Shengtao Tu , Jinyu Li , Yanyun Ren , Qin Jiang , Shisheng Xiong
{"title":"一种新型的忆阻器编程电路","authors":"Shengtao Tu , Jinyu Li , Yanyun Ren , Qin Jiang , Shisheng Xiong","doi":"10.1016/j.mee.2023.112072","DOIUrl":null,"url":null,"abstract":"<div><p><span><span><span>Memristor has attracted a lot of interest due to its high processing speed, </span>low power consumption and high integration ability, which is critical for electronic systems and memory-centric computing. However, the memristor </span>programming circuit and strategy are still inflexible and complex, since the signal generator/collector and stimulate pulse must be carefully matched and designed based on memristor intrinsic characteristics without reconfigurable. Here, a simple and effective circuit only consists a parallel reference-resistor-and-NMOS is designed to program memristor with a >99% memristance precision. And the amplitude and width of stimulate pulse are fixed to ±4 V and 5 ms, respectively. In order to cope with the device variation, such as ±10% tolerance of transition voltage, an optimized programming strategy was proposed and demonstrated great robustness. Additionally, a set of reference resistors and NMOSs have been added to facilitate multi-level memristance operation without requiring any changes to the circuit structure. This program circuit was also employed to program memristor crossbar remains 99% precision. In the end, a memristor-based </span>convolutional neural network which controlled by our optimized programming circuit was used for image recognition, and 89.36% accuracy can be achieved even under 15.8% memristance tolerance. This novel circuit demonstrates a simple and flexible strategy in memristor programming, providing a new way to control memristor crossbar for practical application.</p></div>","PeriodicalId":18557,"journal":{"name":"Microelectronic Engineering","volume":null,"pages":null},"PeriodicalIF":2.6000,"publicationDate":"2023-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"A novel programming circuit for memristors\",\"authors\":\"Shengtao Tu , Jinyu Li , Yanyun Ren , Qin Jiang , Shisheng Xiong\",\"doi\":\"10.1016/j.mee.2023.112072\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><p><span><span><span>Memristor has attracted a lot of interest due to its high processing speed, </span>low power consumption and high integration ability, which is critical for electronic systems and memory-centric computing. However, the memristor </span>programming circuit and strategy are still inflexible and complex, since the signal generator/collector and stimulate pulse must be carefully matched and designed based on memristor intrinsic characteristics without reconfigurable. Here, a simple and effective circuit only consists a parallel reference-resistor-and-NMOS is designed to program memristor with a >99% memristance precision. And the amplitude and width of stimulate pulse are fixed to ±4 V and 5 ms, respectively. In order to cope with the device variation, such as ±10% tolerance of transition voltage, an optimized programming strategy was proposed and demonstrated great robustness. Additionally, a set of reference resistors and NMOSs have been added to facilitate multi-level memristance operation without requiring any changes to the circuit structure. This program circuit was also employed to program memristor crossbar remains 99% precision. In the end, a memristor-based </span>convolutional neural network which controlled by our optimized programming circuit was used for image recognition, and 89.36% accuracy can be achieved even under 15.8% memristance tolerance. This novel circuit demonstrates a simple and flexible strategy in memristor programming, providing a new way to control memristor crossbar for practical application.</p></div>\",\"PeriodicalId\":18557,\"journal\":{\"name\":\"Microelectronic Engineering\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":2.6000,\"publicationDate\":\"2023-08-15\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Microelectronic Engineering\",\"FirstCategoryId\":\"5\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S0167931723001375\",\"RegionNum\":4,\"RegionCategory\":\"工程技术\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"ENGINEERING, ELECTRICAL & ELECTRONIC\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Microelectronic Engineering","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0167931723001375","RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ENGINEERING, ELECTRICAL & ELECTRONIC","Score":null,"Total":0}
Memristor has attracted a lot of interest due to its high processing speed, low power consumption and high integration ability, which is critical for electronic systems and memory-centric computing. However, the memristor programming circuit and strategy are still inflexible and complex, since the signal generator/collector and stimulate pulse must be carefully matched and designed based on memristor intrinsic characteristics without reconfigurable. Here, a simple and effective circuit only consists a parallel reference-resistor-and-NMOS is designed to program memristor with a >99% memristance precision. And the amplitude and width of stimulate pulse are fixed to ±4 V and 5 ms, respectively. In order to cope with the device variation, such as ±10% tolerance of transition voltage, an optimized programming strategy was proposed and demonstrated great robustness. Additionally, a set of reference resistors and NMOSs have been added to facilitate multi-level memristance operation without requiring any changes to the circuit structure. This program circuit was also employed to program memristor crossbar remains 99% precision. In the end, a memristor-based convolutional neural network which controlled by our optimized programming circuit was used for image recognition, and 89.36% accuracy can be achieved even under 15.8% memristance tolerance. This novel circuit demonstrates a simple and flexible strategy in memristor programming, providing a new way to control memristor crossbar for practical application.
期刊介绍:
Microelectronic Engineering is the premier nanoprocessing, and nanotechnology journal focusing on fabrication of electronic, photonic, bioelectronic, electromechanic and fluidic devices and systems, and their applications in the broad areas of electronics, photonics, energy, life sciences, and environment. It covers also the expanding interdisciplinary field of "more than Moore" and "beyond Moore" integrated nanoelectronics / photonics and micro-/nano-/bio-systems. Through its unique mixture of peer-reviewed articles, reviews, accelerated publications, short and Technical notes, and the latest research news on key developments, Microelectronic Engineering provides comprehensive coverage of this exciting, interdisciplinary and dynamic new field for researchers in academia and professionals in industry.