Probabilistic computing and biological applications

概率计算和生物应用

• 批准号: 0726969
• 负责人: David Anderson
• 金额: $ 75万
• 依托单位: Georgia Tech Research Corporation
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2007
• 资助国家: 美国
• 起止时间: 2007-09-01 至 2012-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0726969&HistoricalAwards=false
• 关键词: Probabilistic; computing; biological; applications

Biological systems hold the key to ultra-powerful computing using just microwatts of power; this research explores emulating this behavior in electrical systems. A novel, biologically inspired computing paradigm is being developed which could fundamentally change scientific and multimedia computing. This new computing approach simultaneously addresses a looming problem in semiconductor systems and many nano-electronics systems; that is, as feature sizes in computer chips are scaled down further, ideal behavior cannot always be guaranteed. In these systems, results of individual operations are described only statistically or probabilistically. A principal point of this research is to embrace probabilistic computing elements rather than to devise ways to make them ideal or deterministic. Recent research suggests that biological and other natural systems are probabilistic in nature. Thus, probabilistic technology provides a novel method to simulate biological, chemical, and neurobiological systems to reach previously unattainable simulation speeds and complexity. Additionally, many multi-media signal processing systems can take advantage of this computing approach to achieve tremendous gains in efficiency at the cost of imperceptible degradation in quality. Probabilistic CMOS (PCMOS) allows a computing circuit to operate probabilistically and, as a result, achieve extreme power savings. A PCMOS circuit is a digital circuit where the supply voltage is lowered to sub-threshold levels; the output of the circuit is correct with some probability p 1. The probabilistic nature of the computation may be artificially imposed or it may be a inevitable result of extreme semiconductor scaling. Furthermore, p can be precisely controlled using extant analog floating gate technology. This research uses PCMOS to perform computations and simulations that either require or can tolerate probabilistic behavior, specifically simulation of biological processes which can be extended to Monte Carlo simulations for any dynamical system. These methods allow orders-of-magnitude speed-up and substantially reduced power consumption. The research involves hardware, algorithmic, and theoretical aspects of probabilistic computing.

生物系统掌握着只用微瓦功率就能实现超强计算的关键；本研究探索了在电气系统中模拟这种行为。一种新颖的、受生物学启发的计算范式正在被开发，它可能从根本上改变科学和多媒体计算。这种新的计算方法同时解决了半导体系统和许多纳米电子系统中迫在眉睫的问题；也就是说，随着计算机芯片的特征尺寸进一步缩小，理想的行为不能总是得到保证。在这些系统中，单个操作的结果仅以统计或概率方式描述。这项研究的一个要点是采用概率计算元素，而不是设计使它们理想或确定的方法。最近的研究表明，生物和其他自然系统在本质上是概率性的。因此，概率技术提供了一种新的方法来模拟生物、化学和神经生物学系统，以达到以前无法实现的模拟速度和复杂性。此外，许多多媒体信号处理系统可以利用这种计算方法以难以察觉的质量下降为代价获得巨大的效率收益。概率CMOS （PCMOS）允许计算电路以概率方式运行，从而实现极低的功耗。PCMOS电路是一种数字电路，其中电源电压降低到亚阈值水平；电路的输出以p 1的概率是正确的。计算的概率性质可能是人为强加的，也可能是极端半导体缩放的必然结果。此外，可以使用现有的模拟浮栅技术精确控制p。本研究使用PCMOS来执行需要或可以容忍概率行为的计算和模拟，特别是生物过程的模拟，可以扩展到任何动态系统的蒙特卡罗模拟。这些方法可以实现数量级的加速，并大大降低功耗。研究涉及硬件，算法和理论方面的概率计算。