: Challenges and opportunities. In 2017 IEEE Custom Integrated Circuits Conference (CICC) , 1-8 (IEEE, 2017). · DOI · Paper page
  3. 16. Bojnordi, M. N. & Ipek, E. Memristive boltzmann machine: A hardware accelerator for combinatorial optimization and deep learning. In 2016 IEEE International Symposium on High Performance Computer Architecture (HPCA) , 1-13 (IEEE, 2016). · DOI · Paper page
  4. 19. Shin, J. H., Jeong, Y. J., Zidan, M. A., Wang, Q. & Lu, W. D. Hardware acceleration of simulated annealing of spin glass by rram crossbar array. In 2018 IEEE International Electron Devices Meeting (IEDM) , 3-3 (IEEE, 2018). · DOI · Paper page
  5. 25. Wu, T. F. et al. Brain-inspired computing exploiting carbon nanotube fets and resistive ram: Hyperdimensional computing case study. In 2018 IEEE International Solid-State Circuits Conference-(ISSCC) , 492-494 (IEEE, 2018).
  6. 28. Song, L., Qian, X., Li, H. & Chen, Y. Pipelayer: A pipelined reram-based accelerator for deep learning. In 2017 IEEE International Symposium on High Performance Computer Architecture (HPCA) , 541-552 (IEEE, 2017). · DOI · Paper page
  7. 29. Sebastian, A. et al. Computational memory-based inference and training of deep neural networks. In 2019 Symposium on VLSI Technology , T168-T169 (IEEE, 2019). · DOI
  8. 31. Mahmoodi, M. et al. Ultra-low power physical unclonable function with nonlinear fixed-resistance crossbar circuits. In 2019 IEEE International Electron Devices Meeting (IEDM) , 30-1 (IEEE, 2019). · doi:10.1109/iedm19573.2019.8993618
  9. 38. Yan, B. et al. Rram-based spiking nonvolatile computing-in-memory processing engine with precisionconfigurable in situ nonlinear activation. In 2019 Symposium on VLSI Technology , T86-T87 (IEEE, 2019). · DOI · Paper page
  10. 40. Gupta, I. et al. Real-time encoding and compression of neuronal spikes by metal-oxide memristors. Nat. communications 7 , 1-9 (2016).
  11. 59. Govoreanu, B. et al. 10 × 10nm 2 hf/hfo x crossbar resistive ram with excellent performance, reliability and low-energy operation. In 2011 International Electron Devices Meeting , 31-6 (IEEE, 2011).
  12. 48. Wu, T. F. et al. 14.3 a 43pj/cycle non-volatile microcontroller with 4.7 m s shutdown/wake-up integrating 2.3-bit/cell resistive ram and resilience techniques. In 2019 IEEE International Solid-State Circuits Conference(ISSCC) , 226-228 (IEEE, 2019).
  13. B.Govoreanu G.Kar Y.Chen V.Paraschiv S.Kubicek A.Fantini I.Radu L.Goux S.Clima R.Degraeve N.Jossart O.Richard T.Vandeweyer K.Seo P.Hendrickx G.Pourtois H.Bender L.Altimime D. J.Wouters J. A.Kittl M.Jurczak in2011 Int. Electron Devices Meeting IEEE Piscataway NJ2011 pp.31–36. · DOI · Paper page
  14. 68. Fantini, A. et al. Intrinsic switching variability in hfo 2 rram. In 2013 5th IEEE International Memory Workshop , 30-33 (IEEE, 2013). · DOI · Paper page
  15. 71. Chen, P.-Y. et al. Mitigating effects of non-ideal synaptic device characteristics for on-chip learning. In 2015 IEEE/ACM International Conference on Computer-Aided Design (ICCAD) , 194-199 (IEEE, 2015).
  16. 73. Fantini, A. et al. Intrinsic program instability in hfo2 rram and consequences on program algorithms. In 2015 IEEE International Electron Devices Meeting (IEDM) , 7.5.1-7.5.4 (2015). · DOI
  17. 74. Lee, H. et al. Evidence and solution of over-reset problem for hfo x based resistive memory with sub-ns switching speed and high endurance. In 2010 International Electron Devices Meeting , 19-7 (IEEE, 2010). · DOI · Paper page
  18. 82. Murmann, B. ADC Performance Survey 1997-2020 (2020).
  19. 87. Hashemi, S., Anthony, N., Tann, H., Bahar, R. I. & Reda, S. Understanding the impact of precision quantization on the accuracy and energy of neural networks. In Design, Automation & Test in Europe Conference & Exhibition (DATE), 2017 , 1474-1479 (IEEE, 2017).
  20. 89. Su, F. et al. A 462gops/j rram-based nonvolatile intelligent processor for energy harvesting ioe system featuring nonvolatile logics and processing-in-memory. In 2017 Symposium on VLSI Technology , T260-T261 (IEEE, 2017).
  21. 90. Bocquet, M. et al. In-memory and error-immune differential rram implementation of binarized deep neural networks. In 2018 IEEE International Electron Devices Meeting (IEDM) , 20-6 (IEEE, 2018). · DOI · Paper page
  22. 92. Liu, Q. et al. 33.2 a fully integrated analog reram based 78.4 tops/w compute-in-memory chip with fully parallel mac computing. In 2020 IEEE International Solid-State Circuits Conference-(ISSCC) , 500-502 (IEEE, 2020).
  23. 93. Mochida, R. et al. A 4m synapses integrated analog reram based 66.5 tops/w neural-network processor with cell current controlled writing and flexible network architecture. In 2018 IEEE Symposium on VLSI Technology , 175-176 (IEEE, 2018).
  24. 94. Xue, C.-X. et al. 24.1 a 1mb multibit reram computing-in-memory macro with 14.6 ns parallel mac computing time for cnn based ai edge processors. In 2019 IEEE International Solid-State Circuits Conference-(ISSCC) , 388-390 (IEEE, 2019).
  25. Q.Liu B.Gao P.Yao D.Wu J.Chen Y.Pang W.Zhang Y.Liao C.-X.Xue W.-H.Chen J.Tang Y.Wang M.-F.Chang H.Qian H.Wu in2020 IEEE Int. Solid-State Circuits Conf.-(ISSCC) IEEE Piscataway NJ2020 pp.500–502.
  26. 99. Karatsuba, A. A. & Ofman, Y. P. Multiplication of many-digital numbers by automatic computers. In Doklady Akademii Nauk , vol. 145, 293-294 (Russian Academy of Sciences, 1962).
  27. 100. Huss-Lederman, S., Jacobson, E. M., Johnson, J. R., Tsao, A. & Turnbull, T. Strassen's algorithm for matrix multiplication: Modeling, analysis, and implementation. In Proceedings of Supercomputing , vol. 96, 9-6 (Citeseer, 1996).
  28. 103. Bavandpour, M., Mahmoodi, M. R. & Strukov, D. acortex: An energy-efficient multi-purpose mixed-signal inference accelerator. accepted, IEEE J. Explor. Solid-State Comput. Devices Circuits (2020).
  29. 104. Bavandpour, M., Sahay, S., Mahmoodi, M. R. & Strukov, D. Efficient mixed-signal neurocomputing via successive integration and rescaling. IEEE Transactions on Very Large Scale Integration (VLSI) Syst. (2019).
  30. 108. Moons, B., Uytterhoeven, R., Dehaene, W. & Verhelst, M. 14.5 envision: A 0.26-to-10tops/w subword-parallel dynamic-voltage-accuracy-frequency-scalable convolutional neural network processor in 28nm fdsoi. In 2017 IEEE International Solid-State Circuits Conference (ISSCC) , 246-247 (IEEE, 2017).
  31. 109. Merritt, R. Startup Accelerates AI at the Sensor (2019).
  32. 110. Shin, D., Lee, J., Lee, J. & Yoo, H.-J. 14.2 dnpu: An 8.1 tops/w reconfigurable cnn-rnn processor for general-purpose deep neural networks. In 2017 IEEE International Solid-State Circuits Conference (ISSCC) , 240-241 (IEEE, 2017).
  33. 115. Google. Edge TPU: Google's purpose-built ASIC designed to run inference at the edge (2019).
  34. 117. Cutress, I. Cambricon, maker of hauwei's kirin npu ip, build a big ai chip and pcie card (2018).
  35. 118. Armasu, L. Move Over GPUs: Startup's Chip Claims to Do Deep Learning Inference Better (2018).
  36. 119. Ueyoshi, K. et al. Quest: A 7.49 tops multi-purpose log-quantized dnn inference engine stacked on 96mb 3d sram using inductive-coupling technology in 40nm cmos. In 2018 IEEE International Solid-State Circuits Conference-(ISSCC) , 216-218 (IEEE, 2018).
  37. 121. Hruska, J. New Movidius Myriad X VPU Packs a Custom Neural Compute Engine (2017).
  38. 124. Chen, Y. et al. Dadiannao: A machine-learning supercomputer. In 2014 47th Annual IEEE/ACM International Symposium on Microarchitecture , 609-622 (IEEE, 2014). · DOI · Paper page
  39. 125. Rockchip. Rockchip Released Its First AI Processor RK3399Pro NPU Performance up to 2.4TOPs (2018).
  40. R.Mochida K.Kouno Y.Hayata M.Nakayama T.Ono H.Suwa R.Yasuhara K.Katayama T.Mikawa Y.Gohou in2018 IEEE Symp. on VLSI Technology IEEE Piscataway NJ2018 pp.175–176. · DOI · Paper page
  41. C.-X.Xue W.-H.Chen J.-S.Liu J.-F.Li W.-Y.Lin W.-E.Lin J.-H.Wang W.-C.Wei T.-W.Chang T.-C.Chang T.-Y.Huang H.-Y.Kao S.-Y.Wei Y.-C.Chiu C.-Y.Lee C.-C.Lo Y.-C.King C.-J.Lin R.-S.Liu C.-C.Hsieh K.-T.Tang M.-F.Chang in2019 IEEE Int. Solid-State Circuits Conf.-(ISSCC) IEEE Piscataway NJ2019 pp.388–390.

Cited by

Items 101–200 of 216

Metadata

Authors
Amirsoleimani, A., Alibart, F., Yon, V., Xu, J., Pazhouhandeh, M. R., Ecoffey, S., Beilliard, Y., Genov, R., & Drouin, D.
Affiliations
  • University of Toronto (CA)
  • Centre National de la Recherche Scientifique (FR)
Corresponding author
  • Amirsoleimani, Amirali
  • Alibart, Fabien
Venue
Advanced Intelligent Systems
Publisher
Wiley
Published
2020-08-23
Volume
2
Issue
11
Pages
2000115
ISSN
2640-4567 (ISSN-L)
OpenAlex
W3081302630
Cited by
216
References
88
Citations by year
2026: 21 · 2025: 52 · 2024: 44 · 2023: 41 · 2022: 38 · 2021: 19 · 2020: 1
FWCI
8.49
Type
journal article
Language
en
Open access
Locations
  • Advanced Intelligent Systems (publishedVersion, cc-by)
  • HAL (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
  • LillOA (Université de Lille (University Of Lille)) (submittedVersion, other-oa)
Funders
  • European Research Council (773228)
  • Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada (HIDATA)
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 › Engineering › Electrical and Electronic Engineering › Low-power high-performance VLSI design
Keywords
in-memory computing, matrix-vector multiplication, resistive random access memory, memristors, monolithic integration, near-memory computing, energy efficiency, circuit design

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.

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

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.