Toward Energy-Efficient, Bio-Inspired Circuits and Systems for Error-Resilient and Biomedical Applications

我们中的许多人每隔几年就会更换一次电脑、笔记本电脑或手机。由于电子设备的不断小型化，技术进步，使得更大数量的设备可以封装到单个芯片中，而成本保持相对稳定。然而，这一趋势已经放缓，预计将在不到十年的时间内结束。晶体管，电路中的基本功能单元，现在的尺寸只有几纳米，也就是几十亿分之一米。在如此小的规模下，很难均匀地制造所有晶体管并使它们可靠地工作。目前确保可靠运行的方法是施加比通常所需的更大的功率，因此电子设备仍然消耗大量能量。另一方面，许多计算机应用，例如多媒体、语音识别和网络搜索，并不总是需要完全准确的结果，并且由于诸如人类感知限制的许多因素，“足够好”的结果通常是足够的。此类应用程序被认为是不精确容忍或错误恢复的。该研究计划的一个目标是通过开发新的和创新的计算结构，采用近似，随机和大脑启发的神经形态计算技术，解决纳米级电子产品的能效和容错问题。这些新技术允许计算系统在质量和能量之间进行权衡。

Many of us get our computers, laptops or phones replaced every few years. Technology advances due to the continuous miniaturization of electronic devices, such that a larger number of devices can be packed into a single chip, while the cost has been kept relatively stable. This trend has slowed down, however, and is predicted to end in less than a decade. A transistor, the basic functional unit in a circuit, is now sized in just a few nanometers, that is, a few billionth of a meter. At such a small scale, it is difficult to fabricate all transistors uniformly and make them operate reliably. The current method to ensure a reliable operation is to apply a larger power than it is often necessary, so electronics still consume a lot of energy. On the other hand, many computer applications, such as multimedia, voice recognition and web search, do not always require a fully accurate result and a “good-enough” result is often sufficient due to many factors such as human perceptual limitations. This class of applications is considered imprecision-tolerant or error-resilient. One objective of this research program is to address the energy-efficiency and error-resilience issues in nanometer-scale electronics by developing new and innovative computational structures that employ approximate, stochastic and the brain-inspired neuromorphic computing techniques. These new techniques allow computing systems to trade off quality for energy.