Signal Processing to the Rescue of Moore's Law

该项目寻求提供有效和低成本的解决方案，以解决目前制约着2500亿美元全球半导体行业增长的电力可靠性问题。这是制造集成电路(IC)或微芯片的行业，这些集成电路或微芯片位于消费电子、通信和计算产品的核心，如手机、笔记本电脑、个人数字助理、调制解调器、路由器和许多其他产品。摩尔？S定律是自70年代以来推动该行业创新和增长的自我实现的预言，该定律预测，每18个月，单位面积上的晶体管数量将翻一番。摩尔-S定律正受到两个趋势的威胁：纳米非理想化的出现使设计可靠的芯片变得困难，以及新的应用和标准对功能的要求增加，导致高功耗。这就是威胁摩尔-S定律的电力可靠性问题。在这个项目中，研究人员利用信号处理的能力理论和技术，通过将纳米IC视为有噪声的通信网络，来解决上述宽带通信系统IC的功率可靠性问题。研究分为三个主要活动。首先，将开发用于数字信号处理(DSP)特定块的参数化统计行为模型，例如数据路径、计算、存储器和片上通信块。接下来，通过使用行为模型并使用统计信号处理理论中的概念，例如自适应滤波、检测和估计、差错控制以及分布式和稳健系统理论，将研究信号处理启发的技术，以设计稳健和节能的信号处理和通信系统的实现。最后，一个展示一种或多种技术潜力的原型测试芯片将在项目的最后一年实施和测试。

This project seeks to provide effective and low-cost solutions to the power-reliability problem inhibiting the growth of the $250B global semiconductor industry today. This is the industry that manufactures integrated circuits (ICs) or micro-chips that lie at the heart of consumer electronics, communications and computing products such as cell-phones, lap-tops, personal digital assistants, modems, routers, and many others. Moore?s Law, a self-fulfilling prophesy driving innovation and growth in the industry since the 70s, predicts that the number of transistors per unit area will double every 18 months. Moore?s Law is under threat from two trends: emergence of nanometer non-idealities that make it hard to design reliable chips, and increased functionality demanded by new applications and standards resulting in high power consumption. This is then the power-reliability problem which threatens Moore?s Law. In this project, the investigators are employing the power of signal processing theory and techniques to solve the power-reliability problem described above in the context of ICs for broadband communication systems by viewing nanometer ICs as noisy communication networks. The research is organized into three major activities. First, parameterized statistical behavioral models for digital signal processing (DSP)-specific blocks such as data-path, computation, memory, and on-chip communication blocks will be developed. Next, signal processing-inspired techniques will be investigated for designing robust and energy-efficient implementations of signal processing and communication systems by employing the behavioral models and using concepts from statistical signal processing theory such as adaptive filtering, detection and estimation, error-control and distributed and robust systems theory. Finally, a prototype test chip demonstrating the potential of one or more of the techniques will be implemented and tested in the final year of the project.