期刊
IEEE ELECTRON DEVICE LETTERS
卷 43, 期 1, 页码 142-145出版社
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/LED.2021.3125966
关键词
Silicon; Programming; Tin; Logic gates; Tunneling; Computer architecture; Performance evaluation; Synaptic device; AND flash memory; fin-type flash device; low programming voltage; binary neural network; XNOR operation
资金
- National Research and Development Program through the National Research Foundation of Korea (NRF) - Ministry of Science and ICT [2021M3F3A2A02037889]
- BK21 FOURProgram of the Education and Research Program for Future ICT Pioneers, Seoul National University
An oxide fin-based AND flash memory synaptic device is proposed and fabricated for a hardware-based binary neural network. The device features a fin-like curved channel structure, providing local electric field enhancement for improved programming efficiency. The fin-based AND flash cell exhibits high on/off current ratio, low off current, and low programming voltage, resulting in a sufficient dynamic range of synaptic weights for BNNs. Additionally, a hardware-based BNN using a novel two-cell synaptic device arrangement is proposed, enabling parallel XNOR operation and bit-counting. Experimental results show only slight degradation in classification accuracy compared to the baseline accuracy, making it suitable for off-chip event-driven computation.
An oxide fin-based AND flash memory synaptic device is proposed and fabricated using a spacer patterning technology for a hardware-based binary neural network (BNN). A fin-like curved channel structure provides local electric field enhancement, which improves programming efficiency compared to planar-type flash synaptic devices. The fin-based AND flash cell exhibits a high on/off current ratio (>10(5)) with sub-pA off current, and a low programming voltage (< 9 V) is used to achieve a sufficient dynamic range of synaptic weights (>10(3)) for BNNs. Furthermore, a hardware-based BNN using novel two cell-based synaptic devices arranged in AND array architecture is proposed to implement parallel XNOR operation and bit-counting. Proposed BNN using the synapse model with measured dynamic range and retention property shows only < 0.5 % degradation of classification accuracy compared to the baseline accuracy, which is suitable to perform off-chip event-driven computation using parallel read-out operations.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据