SPARSITY-AWARE NEURAL PROCESSING UNIT FOR PERFORMING CONSTANT PROBABILITY INDEX MATCHING AND PROCESSING METHOD OF THE SAME

• Main Authors: RYU SUNG-JU, KIM JAE-JUN, OH YOUNG-TAEK
• Format: Patent
• Language: Chinese; English
• Published: 01.10.2021
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention relates to a sparsity-aware neural processing unit for performing constant probability index matching and a processing method of the same. A method of processing of a sparsity-aware neural processing unit includes receiving a plurality of input activations (IAs); obtaining a weight having a non-zero value in each weight output channel; obtaining an input channel index, wherein the input channel index stores the weight and the IAs in a memory and includes a memory address position in which the weight and the IAs are stored; and in an index matching unit (IMU) including a buffer memory storing the index of the input channel, arranging the weight of the non-zero value of each weight output channel according to the row size of the IMU, and matching the weight and the IAs. 本发明涉及与IA及加权的密度无关地执行恒定概率的指标匹配的稀疏性感知神经处理单元及其处理方法。根据本发明的实施例，稀疏性感知神经处理单元处理方法可包括：接收多个输入激活(Input Activation，IA)的步骤；从各加权输出通道获取具有非零值的加权(Weight)的步骤；获得输入通道指标的步骤，其中，所述输入通道指标将所述加权及IA存储在存储器并包括存储有所述加权及IA的存储器地址位置；在包括存储所述输入通道指标的缓冲存储器的指标匹配单元(Index