Memristor-based storage system with convolutional autoencoder-based image compression network

• DOI: 10.1038/s41467-024-45312-0
• Publication Date: 2024-02-07T05:02:13Z
• Authors: Yulin Feng; Yizhou Zhang; Zheng Zhou; Peng Huang; Lifeng Liu; Xiaoyan Liu; Jinfeng Kang
• Keywords: Autoencoder

The exponential growth of various complex images is putting tremendous pressure on storage systems. Here, we propose a memristor-based system with an integrated near-storage in-memory computing-based convolutional autoencoder compression network to boost the energy efficiency and speed image compression/retrieval improve density. We adopt 4-bit memristor arrays experimentally demonstrate functions system. step-by-step quantization aware training scheme equivalent transformation for transpose convolution performance. exhibits high (>33 dB) peak signal-to-noise ratio in decompression ImageNet Kodak24 datasets. Benchmark comparison results show that could reduce latency consumption by over 20×/5.6× 180×/91×, respectively, compared server-grade central processing unit-based/the graphics unit-based system, density more than 3 times.