Pub Date : 2024-01-28DOI: 10.1109/ICEIC61013.2024.10457128
Tae Eun Jang, Kyu Hyun Lee, Gi Yeol Kim, Su Yeon Yun, Da-Hyeon Youn, Hyunggu Choi, Jihyang Kim, Soo Youn Kim, Minkyu Song
This paper presents a compute-in-memory (CIM) architecture for MAC operation using 2T1 C dynamic random access memory (DRAM) and a successive-approximation analog-to-digital converter (SAR ADC). The proposed design features CIM analog multiplication and summation architecture consisting of a digital-to-time converter (DTC) and SAR ADC. The DTC converts the input code into clock-based pulse width, and the calculation can be done by passing through pulse into a 2T1C DRAM array in parallel. The proposed structure is implemented using a 28-nm CMOS process, operates four parallel $2-bittimes 4-bit$ multiplication and total summation simultaneously, and a single calculation requires 140ns for 100MHz system clock frequency.
本文介绍了一种用于 MAC 运算的内存计算（CIM）架构，该架构使用 2T1 C 动态随机存取存储器（DRAM）和逐次逼近模数转换器（SAR ADC）。所提出的设计采用 CIM 模拟乘法和求和架构，包括数字到时间转换器（DTC）和 SAR ADC。DTC 将输入代码转换为基于时钟的脉宽，通过将脉冲并行传入 2T1C DRAM 阵列来完成计算。所提出的结构采用 28 纳米 CMOS 工艺实现，可同时进行四个并行的 2 位/次 4 位元乘法和总和运算，在 100MHz 系统时钟频率下，单次运算需要 140ns 的时间。
{"title":"Compute-in-Memory with SAR ADC and 2T1C DRAM for MAC Operations","authors":"Tae Eun Jang, Kyu Hyun Lee, Gi Yeol Kim, Su Yeon Yun, Da-Hyeon Youn, Hyunggu Choi, Jihyang Kim, Soo Youn Kim, Minkyu Song","doi":"10.1109/ICEIC61013.2024.10457128","DOIUrl":"https://doi.org/10.1109/ICEIC61013.2024.10457128","url":null,"abstract":"This paper presents a compute-in-memory (CIM) architecture for MAC operation using 2T1 C dynamic random access memory (DRAM) and a successive-approximation analog-to-digital converter (SAR ADC). The proposed design features CIM analog multiplication and summation architecture consisting of a digital-to-time converter (DTC) and SAR ADC. The DTC converts the input code into clock-based pulse width, and the calculation can be done by passing through pulse into a 2T1C DRAM array in parallel. The proposed structure is implemented using a 28-nm CMOS process, operates four parallel $2-bittimes 4-bit$ multiplication and total summation simultaneously, and a single calculation requires 140ns for 100MHz system clock frequency.","PeriodicalId":518726,"journal":{"name":"2024 International Conference on Electronics, Information, and Communication (ICEIC)","volume":"102 7-8","pages":"1-3"},"PeriodicalIF":0.0,"publicationDate":"2024-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140530387","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-28DOI: 10.1109/ICEIC61013.2024.10457222
Jin Shin, Hyun Kim
Recently, in deep learning research, the importance of domain generalization (DG) for unseen domains has been emphasized. Most of the baseline methodologies for this focus on generating adversarial representations or separating content and style information from intermediate features for learning. However, these approaches inevitably increase the time complexity for both training and inference. In this study, we propose an approach to improve DG performance without excessive bottleneck points. We suggest an auxiliary network structure that places a mapping layer for feature alignment after the stem layer, a generative model based on an adaptive instance normalization that can adjust mean and standard deviation. This structure consistently adjusts the output feature maps of the stem layer to follow a Gaussian distribution regardless of the domain used as the input image. Moreover, both training and inference are possible without iterative routines, making their complexity nearly identical to training without the DG strategies. Experimental results show that our model outperforms the existing DG baseline with the highest performance in image classification tasks by an average accuracy of 0.71% higher on the PACS benchmarking dataset.
{"title":"A Simplified Feature Alignment Strategy for Image Classification Across Domains","authors":"Jin Shin, Hyun Kim","doi":"10.1109/ICEIC61013.2024.10457222","DOIUrl":"https://doi.org/10.1109/ICEIC61013.2024.10457222","url":null,"abstract":"Recently, in deep learning research, the importance of domain generalization (DG) for unseen domains has been emphasized. Most of the baseline methodologies for this focus on generating adversarial representations or separating content and style information from intermediate features for learning. However, these approaches inevitably increase the time complexity for both training and inference. In this study, we propose an approach to improve DG performance without excessive bottleneck points. We suggest an auxiliary network structure that places a mapping layer for feature alignment after the stem layer, a generative model based on an adaptive instance normalization that can adjust mean and standard deviation. This structure consistently adjusts the output feature maps of the stem layer to follow a Gaussian distribution regardless of the domain used as the input image. Moreover, both training and inference are possible without iterative routines, making their complexity nearly identical to training without the DG strategies. Experimental results show that our model outperforms the existing DG baseline with the highest performance in image classification tasks by an average accuracy of 0.71% higher on the PACS benchmarking dataset.","PeriodicalId":518726,"journal":{"name":"2024 International Conference on Electronics, Information, and Communication (ICEIC)","volume":"39 1","pages":"1-3"},"PeriodicalIF":0.0,"publicationDate":"2024-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140530402","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-28DOI: 10.1109/ICEIC61013.2024.10457228
Jie-Wei Gim, Lei-Jun Siau, Jen-Hahn Low, E. Lim, P. Chee
Rapid advances in sensing technologies have led to the rapid development of wearable electronics for biomedical applications. Among these, the triboelectric nanogenerator (TENG) is a promising technology for harvesting energy from the environment. TENG can be used as a self-powered wearable sensor to generate electricity by converting mechanical energy into electrical energy. In this context, a pressure-sensitive single-electrode triboelectric nanogenerator (SE-TENG) is developed. A $4times 4$ array of deformable protruded hemispherical structures is constructed on the elastomer to enhance the contact surface area. This design also allows for the energy harvesting from different magnitudes of hand tapping forces. In our experiments, the output voltage is proportional to the hand tapping forces. Three different diameters of protruded hemispherical structures were compared, and the larger diameter shows a larger output voltage. This proposed SE-TENG has been demonstrated for ball bouncing game, which is useful for rehabilitation applications.
{"title":"Development of A Pressure-Sensitive Triboelectric Self-Powered Sensor Using Protruded Hemispherical Array Structures","authors":"Jie-Wei Gim, Lei-Jun Siau, Jen-Hahn Low, E. Lim, P. Chee","doi":"10.1109/ICEIC61013.2024.10457228","DOIUrl":"https://doi.org/10.1109/ICEIC61013.2024.10457228","url":null,"abstract":"Rapid advances in sensing technologies have led to the rapid development of wearable electronics for biomedical applications. Among these, the triboelectric nanogenerator (TENG) is a promising technology for harvesting energy from the environment. TENG can be used as a self-powered wearable sensor to generate electricity by converting mechanical energy into electrical energy. In this context, a pressure-sensitive single-electrode triboelectric nanogenerator (SE-TENG) is developed. A $4times 4$ array of deformable protruded hemispherical structures is constructed on the elastomer to enhance the contact surface area. This design also allows for the energy harvesting from different magnitudes of hand tapping forces. In our experiments, the output voltage is proportional to the hand tapping forces. Three different diameters of protruded hemispherical structures were compared, and the larger diameter shows a larger output voltage. This proposed SE-TENG has been demonstrated for ball bouncing game, which is useful for rehabilitation applications.","PeriodicalId":518726,"journal":{"name":"2024 International Conference on Electronics, Information, and Communication (ICEIC)","volume":"76 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2024-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140530222","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：