Development of New Materials for Optically Programmable and Re-Programmable Printable Memristors for Neuromorphic Computing

印刷电子学(即使用添加的低温印刷工艺制造电子设备和电路的技术)承诺以新的形状因数(例如，在柔性或共形衬底上，使用生物兼容元件，大面积地)低成本、按需制造电子电路。印刷电子系统的制造需要比制造传统集成电路所需的硬件解决方案简单得多、成本更低的硬件解决方案，物联网中一些最有希望的商业机会是在智能传感器、智能标签和标签以及其他“边缘计算”场景中，在这些场景中，智能处理嵌入到低成本设备本身，而不是转移到云中。这些系统的交付需要低功率、低成本和低组件数量的模拟计算，并结合深度学习算法，这导致了对模拟神经网络计算的兴趣的重新燃起，有时被称为“神经形态”计算方法。实现这些机器学习算法所需的硬件的关键组件需要新的材料来制造坚固、可编程的打印设备。在这个PHD项目中，将开发新的材料和配方，以提供其中材料电阻能够在制造后被编程的阵列和电路。这将首先通过对材料电阻进行光学编码来实现，以使忆阻器设备中的写入和读取过程解耦，从而提供稳定、坚固和可重现的可编程阵列，从而提供电路(例如，用于神经网络的突触权重)。这种方法将被扩展到开发用于光学可再编程阵列的材料，其中将使用不同波长的照明来重复编码和擦除器件阵列的电阻。最后，该研究计划将研究从2D平面打印阵列转移到3D折叠或打印结构的方法，这些结构具有更高的设备密度和更高互连的电路体系结构，如在人类皮质中看到的那样。

Printed electronics (i.e. the technology of fabricating electronic devices and circuits using additive, low-temperature printing processes) promises low-cost, on-demand electronic circuit fabrication, in new form-factors (e.g. on flexible or conformal substrates, using biocompatible components, over large areas). The manufacture of printed electronic systems requires much simpler and cheaper hardware solutions than those needed for making traditional integrated circuits and some of the most promising commercial opportunities in the Internet of Things are in intelligent sensors, smart tags and labels, and other 'edge computing' scenarios, where intelligent processing is embedded in the low-cost device itself not transferred to the cloud. Delivery of these systems requires the low-power, low-cost, and low component count of analogue computing, combined with deep learning algorithms and this has led to a resurgence of interest in analogue neural network computing, sometimes termed a "neuromorphic" approach to computation. Critical components of the hardware needed to implement these machine learning algorithms require new materials for robust, programmable, printed devices. In this PhD project new materials and formulations will be developed to deliver arrays and circuits in which the materials resistance is capable of being programmed after fabrication.This will be initially achieved by optical encoding of materials resistance to decouple the write and read processes in memristor devices to deliver stable, robust and reproducible programmable arrays and hence circuits (e.g. for the synaptic weights of neural networks). This approach will be extended to develop materials for optically re-programmable arrays in which illumination at different wavelengths will be used to repeatedly encode and erase the resistance of device arrays. Finally the research programme will investigate approaches to move from 2D planar printed arrays to 3D folded or printed structures with a greater device density and a more highly interconnected circuit architecture such as that seen in the human cortex.