← 返回 JSSC 论文列表JSSC 2019第1期Digital Circuits28nm
An Always-On 38 uJ per 86 CIFAR-10 Mixed-Signal Binary CNN Processor With All Me
一款超低功耗混合信号二进制CNN处理器，实现CIFAR-10图像分类任务3.8µJ/分类的能效
28nm CMOS, 0.6V/0.8V供电, 237FPS, 0.9mW功耗, 3.8µJ/分类, 86%准确率
混合信号处理器二进制CNN超低功耗开关电容神经元CIFAR-10分类
▸采用BinaryNet算法将权重和激活值约束为+1/-1，简化乘法运算为XNOR操作
▸权重静止数据并行架构结合输入复用技术，大幅降低内存访问能耗
▸创新的开关电容神经元设计，包含1024位温度计编码CDAC和9位二进制加权偏置模块
Abstract
The trend of pushing inference from cloud to
edge due to concerns of latency, bandwidth, and privacy has
created demand for energy-efficient neural network hardware.
This paper presents a mixed-signal binary convolutional neural
network (CNN) processor for always-on inference applications
that achieves 3.8 µJ/classification at 86% accuracy on the
CIFAR-10 image classification data set. The goal of this paper
is to establish the minimum-energy point for the representative
CIFAR-10 inference task, us