Application-specific integrated circuit (ASIC) implementations for Deep
Neural Networks (DNNs) have been adopted in many systems because of their
higher classification speed. However, although they may be characterized by
better accuracy, larger DNNs require significant energy and area, thereby
limiting their wide adoption. The energy consumption of DNNs is driven by both
memory accesses and computation. Binarized Neural Networks (BNNs), as a
trade-off between accuracy and energy consumption, can achieve great energy
reduction, and have good accuracy for large DNNs due to its regularization
effect. However, BNNs show poor accuracy when a smaller DNN configuration is
adopted. In this paper, we propose a new DNN model, LightNN, which replaces the
multiplications to one shift or a constrained number of shifts and adds. For a
fixed DNN configuration, LightNNs have better accuracy at a slight energy
increase than BNNs, yet are more energy efficient with only slightly less
accuracy than conventional DNNs. Therefore, LightNNs provide more options for
hardware designers to make trade-offs between accuracy and energy. Moreover,
for large DNN configurations, LightNNs have a regularization effect, making
them better in accuracy than conventional DNNs. These conclusions are verified
by experiment using the MNIST and CIFAR-10 datasets for different DNN
configurations.

本文提出一种新的 DNN 模型 LightNN，它将乘法替换成一次位移或约束数量的位移和加法，为硬件设计者提供了更多在精度和能耗之间取得平衡的选项，并且验证实验表明，LightNN 在大型 DNN 配置下具有优于传统 DNN 和 BNN 的精度和能效优势。

LightNN: 填补传统深度神经网络与二值化网络之间的差距

LightNN: Filling the Gap between Conventional Deep Neural Networks and  Binarized Networks

We discuss a new class of Micro Pattern Gas Detectors, the Gas Pixel Detector
(GPD), in which a complete integration between the gas amplification structure
and the read-out electronics has been reached. An Application-Specific
Integrated Circuit (ASIC) built in deep sub-micron technology has been
developed to realize a monolithic device that is, at the same time, the
pixelized charge collecting electrode and the amplifying, shaping and charge
measuring front-end electronics. The CMOS chip has the top metal layer
patterned in a matrix of 80 micron pitch hexagonal pixels, each of them
directly connected to the underneath electronics chain which has been realized
in the remaining five layers of the 0.35 micron VLSI technology. Results from
tests of a first prototype of such detector with 2k pixels and a full scale
version with 22k pixels are presented. The application of this device for
Astronomical X-Ray Polarimetry is discussed. The experimental detector response
to polarized and unpolarized X-ray radiation is shown. Results from a full
MonteCarlo simulation for two astronomical sources, the Crab Nebula and the
Hercules X1, are also reported.

提出了一种新的微型图案气体探测器 Gas Pixel Detector (GPD)，该探测器使用 CMOS 芯片作为像素化电荷收集电极和前端电子学，可用于天文 X 射线偏振测量，已完成了包括 2k 和 22k 像素在内的完整模型的实验证明。