适用于权重复用神经网络的脉动阵列单元及脉动阵列结构
By using pulsating array units and structural design, the multiplication and addition operations of two sets of input data for the convolutional neural network were realized within the same cycle, which solved the problem of slow computation speed of convolutional neural networks and improved computational efficiency.
CN116702851BActive Publication Date: 2026-07-17NANJING INST OF INTELLIGENT TECH INST OF MICROELECTRONICS OF THE CHINESE ACAD OF
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANJING INST OF INTELLIGENT TECH INST OF MICROELECTRONICS OF THE CHINESE ACAD OF
- Filing Date
- 2023-06-27
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing convolutional neural networks can only perform a single convolution calculation at a time, resulting in low computation speed.
Method used
Design a pulsating array unit that can simultaneously input two different sets of input data and perform multiplication and addition operations with the same weight data, and perform one more multiplication and addition operation within the same cycle through the pulsating array structure.
Benefits of technology
It improves the computation speed of neural networks with high weight reuse, such as convolutional neural networks.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN116702851B_ABST
Abstract
本发明公开了一种适用于权重复用神经网络的脉动阵列单元及脉动阵列结构，所述脉动阵列单元包括：权重寄存器，第一输入寄存器，第二输入寄存器，第一部分和寄存器，第二部分和寄存器，第一乘法器，第二乘法器，第一累加器，第二累加器；通过上述结构组合，使本发明的脉动阵列单元能够同时输入两组输入数据与同一权重数据进行乘累加计算，比起常规脉动矩阵的单组数据计算的方式，以本发明的脉动阵列单元为基础构建的脉动矩阵更加适用于处理权重复用度高的神经网络的矩阵运算，且由于每个周期能够同时计算两组乘加运算，从而提高了权重复用度高的神经网络(例如卷积神经网络)的计算速度。
Need to check novelty before this filing date? Find Prior Art