Optogenetics-inspired photoelectric memories based on flexible nanogap electrodes

该项目的目的是开发一种融合了光子、电子和离子效应的新型神经形态系统，为记忆计算和人工视觉记忆应用带来新的前景。这将通过开发使用共面纳米间隙电极和多功能溶液处理材料的光电存储器来实现，该材料采用与大面积柔性衬底兼容的低成本工艺制造。神经形态工程准备通过开发能够真实模拟生物神经网络的电子设备来革命信息技术。一个关键组件是“人工突触”，它需要高度可扩展和高能效，同时支持类似生物突触的丰富动态反应。这类平台的一个新兴应用是神经形态视觉，其中光传感器不仅通过将光转化为电信号来模拟人类视觉的时空性质，而且通过以极其高效的方式捕获仅有用的信息并将其发送到处理单元。这对于支持大量应用的实时模式识别任务尤其相关，这些应用从自主移动到护理点诊断，利用传感器在速度、更大动态范围和更低计算成本方面的进步。光遗传学领域已经率先使用了光敏蛋白质，这种蛋白质可以在光照下随意激活，并刺激神经元放电。受这项技术的启发，我将制造可以由光刺激控制的人造突触，与电刺激相比，光刺激可以在空间上受到限制，从而显著减少串扰和噪声，同时它们能够实现更高的灵敏度和信号传播速度。我将使用一种简单的纳米制造方法来设计与实际突触相同维度的原型设备，即大长宽比纳米间隙分离电极，纳米间隙的范围在15 nm，类似于突触裂隙的大小。为了解决目前在大面积柔性衬底上可靠制造纳米级结构的挑战，将使用粘合光刻技术制造模拟神经元网络的互连纳米间隙电极。最后，我将使用光敏多金属氧酸盐和卤化物钙钛矿来制备类突触的金属/半导体/金属结。这些材料的成膜特性及其与金属结构的界面将被定制以展示神经形态功能，例如(A)联想学习，(B)并行寻址设备以模拟生物网络的动态平衡，以及(C)光刺激在阵列中的空间集成以实现依赖于每个像素上的光强度/波长的选择性存储。与现有的记忆技术相比，我的方法呈现出几个优点，这些技术基于纵横杆结构和仅基于电刺激。首先，共面纳米间隙电极由于其低维度，在实现低功耗和快速切换速度方面大有希望，就像其他类型的设备(射频二极管、光电探测器)所展示的那样，而它们的平面几何结构促进了光控制操作，使得能够模拟调节电阻状态和消除阵列配置中的潜伏电流。其次，上述溶液可加工材料具有许多有吸引力的光电性能、化学可调性和可制造性优点，使它们适合达到设定的性能目标。成功实施这一奖学金将代表着神经形态设备制造的范式转变，支持英国的电子和制造业，同时它将使我成为人工智能硬件纳米级光电子领域的领导者。

The aim of this project is to develop a new form of neuromorphic systems that merge photonic, electronic and ionic effects, bringing new prospects for in-memory computing and artificial visual memory applications. This will be achieved upon developing photoelectric memories that employ coplanar nanogap electrodes and multi-functional solution-processed materials, fabricated with low-cost processes compatible with large-area flexible substrates.Neuromorphic engineering is poised to revolutionise information technologies by developing electronic devices that can realistically emulate biological neural networks. A key component is the "artificial synapse" that needs to be highly scalable and power efficient, whilst supporting rich dynamical responses akin to biological synapses. An emerging application of such platforms is in neuromorphic vision, where light sensors mimic the spatio-temporal nature of human vision not only by turning light into electrical signals but also by capturing and sending the useful-only information to the processing unit in an extremely efficient manner. This is particularly relevant for real-time pattern recognition tasks that support a plethora of applications, from autonomous locomotion to point-of-care diagnostics, leveraging the sensors advances in speed, greater dynamic range and decreased computational cost. The field of optogenetics has pioneered the use of light-sensitive proteins that can be activated at will upon illumination and stimulate the neurons to fire. Inspired by this technology, I will fabricate artificial synapses that can be controlled by optical stimuli, which, in contrast to electrical ones, can be spatially confined reducing thus significantly the crosstalk and noise, while they enable higher sensitivity and signal propagation speed. I will employ a simple nanofabrication method to design prototype devices of the same dimensionality as the actual synapse, namely large aspect ratio nanogap-separated electrodes, the nanogap being in the range of 15 nm, similar to the size of the synaptic cleft. Interconnected nanogap electrodes emulating neuronal networks will be fabricated using adhesion lithography technique to address the current challenge of reliable manufacturing of nanoscale structures on large area flexible substrates. Finally, I will employ photosensitive polyoxometalate and halide perovskite to fabricate synaptic-like metal/semiconductor/metal junctions. The film forming properties of these materials and their interfaces with the metal structures will be tailored to demonstrate neuromorphic functionalities, such as (a) associative learning, (b) parallel addressing of devices to emulate homeostasis of biological networks and (c) spatial integration of the optical stimulus in the array to enable selective storage depending on the light intensity/wavelength on each pixel.My approach presents several advantages over the existing memristive technologies, which are based on crossbar architectures and solely electrical stimulus. First, coplanar nanogap electrodes, owing to their low dimensionality, hold great promise for achieving low power consumption and fast switching speeds, as already demonstrated with other types of devices (radiofrequency diodes, photodetectors), while their planar geometry facilitates a light-controlled operation, enabling both analogue tuning of resistance states and elimination of sneak currents in the array configuration. Second, the aforementioned solution-processable materials present many attractive optoelectronic properties, chemical tunability and manufacturability merits that render them suitable to reach the set performance goals.Successful implementation of this fellowship will represent a paradigm shift in the fabrication of neuromorphic devices, supporting the UK-based electronics and manufacturing industry, while it will establish me as a leader in the field of nanoscale optoelectronics for AI hardware.