Title:
SEMICONDUCTOR DEVICE AND METHOD FOR CONFIGURING NEURAL NETWORK
Document Type and Number:
Japanese Patent JP2020166793
Kind Code:
A
Abstract:
To provide a semiconductor device and a method for configuring a neural network that implement a neural network structure without relying on software.SOLUTION: A semiconductor device 10 can construct a neural network at least composed of a synapse and a start point neuron and an end point neuron coupled together by the synapse. The semiconductor device comprises: a storage unit 14 including a memory cell array 142 that stores data for each entry and a partial match search circuit 144 that searches for an entry containing data that partially matches a search key from data stored in the memory cell array 142; a calculation circuit 16 that calculates a value of the end point neuron based on the data searched for by the partial match search circuit 144; an activation calculation circuit 18 that outputs activation data obtained by passing a calculation result through an activation function; and a data update circuit 20 that rewrites part of the data stored in the storage unit 14 when the activation data is input.SELECTED DRAWING: Figure 1
Inventors:
NAKAMURA KAZUYUKI
Application Number:
JP2019069485A
Publication Date:
October 08, 2020
Filing Date:
March 30, 2019
Export Citation:
Assignee:
KYUSHU INST TECH
International Classes:
G06G7/60; G06F17/10; G06N3/063
Attorney, Agent or Firm:
Takayuki Imanaka