基于带符号位的浮点数运算的多位宽3D RRAM设计  

A Multi-Bit 3D RRAM-Based Signed Floating-Point Number Operations

作　　者：王兴华[1] 王天[1] 王乾 李潇然[1] WANG Xinghua;WANG Tian;WANG Qian;LI Xiaoran(School of Integrated Circuits and Electronics,Beijing Institute of Technology,Beijing 100081,China)

机构地区：[1]北京理工大学集成电路与电子学院,北京100081

出　　处：《北京理工大学学报》2022年第12期1299-1304,共6页Transactions of Beijing Institute of Technology

基　　金：国家自然科学基金资助项目(61801027,62101038)。

摘　　要：本文介绍了卷积神经网络(convolutional neutral network,CNN)系统中具有多位存储的三维阻变式存储器(threedimensional resistive random-access memory,3D RRAM)的带符号位的浮点数运算.与其他类型存储器相比,3D RRAM可以在存储器内部进行运算,且具有更高的读取速率和更低的能耗,为解决冯诺依曼架构的瓶颈问题提供新方案.单个RRAM单元的最大和最小电阻分别达到10 GΩ和10 MΩ,可在多级电阻状态下稳定,以存储多比特位宽的数据.测试结果表明,带符号位的浮点数的卷积运算系统的精度可以达到99.8%,测试中3D RRAM模型的峰值读取速度为0.529 MHz.In this paper,a signed floating-point number operation with multi-bit storage three-dimensional resistive random-access memory(3D RRAM)was presented for complex convolution neutral network(CNN)systems.Comparing with other types of memory,3D RRAM can not only perform calculations inside the memory,but also possess a higher reading rate and a lower energy consumption,providing a new solution to the bottleneck problem of the Von Neumann architecture.A single RRAM cell can reach a maximum and minimum resistance of 10 GΩand 10 MΩ,which can be stabilized in multi-level resistance states to store high-bit-width data.The test results show that,the accuracy of the signed floating-point number convolution operation system can reach up to 99.8%,the measured peak reading speed of the 3D RRAM model is 0.529 MHz.

关 键 词：3D RRAM 存算一体 带符号位的浮点数卷积运算 多级电阻 峰值读取速度 

分 类 号：TN47[电子电信—微电子学与固体电子学]