An array of sensors generating a collection of correlated signals can benefit from integration with a “smart” system for autonomous inferencing. Mimicking their biological counterparts, smart sensor systems shall possess the capabilities of sensing, memory, and neuromorphic computation. However, state‐of‐the‐art biomimetic systems either do not employ a full set of devices to cover the complete range of capabilities or incorporate devices that are capable of all but appropriate only for a limited range of sensing applications. Presently proposed is a smart sensor architecture that combines an array of sensing elements with an overlapping array of computing and memory elements, thus emulating an innervated peripheral sensing system (IPSS) capable of local and autonomous neuromorphic in‐sensor data pre‐processing. Compatibility of the proposed architecture with functionally distinct elements for sensing, memory, and computing removes the restrictive demand for a single element simultaneously capable of all, thus making this architecture more generally applicable to a wider range of sensors and usage scenarios. An artificial synapse as a computing element is implemented using dual‐gate (DG) thin‐film transistors (TFTs) and the low‐leakage current of transistors based on metal‐oxide semiconductors allows the deployment of capacitors as memory elements. The outputs of the IPSS are passed on to an adjacent artificial neural network (ANN) for near‐sensor inferencing. Monolithic integration of the IPSS and the ANN is made possible by the deployment of the same memory and computing elements in their construction. A smart tactile sensing system based on the proposed architecture is constructed and characterized. The functionality of the system is demonstrated by its application to the classification of a set of tactile images of 3‐dimensionally printed alphabet stamps.

中文翻译：

---

使用双栅极金属氧化物薄膜晶体管技术实现基于传感器内计算的智能传感架构


生成相关信号集合的传感器阵列可以从与“智能”系统的集成中受益，以实现自主推理。智能传感器系统模仿它们的生物对应物，应具有传感、记忆和神经形态计算的能力。然而，最先进的仿生系统要么没有采用全套设备来涵盖全部功能，要么采用能够实现所有功能但仅适用于有限范围传感应用的设备。目前提出的是一种智能传感器架构，它将一系列传感元件与重叠的计算和存储元件阵列相结合，从而模拟一个神经支配的外围传感系统 （IPSS），该系统能够进行局部和自主的神经形态传感器内数据预处理。所提出的架构与用于传感、存储和计算的不同功能元件的兼容性消除了对单个元件同时能够同时满足所有元件的限制性需求，从而使该架构更普遍地适用于更广泛的传感器和使用场景。使用双栅极 （DG） 薄膜晶体管 （TFT） 实现作为计算元件的人工突触，基于金属氧化物半导体的晶体管的低泄漏电流允许部署电容器作为存储元件。IPSS 的输出被传递给相邻的人工神经网络 （ANN） 进行近传感器推理。IPSS 和 ANN 的单体集成是通过在它们的构造中部署相同的内存和计算元素来实现的。基于所提出的架构构建了智能触觉传感系统并进行了表征。 该系统的功能通过应用于 3D 印刷字母邮票的一组触觉图像的分类来证明。