阅读说明：本技术 仿神经网络及其制造方法 (Neural network and manufacturing method thereof ) 是由 埃马纽埃尔·比布 朱莉·格罗利耶 樊尚·加西亚 尼古拉·洛卡泰利 于 2018-11-30 设计创作，主要内容包括：本发明涉及一种仿神经网络(10),包括衬底(15)、神经元(20)组和突触(25)组；至少一个神经元(20)包括叠置层的第一堆叠(35),所述第一堆叠(35)依次包括第一电极(40)、由电绝缘材料制成的第一阻挡层(50)以及第二电极(45),所述第一电极(40)、第一阻挡层(50)和第二电极(45)形成第一铁电隧道结；至少一个突触(25)包括叠置层的第二堆叠(52),所述第二堆叠(52)依次包括第三电极(55)、由电绝缘材料制成的第二阻挡层(65)以及第四电极(60),所述第三电极(55)、第二阻挡层(65)和第四电极(60)形成第二铁电隧道结。(The invention relates to an artificial neural network (10) comprising a substrate (15), a set of neurons (20) and a set of synapses (25); at least one neuron (20) comprises a first stack (35) of stacked layers, the first stack (35) comprising, in order, a first electrode (40), a first barrier layer (50) made of an electrically insulating material, and a second electrode (45), the first electrode (40), the first barrier layer (50), and the second electrode (45) forming a first ferroelectric tunnel junction; at least one synapse (25) comprises a second stack (52) of stacked layers, said second stack (52) comprising, in order, a third electrode (55), a second barrier layer (65) made of an electrically insulating material, and a fourth electrode (60), said third electrode (55), second barrier layer (65), and fourth electrode (60) forming a second ferroelectric tunnel junction.)

1. An artificial neural network (10) comprising a substrate (15), a set of neurons (20) and a set of synapses (25),

at least one neuron (20) comprises, in a first stacking direction (D1), a first stack (35) of stacked layers, the first stack (35) comprising, in order in the first stacking direction (D1):

a first electrode (40) supported by the substrate (15),

a first barrier layer (50) made of an electrically insulating material, and

a second electrode (45);

the first electrode (40), first barrier layer (50) and second electrode (45) form a first ferroelectric tunnel junction;

at least one synapse (25) comprises, in at least a second stacking direction (D2), a second stack (52) of superposed layers, said second stack (52) comprising, in order in the second stacking direction (D2):

a third electrode (55) supported by the substrate (15),

a second barrier layer (65) made of an electrically insulating material, an

A fourth electrode (60);

the third electrode (55), the second barrier layer (65), and the fourth electrode (60) form a second ferroelectric tunnel junction.

2. The neuromorphic network (10) of claim 1, wherein the first barrier layer (50) is made of a ferroelectric material and has a period of unipolar polarization.

3. The simulated neural network (10) of claim 1, wherein the first barrier layer (50) is made of an antiferroelectric material.

4. The neuromorphic network (10) of claim 3, wherein the second barrier layer (65) is made of a ferroelectric material and the ferroelectric material is composed of atoms of a group of elements, and the antiferroelectric material making up the first barrier layer (50) contains atoms of each element of the group, the antiferroelectric material also containing atoms of other elements not belonging to the group.

5. The neuromorphic network (10) of any one of claims 1-4, wherein the first electrode (40) is made of a first electrically conductive material and the second electrode (45) is made of a second electrically conductive material different from the first electrically conductive material.

6. The neuromorphic network (10) of any one of claims 1 to 5, wherein at least one neuron comprises a resistive component (70) having a variable resistance, the resistive component (70) being electrically connected with the electrodes (40, 45) of the respective first stack (35), the neuron (20) being configured to receive a current through the resistive component (70), the current further passing through all layers of the first stack (35) in sequence along a first stacking direction (D1).

7. A method for manufacturing a mock neural network (10) comprising a substrate (15), a set of neurons (20) and a set of synapses (25), the method comprising the steps of:

-obtaining (100) a first set of first electrodes (40) supported by the substrate (15) and a second set of electrodes (55), called third electrodes (55), supported by the substrate (15),

-depositing (110) an electrically insulating barrier layer (50, 65) on each first electrode (40) and each third electrode (55),

-forming (120) an electrode (45), called second electrode (45), on each barrier layer (55) of the first group to form a group of neurons (20), and a fourth electrode (60) on each barrier layer (65) of the second group to form a group of synapses (25), each barrier layer (50, 65) forming a ferroelectric tunnel junction with a respective electrode (40, 45, 55, 60).

8. The method for manufacturing an artificial neural network (10) as claimed in claim 7, wherein each barrier layer (50, 65) deposited during the depositing step (110) is made of a ferroelectric material,

the manufacturing method further comprises, before the forming step (120, 130), the step of inserting atoms of at least one further element into each barrier layer (50) of the first group to transform the ferroelectric material of the barrier layers (50) of the first group into an antiferroelectric material.

9. The method for manufacturing the simulated neural network (10) of claim 8, wherein the inserting step comprises implanting atoms of the further element into each barrier layer (50) of the first set.

10. The method for manufacturing a simulated neural network (10) as claimed in claim 7, wherein the forming step (120, 130) comprises:

-depositing a conductive material, called second conductive material, on each barrier layer (50) of the first group to form a respective second electrode (45), and

-depositing on each barrier layer (65) of the second set a conductive material, called third conductive material, different from said second conductive material, to form a respective fourth electrode (60).