To reduce the energy-consuming and time latency incurred by Von Neumann architecture, this brief developed a complete computing-in-memory (CIM) convolutional macro based on ReRAM array for the convolutional layers of a LeNet-like convolutional neural network (CNN). We binarized the input layer and the first convolutional layer to get higher accuracy. The proposed ReRAM-CIM convolutional macro is suitable as an IP core for any binarized neural networks’ convolutional layers. This brief customized a bit-cell consisting of 2T2R ReRAM cells, regarded ${9 \times 8}$ bit-cells as one unit to achieve high hardware compute accuracy, great read/compute speed, and low power consuming. The ReRAM-CIM convolutional macro achieved 50 ns product-sum computing time for one complete convolutional operation in a convolutional layer in the customized CNN, with an accuracy of 96.96% on MNIST database and a peak energy efficiency of 58.82 TOPS/W.

中文翻译：

---

基于 ReRAM 的内存计算卷积宏，具有用于 AIoT 芯片 IP 应用的定制 2T2R 位单元

为了减少冯诺依曼架构带来的能耗和时间延迟，本简报为类 LeNet 卷积神经网络 (CNN) 的卷积层开发了一个基于 ReRAM 阵列的完整内存计算 (CIM) 卷积宏。我们将输入层和第一个卷积层二值化以获得更高的精度。提出的 ReRAM-CIM 卷积宏适合作为任何二值化神经网络卷积层的 IP 核。本简报定制了一个由 2T2R ReRAM 单元组成的位单元，将 ${9 \times 8}$ 位单元视为一个单元，以实现高硬件计算精度、出色的读取/计算速度和低功耗。ReRAM-CIM 卷积宏在定制的 CNN 的卷积层中实现了 50 ns 乘积和计算时间，用于一个完整的卷积操作，