A fast training method for memristor crossbar based multi-layer neural networks
作者: Raqibul HasanTarek M. TahaChris Yakopcic
作者单位: 1University of Dayton
刊名: Analog Integrated Circuits and Signal Processing, 2017, Vol.93 (3), pp.443-454
来源数据库: Springer Nature Journal
DOI: 10.1007/s10470-017-1051-y
关键词: Neural networksMemristor crossbarsTraining algorithmOn-chip training
英文摘要: Memristor crossbar arrays carry out multiply–add operations in parallel in the analog domain which is the dominant operation in a neural network application. On-chip training of memristor neural network systems have the significant advantage of being able to get around device variability and faults. This paper presents a novel technique for on-chip training of multi-layer neural networks implemented using a single crossbar per layer and two memristors per synapse. Using two memristors per synapse provides double the synaptic weight precision when compared to a design that uses only one memristor per synapse. Proposed system utilizes a novel variant of the back-propagation (BP) algorithm to reduce both circuit area and training time. During training, all the memristors in a crossbar are...
原始语种摘要: Memristor crossbar arrays carry out multiply–add operations in parallel in the analog domain which is the dominant operation in a neural network application. On-chip training of memristor neural network systems have the significant advantage of being able to get around device variability and faults. This paper presents a novel technique for on-chip training of multi-layer neural networks implemented using a single crossbar per layer and two memristors per synapse. Using two memristors per synapse provides double the synaptic weight precision when compared to a design that uses only one memristor per synapse. Proposed system utilizes a novel variant of the back-propagation (BP) algorithm to reduce both circuit area and training time. During training, all the memristors in a crossbar are...
全文获取路径: Springer Nature  (合作)
分享到:
来源刊物:

×
关键词翻译
关键词翻译
  • training 培养
  • crossbar 横撑
  • neural 神经系统的
  • layer 
  • separable 可分的
  • algorithm 算法
  • trained 受过训练的
  • precision 精度
  • steps 无扶手步梯
  • method 方法