A 12-Nm 0.62-1.61 Mw Ultra-Low Power Digital CIM-based Deep-Learning System for End-to-End Always-on Vision.

En-Jui Chang,Cheng-Xin Xue,Chetan Deshpande,Gajanan Jedhe,Jenwei Liang,Chih-Chung Cheng,Hung-Wei Lin,Chia-Da Lee,Sushil Kumar,Kim Soon Jway,Zijie Guo,Ritesh Garg,Allen-Cl Lu,Chien-Hung Lin, Meng-Han Hsieh,Tsung-Yao Lin,Chih-Cheng Chen

2023 IEEE Symposium on VLSI Technology and Circuits (VLSI Technology and Circuits)（2023）

MediaTek Inc.

Cited 0|Views15

Abstract

This work proposes an ultra-low power DCIM-based DL system (DCIM-DLS) for end-to-end AoV with the power range from 0.62 to 1.61 mW (INT8, 2-15 fps). Compared to the prior art [3], the power consumption of DCIM-DLS can be reduced by 70.9% based on the following techniques: 1) an area and energy efficient DCIM that reduces compute RC loading by using pushed-rule 2p8T SRAM bitcell with folded kernels selector, 2) a DCIM-friendly dataflow strategy with dual accumulators that minimizes the DCIM power of weight update and avoids redundant data movement for power saving, and 3) a reconfigurable DCIM control scheme that supports mixed-precision to further reduce power consumption.

Translated text

Key words

2-15 fps,compute RC loading,DCIM power,DCIM-DLS,DCIM-friendly dataflow strategy,end-to-end AoV,mW ultra-low power digital CIM-based,power 0.62 mW to 1.61 mW,power consumption,pushed-rule 2p8T SRAM bitcell,reconfigurable DCIM control scheme,ultra-low power DCIM-based

求助PDF

上传PDF

Bibtex

AI Read Science

AI Summary

AI Summary is the key point extracted automatically understanding the full text of the paper, including the background, methods, results, conclusions, icons and other key content, so that you can get the outline of the paper at a glance.

Example

Background

Key content

Introduction

Methods

Results

Related work

Fund

Key content

Pretraining has recently greatly promoted the development of natural language processing (NLP)
We show that M6 outperforms the baselines in multimodal downstream tasks, and the large M6 with 10 parameters can reach a better performance
We propose a method called M6 that is able to process information of multiple modalities and perform both single-modal and cross-modal understanding and generation
The model is scaled to large model with 10 billion parameters with sophisticated deployment, and the 10 -parameter M6-large is the largest pretrained model in Chinese
Experimental results show that our proposed M6 outperforms the baseline in a number of downstream tasks concerning both single modality and multiple modalities We will continue the pretraining of extremely large models by increasing data to explore the limit of its performance

Upload PDF to Generate Summary

Must-Reading Tree

Example

Generate MRT to find the research sequence of this paper

Data Disclaimer

The page data are from open Internet sources, cooperative publishers and automatic analysis results through AI technology. We do not make any commitments and guarantees for the validity, accuracy, correctness, reliability, completeness and timeliness of the page data. If you have any questions, please contact us by email: report@aminer.cn

Chat Paper

【要点】：本研究提出了一种12纳米工艺的0.62至1.61毫瓦超低功耗数字内容智能管理（DCIM）基深度学习系统，用于端到端的始终开启视觉（AoV），通过多种技术降低了70.9%的能耗，实现了深度学习在边缘计算设备上的高效运行。

【方法】：研究采用了一种面积和能耗效率优化的DCIM，结合了2p8T SRAM位单元的pushed-rule技术和折叠内核选择器，以及一种DCIM友好的数据流策略和可重构的DCIM控制方案。

【实验】：实验使用了INT8精度，帧率在2至15 fps范围内，通过比较现有技术，验证了所提出的DCIM基深度学习系统的能耗优势，具体数据集名称未在摘要中提及。