Information: Join today and make your research activities more affordable! Technical workshop participation fees and annual registration fees are available at member rates.
Notice: [Important] Announcement of Changes to Registration Fee Payment and Manuscript Upload Procedures for IEICE Technical Meetings
IEICE Technical Committee Submission System
Conference Paper's Information		 Online Proceedings
[Sign in]
Tech. Rep. Archives
 Go Top Page Go Previous   [Japanese] / [English] 

Paper Abstract and Keywords
Presentation 2025-12-03 10:25
Fault-Injected Weight-Adjusting Training for Reliable Memristor-Based Neural Networks.
Md. Sihabul Islam, Taisho Sasada, Michiko Inoue (NAIST) VLD2025-54 ICD2025-74 DC2025-87 RECONF2025-105
Abstract (in Japanese) (See Japanese page) 
(in English) Neural networks (NNs) have emerged as essential computational models for learning and inference across a wide range of tasks in artificial intelligence and neuromorphic computing systems (NCS). Memristor crossbars (MCs) offer promising hardware for accelerating NNs due to their efficient computation and low power consumption. However, device-level defects, particularly fabrication-induced stuck-at faults (SAFs), severely degrade inference accuracy and undermine the reliability of such systems. Existing SAFs mitigating methods primarily rely on fault-tolerant mapping or retraining. Mapping methods, though effective, introduce substantial hardware overhead, while retraining is computationally expensive and time-consuming. To address these limitations, we propose a framework that integrates fault-injected weight-adjusting (FIWA) training with a weighted bipartite matching and weight-adjusting (WBM+WA) mapping. By injecting SAFs into the training and adjusting weights, our method enhances the fault resiliency of the model. The proposed WBM+WA mapping, after training, assigns weights to the crossbars so that the faulty weights are placed in the faulty memristors by minimizing the sensitivity of weights to faults and considering the weights adjustment. The effectiveness of this method is evaluated across two different datasets and NN architectures. The results of our experiments demonstrate that our method achieves high inference accuracy, although memristor crossbars have SAFs. Thus, the proposed method improves the reliability of memristor-based NCS.
Keyword (in Japanese) (See Japanese page) 
(in English) neuromorphic computing / memristor crossbar / stuck-at-faults / fault-injected weight-adjusting training / mapping / / /  
Reference Info. IEICE Tech. Rep., vol. 125, no. 262, DC2025-87, pp. 198-203, Dec. 2025.
Paper # DC2025-87 
Date of Issue 2025-11-24 (VLD, ICD, DC, RECONF) 
ISSN Online edition: ISSN 2432-6380
Copyright
and
reproduction		 All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034)
Download PDF VLD2025-54 ICD2025-74 DC2025-87 RECONF2025-105

Conference Information
Committee VLD DC RECONF ICD IPSJ-SLDM  
Conference Date 2025-12-01 - 2025-12-03 
Place (in Japanese) (See Japanese page) 
Place (in English) Toyama International Conference Center 
Topics (in Japanese) (See Japanese page) 
Topics (in English) Design Gaia 2025 -New Field of VLSI Design- 
Paper Information
Registration To DC 
Conference Code 2025-12-VLD-DC-RECONF-ICD-SLDM 
Language English 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Fault-Injected Weight-Adjusting Training for Reliable Memristor-Based Neural Networks. 
Sub Title (in English)  
Keyword(1) neuromorphic computing  
Keyword(2) memristor crossbar  
Keyword(3) stuck-at-faults  
Keyword(4) fault-injected weight-adjusting training  
Keyword(5) mapping  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Md. Sihabul Islam  
1st Author's Affiliation Nara Institute of Science and Technology (NAIST)
2nd Author's Name Taisho Sasada  
2nd Author's Affiliation Nara Institute of Science and Technology (NAIST)
3rd Author's Name Michiko Inoue  
3rd Author's Affiliation Nara Institute of Science and Technology (NAIST)
4th Author's Name  
4th Author's Affiliation ()
5th Author's Name  
5th Author's Affiliation ()
6th Author's Name  
6th Author's Affiliation ()
7th Author's Name  
7th Author's Affiliation ()
8th Author's Name  
8th Author's Affiliation ()
9th Author's Name  
9th Author's Affiliation ()
10th Author's Name  
10th Author's Affiliation ()
11th Author's Name  
11th Author's Affiliation ()
12th Author's Name  
12th Author's Affiliation ()
13th Author's Name  
13th Author's Affiliation ()
14th Author's Name  
14th Author's Affiliation ()
15th Author's Name  
15th Author's Affiliation ()
16th Author's Name  
16th Author's Affiliation ()
17th Author's Name  
17th Author's Affiliation ()
18th Author's Name  
18th Author's Affiliation ()
19th Author's Name  
19th Author's Affiliation ()
20th Author's Name  
20th Author's Affiliation ()
21st Author's Name  
21st Author's Affiliation ()
22nd Author's Name  
22nd Author's Affiliation ()
23rd Author's Name  
23rd Author's Affiliation ()
24th Author's Name  
24th Author's Affiliation ()
25th Author's Name  
25th Author's Affiliation ()
26th Author's Name / /
26th Author's Affiliation ()
()
27th Author's Name / /
27th Author's Affiliation ()
()
28th Author's Name / /
28th Author's Affiliation ()
()
29th Author's Name / /
29th Author's Affiliation ()
()
30th Author's Name / /
30th Author's Affiliation ()
()
31st Author's Name / /
31st Author's Affiliation ()
()
32nd Author's Name / /
32nd Author's Affiliation ()
()
33rd Author's Name / /
33rd Author's Affiliation ()
()
34th Author's Name / /
34th Author's Affiliation ()
()
35th Author's Name / /
35th Author's Affiliation ()
()
36th Author's Name / /
36th Author's Affiliation ()
()
Speaker Author-1 
Date Time 2025-12-03 10:25:00 
Presentation Time 20 minutes 
Registration for DC 
Paper # VLD2025-54, ICD2025-74, DC2025-87, RECONF2025-105 
Volume (vol) vol.125 
Number (no) no.260(VLD), no.261(ICD), no.262(DC), no.263(RECONF) 
Page pp.198-203 
#Pages 6 
Date of Issue 2025-11-24 (VLD, ICD, DC, RECONF) 

[Return to Top Page]

[Return to IEICE Web Page]
The Institute of Electronics, Information and Communication Engineers (IEICE), Japan