Sensor fusion enabled by photonic neuromorphic computing using phase-change materials

使用功能材料可以“积累”信息，在现场执行记忆和计算任务，这是一个不断增长的领域。在硅芯片上使用光学集成提供了一个独特的机会，可以将硅缩放的好处和光学的波长多路复用结合到一个平台上。在这项提案中，我们将显著扩展我们进行非冯·诺依曼计算(如直接在硬件中进行向量矩阵乘法)的初始工作，以证明实验室规模的此类硬件在国防相关计算任务中的潜力。相变器件在Ge2Sb2Te5(GST)等硫系合金中利用非晶态和晶态之间的可逆循环，已经存在了一段时间，在光域(例如DVD-RAM和蓝光RW盘)和最近的电域(例如英特尔/美光于2015年宣布的3D-XPoint存储器)中提供二进制非易失性存储器。与其他人一样，我们最近也展示了基于相变的设备可以提供许多重要的非von Neumann计算/处理功能，即：1)通过利用多级记忆操作模式，相变设备可以模仿生物突触的基本操作。2)通过以‘能量’积累模式操作，相变设备可以模仿生物神经元的基本操作。3)通过利用累积模式，相变设备还可以提供一种形式的非von Neumann计算，其中记忆和(算术)处理在同一设备中同时进行。在最近的工作(Li等，Optica 7(3)，218-225)中，我们展示了硅和氮化硅波导之间的定量实验差异，并演示了在SOI平台上进行多位光学写入。在这份博士提案中，我们将建立在已证实的基本原则的坚实基础上，开发(设计、制造和测试)基于相变的计算原语，特别是突触和神经元模拟以及(二进制和多级)非易失性存储器，并将这些原语结合到非冯·诺伊曼计算网络(体系结构)中。我们将使用它们来演示与传感器处理和管理相关的国防相关计算任务的概念验证。

The use of functional materials that can "accumulate" information to carry out both memory and computing tasks in-situ is a growing field. Using optical integration onto silicon chips provides a unique opportunity to combine the benefits of silicon scaling and the wavelength multiplexing of optics onto a single platform. In this proposal, we will significantly expand our initial work on carrying out non-von Neumann computations (such as Vector-Matrix Multiplication directly in hardware) to a larger matrix to prove a lab scale demonstration of the potential for such hardware in defence-relevant computing tasks. Phase-change devices, that exploit reversible cycling between amorphous and crystalline states in chalcogenide alloys such as Ge2Sb2Te5 (GST), have been with us for some time now, providing binary non-volatile memories in both the optical domain (e.g. DVD-RAM and Blu-Ray RW disks) and more recently the electrical domain (e.g. in the 3D-XPoint memory announced by Intel/Micron in 2015). Along with others, we have also recently shown that phase-change based devices can provide a number of important non-von Neumann computing/processing functionalities, namely: 1) By exploiting a multi-level memory mode of operation, phase-change devices can mimic the basic operation of a biological synapse.2) By operating in an 'energy' accumulation mode, phase-change devices can mimic the basic operation of a biological neuron.3) By exploiting the accumulation mode, phase-change devices can also provide a form of non-von Neumann computing in which memory and (arithmetic) processing are carried out simultaneously in the same device. In recent work (Li et al, Optica 7 (3), 218-225), we have shown the quantitative experimental differences between Silicon and Silicon Nitride waveguides, and demonstrated multibit optical writing on a SOI platform. In this PhD proposal we will build on this firm foundation of proven basic principles and develop (design, fabricate and test) phase-change-based computing primitives, specifically synapse and neuron mimics and (binary and multi-level) non-volatile memories and combine such primitives into non-von Neumann computing networks (architectures). We will use these to demonstrate proof-of-concept for defence-relevant computational tasks related to sensor processing and management.