Analog Signal Processing Using Stochastic Magnets
Analog Signal Processing Using Stochastic Magnets
复制标题
DOI：
10.1109/access.2021.3075839
复制
发表时间：
2021-01-01
期刊：
影响因子：
3.9
通讯作者：
Ghosh, Avik W.
中科院分区：
文献类型：
--
作者：
Ganguly, Samiran;Camsari, Kerem Y.;Ghosh, Avik W.
我们提出了一种基于低能垒磁铁的模拟随机神经元(ASN)的紧凑型硬件单元，并演示了它作为神经形态硬件的构建块的用途。由这些单元组成的网络特别适合于时间推断和模式识别。我们展示了这些ASN的实例应用，包括多层感知器、卷积神经元和存储计算机，这些计算机展示了诸如时间序列学习、处理和预测任务，证明这些单元可以用来构建高效、可扩展和自适应的基于神经网络的信号处理器。我们还提供了与基于数字CMOS的电路的说明性比较，这些电路实现了与使用所示单元构建的网络类似的功能，展示了元件数量可能减少两个数量级，并伴随着能效的提高。这种信号处理器的高效非冯-诺依曼硬件实现可以为物联网、工业控制、生物和光传感器、自动驾驶汽车和无人驾驶飞行器等各种新兴系统中基于硬件的认知集成开辟一条道路。
We present a low energy-barrier magnet based compact hardware unit for analog stochastic neurons (ASNs) and demonstrate its use as a building-block for neuromorphic hardware. Networks assembled from these units are particularly suited for temporal inferencing and pattern recognition. We demonstrate example applications of these ASNs including multi-layer perceptrons, convolutional neurons, and reservoir computers showing tasks such as temporal sequence learning, processing, and prediction tasks which prove that these units can be used to build efficient, scalable, and adaptive neural network based signal-processors. We also provide an illustrative comparison with digital CMOS based circuits that implement similar functionality with networks built using the presented units, demonstrating a possible two orders of magnitude reduction in component-count and concomitant increase in energy efficiency. Efficient non von-Neumann hardware implementation of such signal-processors can open up a pathway for integration of hardware based cognition in a wide variety of emerging systems such as IoT, industrial controls, bio- and photo-sensors, self-driving automotives, and unmanned aerial vehicles.