A Near-Sensor Processing Accelerator for Approximate Local Binary Pattern Networks
A Near-Sensor Processing Accelerator for Approximate Local Binary Pattern Networks
复制标题
DOI：
10.1109/tetc.2023.3285493
复制
发表时间：
2022-10
影响因子：
5.9
通讯作者：
Shaahin Angizi;Mehrdad Morsali;Sepehr Tabrizchi;A. Roohi
中科院分区：
文献类型：
--
作者：
Shaahin Angizi;Mehrdad Morsali;Sepehr Tabrizchi;A. Roohi
在这项工作中，提出了基于高速和节能的近传感器局部二进制二进制模式加速器体系结构（NS-LBP），以执行新型的本地二进制模式深神经网络，这是受我们最近的LBP网络的启发。设计一个近似，面向硬件的且乘以乘以的（MAC）的网络，名为AP-LBP，用于有效的特征提取，进一步降低了计算复杂性。并行内存LBP算法在缓存中处理传感器附近的图像，这显着将数据传输的功率降低到芯片外处理器上。与基线CNN和LBP - 网络模型相比最近的基于LBP的网络。
In this work, a high-speed and energy-efficient comparator-based Near-Sensor Local Binary Pattern accelerator architecture (NS-LBP) is proposed to execute a novel local binary pattern deep neural network. First, inspired by recent LBP networks, we design an approximate, hardware-oriented, and multiply-accumulate (MAC)-free network named Ap-LBP for efficient feature extraction, further reducing the computation complexity. Then, we develop NS-LBP as a processing-in-SRAM unit and a parallel in-memory LBP algorithm to process images near the sensor in a cache, remarkably reducing the power consumption of data transmission to an off-chip processor. Our circuit-to-application co-simulation results on MNIST and SVHN datasets demonstrate minor accuracy degradation compared to baseline CNN and LBP-network models, while NS-LBP achieves 1.25 GHz and an energy-efficiency of 37.4 TOPS/W. NS-LBP reduces energy consumption by 2.2× and execution time by a factor of 4× compared to the best recent LBP-based networks.