Oxide film devices for AI processing of temporal signals

可以运行机器学习算法来分类和预测来自诸如语言处理、环境、工程或医疗监测等应用领域中的传感器的依赖于时间的输入信号。然而，与最先进的神经网络硬件相关的高能耗阻碍了移动、紧凑型传感器的发展，这些传感器可以独立操作，也可以离线操作。这是一个设计使用全新的物理方法来处理神经网络的设备的机会，以解决这个问题。该项目的目标是实验开发能够处理时间相关电信号的薄膜设备，作为神经网络的一部分。该项目包括氧化物材料薄膜的制备；应用不同的表征技术来研究材料特性；设计新型电子设备；按照行业标准基准测试其性能。这项研究项目处于人工智能和设备物理的交叉点，涉及物理、化学、计算机科学领域的学术研究人员之间的合作，并与EPSRC的EP/T027479/1和EP/S032843/1相关联。

It is possible to run machine learning algorithms to classify and predict time-dependent input signals from sensors in application areas such as language processing, environmental, engineering, or medical monitoring. However, high energy consumption associated with the state-of-the-art hardware for neural networks is hindering development of mobile, compact sensors that can be operated stand-alone and offline. This is an opportunity to design devices that use fundamentally new physical approaches to processing neural networks in order to tackle this issue.The goal of this project is experimental development of thin film devices capable of processing time-dependent electrical signals as part of a neural network. The project includes preparation of thin films of oxide materials; application of different characterisation techniques to study the material properties; designing novel electronic devices; testing their performance with respect to the industry standard benchmarks. This research project is at the intersection of artificial intelligence and device physics, and involves collaborative work between academic researchers in Physics, Chemistry, Computer Science and is linked to the EPSRC grants EP/T027479/1 and EP/S032843/1.