中文 | English

J. Hefei Univ. Tech. (Nat. Sci.)

合肥工业大学学报(自然科学版)

JOURNAL OF HEFEI UNIVERSITY OF TECHNOLOGY (NATURAL SCIENCE)

Competent Authorities: Ministry of Education of the People's Republic of China

Sponsor by: Hefei University of Technology

ISSN 1000-1646, CN 34-1001/N

Gated recurrent unit circuit based on memristor

Download Full Text

基于忆阻器的门控循环单元电路

Journal Information

J. Hefei Univ. Tech. (Nat. Sci.), Feb.2024, Vol.47 No.2: 189-194

DOI: 10.3969/j.issn.1003-5060.2024.02.007

Authors

HAN Tingting, ZHANG Zhang, CHEN Sikai

(School of Microelectronics, Hefei University of Technology, Hefei 230601, China)

Abstract and Keywords

Abstract: Making accurate predictions and identifying long-term patterns are both possible through the examination of time series data. Gated recurrent unit(GRU), which can process time series signals in comparison to artificial neural network(ANN), is frequently employed in natural language processing, speech recognition, machine translation, etc. The intricacy of the parameters and model, however, has caused a hardware implementation bottleneck for the GRU model. In this study, a memristor-based GRU with full circuit functionality and fewer input/output parameters is built. According to the simulation findings, the average error of the circuit is 0.0075, effectively realizing the purpose of GRU network. The GRU circuit design is applied to the sequence prediction model, which can forecast the trend of stock price fluctuations, and the anticipated R2 score reaches 0.9234. The construction of GRU hardware circuits based on memristor offers potential for use in artificial intelligence and machine learning.

Keywords: memristor; recurrent neural network(RNN); gated recurrent unit(GRU); sequence prediction

作者信息

韩婷婷，张章，陈思锴

（合肥工业大学微电子学院，安徽合肥230601）

摘要和关键词

摘要: 时间序列数据分析可用于识别长期趋势并进行正确的预测,与人工神经网络(artificial neural network, ANN)相比,门控循环单元(gated recurrent unit, GRU)可以处理时间序列信号,在自然语言处理、语音识别、机器翻译等方面有着广泛的应用。然而,由于参数和模型的复杂性,GRU模型在硬件实现中遇到了瓶颈。文章构建一个基于忆阻器的GRU硬件电路,具有完整的GRU功能,而且输入/输出参数更少。仿真结果表明,电路的平均误差为0.0075,能够有效地实现GRU网络的功能。将设计的GRU电路应用在搭建的序列预测模型中,可以预测股票价格变化趋势,且其预测的R2分数达到0.9234。因此基于忆阻器的GRU硬件电路的设计在机器学习和人工智能方面具有一定的应用潜力。

关键词: 忆阻器；循环神经网络（RNN）；门控循环单元（GRU）；序列预测

Funding

国家自然科学基金区域创新发展联合基金资助项目（U19A2053）

Previous Next

Back to Contents