Approximate Computing: An Energy-Efficient Computing Technique for Error Resilient Applications
Approximate Computing: An Energy-Efficient Computing Technique for Error Resilient Applications
复制标题
近似计算：一种用于容错应用的节能计算技术
DOI：
10.1109/isvlsi.2015.130
复制
发表时间：
2015
期刊：
影响因子：
--
通讯作者：
A. Raghunathan
中科院分区：
文献类型：
--
作者：
K. Roy;A. Raghunathan
近年来，近似计算已成为众所周知的计算技术。它依赖于许多系统和应用程序在计算结果中的质量或最佳性丧失的能力。主要思想是利用系统的固有误差弹性或误差耐性，以实现能源效率，换句话说，通过能源消耗进行交易准确性。在大多数情况下，这种折衷也与绩效改进有关，例如更快的操作，减少区域等。幸运的是，如今的大多数重型工作负载都具有内在的应用程序弹性。在大多数多媒体应用中，最终输出是通过人类感官来解释的，这并不是完美。这一事实避免了产生精确输出的需求。最近的研究工作已定量确定了许多应用中的高度固有弹性。例如，我们对12个识别，视觉和多媒体应用的基准套件的分析表明，平均而言，在可以忍受至少一定程度的近似值的计算中，有83％的运行时间用于计算。因此，有可能在广泛的背景下利用固有的弹性。让我们考虑数字信号处理（DSP）。在多媒体应用程序中，DSP块实现了图像，声音和视频处理算法，其中最终输出是图像或声音或视频的人类感官。在解释图像，声音或视频时，人类的感知能力有限。这使系统可以灵活地产生质量输出。例如，数值精确度的放松提供了进行不精确或近似计算的自由度。近似信号处理的另一个例子是增量的利用。显然，权衡可能允许近似计算来处理超出我们可以通过传统计算的任务。设计抽象的不同级别可以实现近似计算方法。在本文中，我们将简要介绍为实现错误弹性应用程序实施近似硬件的不同方法。
Approximate computing has become a well-known computing technique in recent years. It relies on the ability of many systems and applications to self-heal or to tolerate some loss of quality or optimality in the computed result. The main idea is to exploit the inherent error resiliency or error tolerance of the system to achieve energy efficiency, or in other words, trading accuracy with energy consumption. Such trade-off, in most cases, is also associated with performance improvements like faster operations, area reduction etc. Fortunately, most of the heavy workloads nowadays exhibit intrinsic application resilience. In most multimedia applications, the final output is interpreted by human senses, which are not perfect. This fact averts the need to produce exact outputs. Recent research efforts have quantitatively ascertained the high degree of inherent resilience in many applications. For example, our analysis of a benchmark suite of 12 recognition, vision and multimedia applications shows that on average, 83% of the runtime is spent in computations that can tolerate at least some degree of approximation. Therefore, there is a potential to leverage inherent resilience in a broad context. Let us consider digital signal processing (DSP). In multi-media applications, DSP blocks implement image, sound and video processing algorithms, where the final output is either an image or sound or a video for human senses. When interpreting an image or a sound or a video, human beings have limited perceptual capacities. This allows the system to be flexible in producing quality outputs. As an example, the relaxation on numerical precision provides some freedom to carry out imprecise or approximate computation. Another example of approximate signal processing is the utilization of incremental refinement. It is evident that the tradeoffs may allow approximate computing to handle tasks beyond what we can do with traditional computing.There are different levels of design abstraction where approximate computing methods can be implemented. In this paper we will briefly describe different approaches that we have developed to implement approximate hardware for error resilient applications.