標題: | Mitigating Effects of Non-ideal Synaptic Device Characteristics for On-chip Learning |
作者: | Chen, Pai-Yu Lin, Binbin Wang, I-Ting Hou, Tuo-Hung Ye, Jieping Vrudhula, Sarma Seo, Jae-sun Cao, Yu Yu, Shimeng 交大名義發表 National Chiao Tung University |
關鍵字: | machine learning;neuromorphic computing;cross-point array;resistive memory;synaptic device |
公開日期: | 1-一月-2015 |
摘要: | The cross-point array architecture with resistive synaptic devices has been proposed for on-chip implementation of weighted sum and weight update in the training process of learning algorithms. However, the non-ideal properties of the synaptic devices available today, such as the nonlinearity in weight update, limited ON/OFF range and device variations, can potentially hamper the learning accuracy. This paper focuses on the impact of these realistic properties on the learning accuracy and proposes the mitigation strategies. Unsupervised sparse coding is selected as a case study algorithm. With the calibration of the realistic synaptic behavior from the measured experimental data, our study shows that the recognition accuracy of MNIST handwriting digits degrades from similar to 97 % to similar to 65 %. To mitigate this accuracy loss, the proposed strategies include 1) the smart programming schemes for achieving linear weight update; 2) a dummy column to eliminate the off-state current; 3) the use of multiple cells for each weight element to alleviate the impact of device variations. With the improved synaptic behavior by these strategies, the accuracy increases back to similar to 95 %, enabling the reliable integration of realistic synaptic devices in the neuromorphic systems. |
URI: | http://hdl.handle.net/11536/129821 |
ISBN: | 978-1-4673-8388-2 |
ISSN: | 1933-7760 |
期刊: | 2015 IEEE/ACM INTERNATIONAL CONFERENCE ON COMPUTER-AIDED DESIGN (ICCAD) |
起始頁: | 194 |
結束頁: | 199 |
顯示於類別: | 會議論文 |