Memristive Devices for Time Domain Compute-in-Memory
Analog compute schemes and compute-in-memory (CIM) have emerged in an effort to reduce the increasing power hunger of convolutional neural networks (CNNs), which exceeds the constraints of edge devices. Memristive device types are a relatively new offering with interesting opportunities for unexplor...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2022-01-01
|
| Series: | IEEE Journal on Exploratory Solid-State Computational Devices and Circuits |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/9930136/ |
| _version_ | 1828097430272344064 |
|---|---|
| author | Florian Freye Jie Lou Christopher Bengel Stephan Menzel Stefan Wiefels Tobias Gemmeke |
| author_facet | Florian Freye Jie Lou Christopher Bengel Stephan Menzel Stefan Wiefels Tobias Gemmeke |
| author_sort | Florian Freye |
| collection | DOAJ |
| description | Analog compute schemes and compute-in-memory (CIM) have emerged in an effort to reduce the increasing power hunger of convolutional neural networks (CNNs), which exceeds the constraints of edge devices. Memristive device types are a relatively new offering with interesting opportunities for unexplored circuit concepts. In this work, the use of memristive devices in cascaded time-domain CIM (TDCIM) is introduced with the primary goal of reducing the size of fully unrolled architectures. The different effects influencing the determinism in memristive devices are outlined together with reliability concerns. Architectures for binary as well as multibit multiply and accumulate (MAC) cells are presented and evaluated. As more involved circuits offer more accurate compute result, a tradeoff between design effort and accuracy comes into the picture. To further evaluate this tradeoff, the impact of variations on overall compute accuracy is discussed. The presented cells reach an energy/OP of 0.23 fJ at a size of <inline-formula> <tex-math notation="LaTeX">$1.2~{\mu{ }}\text{m}^{2}$ </tex-math></inline-formula> for binary and 6.04 fJ at <inline-formula> <tex-math notation="LaTeX">$3.2~\mu \text{m}^{2}$ </tex-math></inline-formula> for <inline-formula> <tex-math notation="LaTeX">$4\times 4$ </tex-math></inline-formula> bit MAC operations. |
| first_indexed | 2024-04-11T07:46:15Z |
| format | Article |
| id | doaj.art-7cc4c0186508466ea6ea37243f3d9cb9 |
| institution | Directory Open Access Journal |
| issn | 2329-9231 |
| language | English |
| last_indexed | 2024-04-11T07:46:15Z |
| publishDate | 2022-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Journal on Exploratory Solid-State Computational Devices and Circuits |
| spelling | doaj.art-7cc4c0186508466ea6ea37243f3d9cb92022-12-22T04:36:18ZengIEEEIEEE Journal on Exploratory Solid-State Computational Devices and Circuits2329-92312022-01-018211912710.1109/JXCDC.2022.32170989930136Memristive Devices for Time Domain Compute-in-MemoryFlorian Freye0https://orcid.org/0000-0003-3025-8910Jie Lou1https://orcid.org/0000-0003-0380-8585Christopher Bengel2https://orcid.org/0000-0002-2892-9837Stephan Menzel3https://orcid.org/0000-0002-4258-2673Stefan Wiefels4https://orcid.org/0000-0003-2820-9677Tobias Gemmeke5https://orcid.org/0000-0003-1583-3411Chair of Integrated Digital Systems and Circuit Design, RWTH Aachen University, Aachen, GermanyChair of Integrated Digital Systems and Circuit Design, RWTH Aachen University, Aachen, GermanyInstitut für Werkstoffe der Elektrotechnik 2, RWTH Aachen University, Aachen, GermanyForschungszentrum Jülich GmbH, Peter Gruenberg Institut (PGI-7), Jülich, GermanyForschungszentrum Jülich GmbH, Peter Gruenberg Institut (PGI-7), Jülich, GermanyChair of Integrated Digital Systems and Circuit Design, RWTH Aachen University, Aachen, GermanyAnalog compute schemes and compute-in-memory (CIM) have emerged in an effort to reduce the increasing power hunger of convolutional neural networks (CNNs), which exceeds the constraints of edge devices. Memristive device types are a relatively new offering with interesting opportunities for unexplored circuit concepts. In this work, the use of memristive devices in cascaded time-domain CIM (TDCIM) is introduced with the primary goal of reducing the size of fully unrolled architectures. The different effects influencing the determinism in memristive devices are outlined together with reliability concerns. Architectures for binary as well as multibit multiply and accumulate (MAC) cells are presented and evaluated. As more involved circuits offer more accurate compute result, a tradeoff between design effort and accuracy comes into the picture. To further evaluate this tradeoff, the impact of variations on overall compute accuracy is discussed. The presented cells reach an energy/OP of 0.23 fJ at a size of <inline-formula> <tex-math notation="LaTeX">$1.2~{\mu{ }}\text{m}^{2}$ </tex-math></inline-formula> for binary and 6.04 fJ at <inline-formula> <tex-math notation="LaTeX">$3.2~\mu \text{m}^{2}$ </tex-math></inline-formula> for <inline-formula> <tex-math notation="LaTeX">$4\times 4$ </tex-math></inline-formula> bit MAC operations.https://ieeexplore.ieee.org/document/9930136/Compute-in-memory (CIM)convolutional neural networks (CNNs)memristive devicestime-domain (TD) computingtime-domain CIM (TDCIM) |
| spellingShingle | Florian Freye Jie Lou Christopher Bengel Stephan Menzel Stefan Wiefels Tobias Gemmeke Memristive Devices for Time Domain Compute-in-Memory IEEE Journal on Exploratory Solid-State Computational Devices and Circuits Compute-in-memory (CIM) convolutional neural networks (CNNs) memristive devices time-domain (TD) computing time-domain CIM (TDCIM) |
| title | Memristive Devices for Time Domain Compute-in-Memory |
| title_full | Memristive Devices for Time Domain Compute-in-Memory |
| title_fullStr | Memristive Devices for Time Domain Compute-in-Memory |
| title_full_unstemmed | Memristive Devices for Time Domain Compute-in-Memory |
| title_short | Memristive Devices for Time Domain Compute-in-Memory |
| title_sort | memristive devices for time domain compute in memory |
| topic | Compute-in-memory (CIM) convolutional neural networks (CNNs) memristive devices time-domain (TD) computing time-domain CIM (TDCIM) |
| url | https://ieeexplore.ieee.org/document/9930136/ |
| work_keys_str_mv | AT florianfreye memristivedevicesfortimedomaincomputeinmemory AT jielou memristivedevicesfortimedomaincomputeinmemory AT christopherbengel memristivedevicesfortimedomaincomputeinmemory AT stephanmenzel memristivedevicesfortimedomaincomputeinmemory AT stefanwiefels memristivedevicesfortimedomaincomputeinmemory AT tobiasgemmeke memristivedevicesfortimedomaincomputeinmemory |