NEURAL NETWORK PROCESSING SYSTEM

專利類型

發明

專利國別 (專利申請國家)

美國

專利申請案號

15/649,340

專利證號

US 10,902,317 B2

專利獲證名稱

NEURAL NETWORK PROCESSING SYSTEM

專利所屬機關 (申請機關)

國立交通大學

獲證日期

2021/01/26

技術說明

本發明提供一種神經網路處理系統，包括至少一突觸，突觸係接收一輸入信號，突觸具有一外部權重值及一內部權重值，內部權重值會經由外部刺激而產生變化，當內部權重值之變化累積至一臨界值時，會同時改變外部權重值，使得輸入信號乘上突觸的外部權重值會產生一權重訊號，一神經元電路係連接突觸，以接收突觸所傳輸的權重訊號，並可計算這些權重訊號以輸出。本發明可以同時加速深度學習中的預測及學習功能，並可實現高精確度且具即時學習能力的硬體神經網路。 A neural network processing system is disclosed, which comprises at least one synapse, the synapse receives an input signal, and the synapse has an external weight value and an internal weight value, the internal weight value changes through external stimuli, when the change of the internal weight value is accumulated to a critical value, the external weight value is changed at the same time so that the input signal multiplied by the external weight of the synapse will produce a weight signal. A neuron circuit is connected to the synapse to receive the weight signal transmitted by the synapse and to compute the weight signal to output. The present invention can simultaneously accelerate the prediction and learning function in the deep learning and realize the hardware neural network with high precision and instant learning ability.

備註

連絡單位 (專責單位/部門名稱)

智慧財產權中心

連絡電話

03-5738251

網址