EP3718054A1 - Réseau neuromimétique et procédé de fabrication associé - Google Patents
Réseau neuromimétique et procédé de fabrication associéInfo
- Publication number
- EP3718054A1 EP3718054A1 EP18807648.3A EP18807648A EP3718054A1 EP 3718054 A1 EP3718054 A1 EP 3718054A1 EP 18807648 A EP18807648 A EP 18807648A EP 3718054 A1 EP3718054 A1 EP 3718054A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- electrode
- barrier layer
- conductive material
- ferroelectric
- stack
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/061—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using biological neurons, e.g. biological neurons connected to an integrated circuit
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
- G06N3/065—Analogue means
-
- G—PHYSICS
- G11—INFORMATION STORAGE
- G11C—STATIC STORES
- G11C11/00—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor
- G11C11/21—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using electric elements
- G11C11/22—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using electric elements using ferroelectric elements
- G11C11/221—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using electric elements using ferroelectric elements using ferroelectric capacitors
-
- G—PHYSICS
- G11—INFORMATION STORAGE
- G11C—STATIC STORES
- G11C11/00—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor
- G11C11/21—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using electric elements
- G11C11/22—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using electric elements using ferroelectric elements
- G11C11/223—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using electric elements using ferroelectric elements using MOS with ferroelectric gate insulating film
-
- G—PHYSICS
- G11—INFORMATION STORAGE
- G11C—STATIC STORES
- G11C11/00—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor
- G11C11/54—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using elements simulating biological cells, e.g. neuron
-
- H—ELECTRICITY
- H10—SEMICONDUCTOR DEVICES; ELECTRIC SOLID-STATE DEVICES NOT OTHERWISE PROVIDED FOR
- H10B—ELECTRONIC MEMORY DEVICES
- H10B51/00—Ferroelectric RAM [FeRAM] devices comprising ferroelectric memory transistors
-
- H—ELECTRICITY
- H10—SEMICONDUCTOR DEVICES; ELECTRIC SOLID-STATE DEVICES NOT OTHERWISE PROVIDED FOR
- H10D—INORGANIC ELECTRIC SEMICONDUCTOR DEVICES
- H10D64/00—Electrodes of devices having potential barriers
- H10D64/60—Electrodes characterised by their materials
- H10D64/66—Electrodes having a conductor capacitively coupled to a semiconductor by an insulator, e.g. MIS electrodes
- H10D64/68—Electrodes having a conductor capacitively coupled to a semiconductor by an insulator, e.g. MIS electrodes characterised by the insulator, e.g. by the gate insulator
- H10D64/689—Electrodes having a conductor capacitively coupled to a semiconductor by an insulator, e.g. MIS electrodes characterised by the insulator, e.g. by the gate insulator having ferroelectric layers
Definitions
- Neuromimetic network and method of manufacturing the same
- the present invention relates to a neuromimetic network.
- the present invention also relates to a method of manufacturing such a neuromimetic network.
- Networks reproducing the functioning of a human brain are used for specific tasks for which classical architectures, such as Von Neumann architectures, are poorly adapted.
- tasks include recognizing objects or people in images.
- Such networks include two types of components or main circuits performing two distinct functions.
- One type of component exhibits oscillator behavior similar to that of a neuron, and a second type of component with controllable resistance acts as a synapse connecting two neurons to each other.
- networks of this type are called "neuromimetic networks", “neural networks” or “neuromorphic networks”
- Neuromorphic networks are frequently computer-emulated, that is, their functions are reproduced by a computer program. However, such networks then assume the use of a computer having a complex type of conventional architecture, even if the emulated neuromorphic network is simple.
- Neuromorphic networks in which the roles of neurons and synapses are played by dedicated physical components are also known.
- the synapses comprise controllable resistors of the memristor type, using materials such as phase change materials, oxides or ferroelectric tunnel junctions.
- Such synapses have a variable electrical resistance to change the connection between two neurons, and thus control the propagation of electrical pulses between neurons.
- Neurons they are frequently made from conventional technologies of silicon-based microelectronics, especially CMOS technology (acronym from the English term "Complementary Metal-Oxide-Semiconductor").
- a neuromimetic network comprising a substrate, a set of neurons and a set of synapses, at least one neuron comprising a first stack of superimposed layers in a first stacking direction, the first stack comprising successively according to the first direction of stacking:
- a first electrode carried by the substrate is
- a first barrier layer made of an electrically insulating material
- the first electrode, the first barrier layer and the second electrode forming a first ferroelectric tunnel junction
- At least one synapse comprising a second stack of superimposed layers in a second stacking direction, the second stack comprising successively in the second stacking direction:
- a third electrode carried by the substrate is
- a second barrier layer made of an electrically insulating material
- the third electrode, the second barrier layer and the fourth electrode forming a second ferroelectric tunnel junction.
- the neuromorphic network comprises one or more of the following characteristics, taken in isolation or in any technically possible combination:
- the first barrier layer is made of a ferroelectric material and has a unipolar polarization cycle.
- the first barrier layer is made of antiferroelectric material.
- the second barrier layer is made of a ferroelectric material composed of atoms of a set of elements and the antiferroelectric material constituting the first barrier layer comprises atoms of each element of the set of elements, the antiferroelectric material comprising, in addition, atoms of an additional element not belonging to the set of elements.
- the first electrode is made of a first conductive material and the second electrode is made of a second conductive material different from the first conductive material.
- at least one neuron comprises an electroresistive component having a variable electrical resistance, the electroresistive component being electrically connected to an electrode of the first corresponding stack, the neuron being configured to receive an electric current flowing through the electroresistive component, the electric current passing through, , successively all the layers of the first stack according to the first stacking direction.
- a neuromimetic network comprising a substrate, a set of neurons and a set of synapses, the manufacturing method comprising steps of:
- each barrier layer of the first set Forming an electrode, called a second electrode, on each barrier layer of the first set to form a set of neurons, and forming a fourth electrode on each barrier layer of the second set to form a set of synapses, each barrier layer forming , with the corresponding electrodes, a ferroelectric tunnel junction.
- a second electrode Forming an electrode, called a second electrode, on each barrier layer of the first set to form a set of neurons, and forming a fourth electrode on each barrier layer of the second set to form a set of synapses, each barrier layer forming , with the corresponding electrodes, a ferroelectric tunnel junction.
- the method comprises one or more of the following characteristics, taken in isolation or in any technically possible combination:
- each barrier layer deposited during the deposition step is made of a ferroelectric material, the manufacturing method further comprising, before the forming step, a step of inserting atoms of at least one element; additional in each barrier layer of the first set to transform the ferroelectric material of the barrier layers of the first set into an antiferroelectric material.
- the insertion step comprises the implantation of atoms of the additional element in each barrier layer of the first set.
- the training stage includes:
- third conductive material Depositing an electrically conductive material, called third conductive material, on each barrier layer of the second set to form the corresponding fourth electrode, the third conductive material being different from the second conductive material.
- FIG. 1 is a partial schematic representation of an example of a neuromimetic network comprising a set of neurons and a set of synapses,
- FIG. 2 is a schematic partial sectional view of a neuron and a synapse of FIG. 1, including ferroelectric tunnel junctions,
- FIG. 3 is a graph showing the variation of the electrical resistance of an example of a ferroelectric tunnel junction of FIG. 2;
- FIG. 4 is a set of graphs presenting examples of polarization cycles and ferroelectric tunnel junction resistance.
- FIG. 5 is a flow chart of the steps of a method of manufacturing the neuromimetic network of FIG. 1;
- FIG. 6 is a diagram of an electrical circuit equivalent to a neuron of FIG. 1,
- FIG. 7 is a set of graphs describing the temporal variation of electrical parameters of the circuit of FIG. 6,
- FIG. 8 is a diagram of an electrical circuit equivalent to another example of a neuron.
- FIG. 9 is a graph showing the electrical behavior of the circuit of FIG. 8.
- FIG. 10 An example of a neuromimetic network 10 has been shown in FIG.
- the neuromimetic network 10 comprises a substrate 15, a set of neurons 20, and a set of synapses 25.
- the neuromimetic network 10 is configured to receive at least one input electrical current CE and to generate in response at least one output electrical current CS.
- the neuromimetic network 10 is configured so that each EC input electric current is distributed over the set of neurons 20 by the set of synapses 25, each output electrical current CS being obtained at the output of at least one neuron 20.
- the substrate 15 is configured to support the set of neurons 20 and the set of synapses 25.
- the substrate 15 has a flat top face.
- the upper face 30 is perpendicular to a direction called normal direction DN.
- the substrate 15 is made of an inorganic crystalline material.
- the substrate 15 is made of a material selected from Si, SiO 2, Al 2 O 3, SrTiO 3, MgO, NdGaO 3, GdScO 3, YaAlO 3, LaAlO 3 and mica.
- the substrate 15 is made of a non-crystalline inorganic material such as a glass or a ceramic.
- the substrate 15 is made of an organic material such as a plastics material.
- the substrate 15 comprises, for example, additional elements such as a set of electrical interconnections, one or more power supplies of the neuromimetic network 10 or means for programming the neuromimetic network 10.
- All or part of the additional elements are, for example, carried by a lower face of the substrate 15.
- the additional elements are, for example, made by a CMOS technology.
- Each neuron 20 is configured to be traversed by an electric current C.
- Each neuron 20 comprises a first stack 35 of superimposed layers in a first stacking direction D1.
- the first stack 35 is configured to be traversed by an electric current C in the first stacking direction D1.
- the first stacking direction D1 is, for example, parallel to the normal direction DN.
- FIG. 35 An example of first stack 35 has been shown in FIG.
- the first stack 35 comprises a first electrode 40, a second electrode 45 and a first barrier layer 50.
- the first electrode 40 is carried by the substrate 15.
- the first electrode 40 is delimited, according to the first stacking direction D1, by the upper face 30 of the substrate 15 and by the first barrier layer 50.
- the first electrode 40 is flat.
- the first electrode 40 has a first thickness e1 measured in the first stacking direction D1.
- the first thickness e1 is, for example, between 1 nm and 100 nm.
- the first electrode 40 is made of a first conductive material M1. It is understood by "conductive material” an electrically conductive material.
- the first conductive material M1 is, for example, a metallic material. According to one embodiment, the first material M1 is platinum.
- the second electrode 45 is carried by the first barrier layer 50.
- the second electrode 45 has a second thickness e2 measured in the first stacking direction D1. The second thickness e2 is between 5 nm and 100 nm.
- the second electrode 45 is made of a second conductive material M2.
- the second conductive material M2 is, for example, a metallic material.
- the second conductive material M2 is different from the first conductive material M1.
- the first barrier layer 50 is delimited, in the first stacking direction D1, by the first electrode 40 and by the second electrode 45.
- the first barrier layer 50 has a third thickness e3, measured according to the first stacking direction D1.
- the third thickness e3 is between one nanometer and five nanometers.
- the first barrier layer 50 is able to form a barrier between the first electrode 40 of the second electrode 45.
- the first barrier layer 50 is made of an electrically insulating material.
- the first barrier layer 50 is configured so that the first electrode 40, the first barrier layer 50 and the second electrode 45 form a first ferroelectric tunnel junction. This means that the first barrier layer 50 is configured to be traversed by the tunnel effect by an electric current C successively passing through the first electrode 40, the first barrier layer 50 and the second electrode 45, and that the first barrier layer 50 is made of a ferroelectric material or antiferroelectric material.
- the first barrier layer 50 is made of a ferroelectric material MF. It is understood by "ferroelectric" that the ferroelectric material MF has a plurality of electric dipoles each generating an electrical moment, and, in the absence of an external electric field, the first barrier layer 50 has a non-zero electrical bias PE.
- the electric polarization PE is defined as the average, per unit volume, of the electrical moments.
- the electric polarization PE is therefore a vector quantity.
- the electric polarization PE is parallel to the first direction D1.
- the electric polarization PE is variable.
- the electric polarization PE is capable of being modified by a potential difference V applied between the electrodes of a ferroelectric tunnel junction.
- PE electric polarization is movable between a first orientation and a second orientation opposite to the first orientation.
- the electric polarization PE has a polarization value P.
- the absolute value of the polarization value P is equal to a standard of the electric polarization PE, and the sign of the polarization value P is positive if the electric polarization PE presents the first orientation and negative if the electric polarization PE has the second orientation.
- a representation of the variation of the electric polarization PE or the polarization value P as a function of the potential difference V is called a "polarization cycle".
- the bias value P is likely to vary between a first extreme value ve1 and a second extreme value ve2.
- the first extreme value ve1 and the second extreme value ve2 have identical absolute values, but different signs.
- An electrical resistance R is defined for each ferroelectric tunnel junction. In a ferroelectric tunnel junction, the electrical resistance R depends on the polarization value P.
- a first resistance value Roff is defined for the ferroelectric tunnel junction considered.
- the first resistance value Roff is, for example, between 100 KiloOhm (kOhm) and 100 gigaOhm (GOhm).
- a second resistance value Ron is defined for the ferroelectric tunnel junction considered.
- the second resistance value Ron is strictly less than the first resistance value Roff.
- the second resistance value Roff is, for example, between 100 Ohm and 100 kOhm.
- each ferroelectric tunnel junction exhibits a hysteresis, ie, the ferroelectric tunnel junction tends to remain in a certain state when the external cause that produced the change of state has ceased.
- the polarization value P In a first state, the polarization value P is equal to the first extreme value ve1. In a second state, the polarization value P is equal to the second extreme value ve2.
- a first coercive voltage Vc1 and a second coercive voltage Vc2 are defined for each ferroelectric tunnel junction.
- the ferroelectric tunnel junction considered When the ferroelectric tunnel junction considered is in the first state, and the potential difference V is increased progressively, the ferroelectric tunnel junction considered switches to the second state when the potential difference V reaches the first coercive voltage Vc1.
- the ferroelectric tunnel junction considered When the ferroelectric tunnel junction considered is in the second state, and the potential difference V is progressively diminished, the ferroelectric tunnel junction considered switches to the first state when the potential difference V reaches the second coercive voltage Vc2.
- the first coercive voltage Vc1 is strictly greater than the second coercive voltage Vc2.
- Vc2 there is a range of potential differences V for which the ferroelectric tunnel junction considered is likely to have either the first extreme value ve1 or the second extreme value ve2.
- the first stack 35 is configured so that the first barrier layer 50 has a unipolar polarization cycle.
- FIG. 4 An example of a bipolar polarization cycle Cb and an example of unipolar polarization cycle Cu are shown in Figure 4 for comparison.
- FIG. figure 4 A graph showing the variation of the resistor R of a ferroelectric tunnel junction having a unipolar polarization cycle, on which the first and second resistor values Ron, Roff and the first and second coercive voltages Vc1, Vc2 are indicated is shown in FIG. figure 4.
- the first conductive material M1 and the second conductive material M2 are chosen so that the polarization cycle of the first barrier layer 50 is unipolar.
- the first conductive material M1 has a first output work
- the second conductive material M2 presents a second output work and the difference between the first output job and the second output job is greater than or equal to the difference between the two coercive voltages Vc2, Vc1.
- a first stack 35 in which the first material M1 is SrRu0 3 , the ferroelectric material is BaTi0 3 and the second material M2 is Al is an example of a first stack having a unipolar polarization cycle.
- first conductive material M1 and the second conductive material M2 have different electrical conductivities.
- the first conductive material M1 and the second conductive material M2 have different crystalline structures.
- Each synapse 25 is configured to receive, from a neuron 20, the electric current C and to transmit the electric current C to a second neuron 20.
- Each synapse 25 comprises a second stack 52 of superimposed layers in a second stacking direction D2.
- the second stacking direction D2 is parallel to the first stacking direction D1.
- the second stack 52 is configured to be traversed by the electric current C along the second stacking direction D2.
- the second stack 52 comprises a third electrode 55, a fourth electrode 60, and a second barrier layer 65.
- the second stack 52 is connected to the first stacks 35 of the two corresponding neurons 20.
- the third electrode 55 is connected to the second electrode 45 of one of the neurons 20 and the fourth electrode 60 is connected to the first electrode 40 of the other neuron 20 considered.
- the third electrode 55 is carried by the substrate 15.
- the third electrode 55 is delimited, according to the second stacking direction D2, by the upper face 30 of the substrate 15 and by the second barrier layer 65.
- the third electrode 55 is made of an electrically conductive material.
- the third electrode 55 is made of the first conductive material M1.
- the third electrode 55 has a fourth thickness e4.
- the fourth thickness e4 is identical to the first thickness e1.
- Each fourth electrode 60 is carried by the corresponding second barrier layer 65.
- Each fourth electrode 60 is made of a third conductive material M3.
- the third conductive material M3 is different from the second conductive material M2.
- the third conductive material M3 is identical to the first conductive material M1.
- the fourth electrode 60 has a fifth thickness e5.
- the fifth thickness e5 is between 5 nanometers and 100 nanometers.
- the second barrier layer 65 is configured so that the third electrode 55, the second barrier layer 65 and the fourth electrode 60 form a second ferroelectric tunnel junction.
- the second barrier layer 65 is made of a ferroelectric material.
- the second barrier layer 65 is made of the same ferroelectric material MF as the first barrier layer 50.
- the ferroelectric materials are different from each other.
- the second barrier layer 65 has a sixth thickness e6, measured along the second stacking direction D2.
- the sixth thickness e6 is identical to the third thickness e3.
- the second stack 52 is configured so that the second barrier layer 65 has a bipolar polarization cycle.
- the second stack 52 has an electric memristor behavior.
- FIG. 10 A flow chart of the steps of a method of manufacturing the neuromimetic network 10 has been shown in FIG.
- the manufacturing method comprises a step 100 of obtaining, a step 1 10 deposition, a step 120 of forming a second electrode 45, a step 130 of forming a fourth electrode 60, and a step 140 of finalization.
- a first set of first electrodes 40 and a second set of second electrodes 55 are obtained.
- the first electrodes 40 and the third electrodes 55 are obtained by deposition of the first material M1 on the upper face 30 of the substrate 15.
- ALD alternating-flow chemical vapor deposition
- a barrier layer 50, 65 is deposited on each first electrode 40 and on each third electrode 55.
- the ferroelectric material MF is deposited on each first electrode 40 to form a corresponding first barrier layer 50, and on each third electrode 55 to form a corresponding second barrier layer 65.
- the second conductive material M2 is deposited on each first barrier layer 50 of the first set.
- the second conductive material M2 is not deposited on the second barrier layers 65 of the second set.
- each second barrier layer 65 of the second assembly is covered with a layer of resin preventing the deposition of the second conductive material M2 on the second barrier layer 65.
- the resin covering the second barrier layers 65 is removed, for example by dissolution in a chemical bath.
- a fourth electrode 60 is formed on each second barrier layer 65.
- the third conductive material M3 is deposited on each second barrier layer 65 of the second set to form a corresponding fourth electrode 60.
- each second electrode 45 is covered with a resin layer preventing the deposition of the third conductive material M3 on the second conductive material M2.
- the resin still present on the substrate 15, or on one of the stacks 35, 52 is removed.
- each synapse 25 is connected to at least two distinct neurons.
- the third electrode 55 of each synapse 25 is connected to at least one neuron 20 and the fourth electrode 60 of each synapse 25 is connected to at least one neuron 20.
- the steps of obtaining 100, depositing 1 10, forming 120 of a second electrode 45, forming 130 of a fourth electrode 65, and a finalizing step 140 are implemented after the manufacturing additional elements carried by the substrate 15.
- steps 100 to 1 10 are implemented during a "back-end of line" phase of the manufacturing process of the neuromimetic network 10.
- a "back-end of line” phase is the second phase of a method for producing an integrated circuit, in a first phase during which the non-metallic portions of the transistors of the integrated circuit are formed.
- the first phase comprises the formation of oxide or nitride layers, the deposition of semiconductor layers, the doping of portions of semiconductor layers or even etchings.
- the metal interconnections are, among others, formed.
- FIG. 6 A diagram of an electrical circuit equivalent to a first stack 35 is shown in FIG. 6.
- the first stack 35 has been shown electrically connected to a voltage source Vdc.
- the first stack 35 is electrically equivalent to a circuit formed of a load resistor Rs placed in series with an assembly formed of a capacitor having a capacitor Cd connected in parallel with the resistor R of the first ferroelectric tunnel junction.
- the load resistance Rs corresponds to the electrical resistance of the first and second electrodes 40, 45.
- the capacitor Cd corresponds to the capacitive behavior of the first stack 35, caused by the juxtaposition of the first and second metal electrodes 40, 45 separated by the first insulating barrier layer 50.
- Resistance R of the first ferroelectric tunnel junction has a unipolar resistance cycle.
- the first stack 35 considered in the simulations has a first resistance value Roff equal to 10 megohm, a second resistance value Ron equal to 1 kiloohm, a capacitance Cd equal to 450 femtofarad, a load resistance Rs equal to 5 kiloohm, a first coercive voltage Vc1 equal to 0.8 volts and a second coercive voltage Vc2 equal to 0.2 volts.
- the voltage source Vdc imposes on the first stack 35 a potential difference equal to 1 volt.
- the results of simulations comprise a graph 200 showing on the ordinate the variation over time of the resistance R (in Ohm) of the first ferroelectric tunnel junction, a graph 210 presenting on the ordinate the variation over time of the intensity (in milliamperes) of the current C passing through the first stack 35 and a graph 220 having as ordinate the variation over time of the potential difference V (in volts) between the first electrode 40 and the second electrode 45.
- the three graphs 200, 210 and 220 have a common abscissa scale graduated in microseconds.
- the graphs 200 to 220 show that the first stack 35 has an oscillating behavior in which the resistance R, the intensity of the current C and the potential difference V vary periodically over time with a frequency F.
- the first stack 35 behaves electrically as a relaxation oscillator. Such behavior is similar to that of a neuron.
- the first stack 35 is therefore adapted to play the role of a neuron 20 in the neuromimetic network 10.
- the neuromimetic network 10 is easier to manufacture than the neuromimetic networks of the state of the art.
- the neuromimetic network 10 does not involve combining synapses and neurons derived from separate manufacturing technologies.
- neurons and synapses are directly formed on the same substrate 15.
- the neuromimetic network is therefore easier to manufacture than the neuromimetic networks of the state of the art.
- the neuromimetic network 10 has, in addition, a greater density of neurons and synapses than the neuromimetic networks of the state of the art.
- neuromimetic network 10 therefore has better reliability than the neuromimetic networks of the state of the art and high durability.
- a second example of a neuromimetic network 10 will now be described.
- the elements identical to the first example of a neuromimetic network are not described again. Only the differences are highlighted.
- the first conductive material M1 is identical to the second conductive material M2.
- Each first barrier layer 50 is made of antiferroelectric material MA.
- the dipoles are oriented antiparallel to each other without completely compensating each other.
- the antiferroelectric material MA has an antiferroelectric behavior at zero voltage and a hysteretic transition to a ferroelectric state with positive or negative voltage.
- the antiferroelectric material MA has a polarization value P of less than or equal to 5 microcoulomb per square centimeter.
- the first ferroelectric tunnel junction exhibits a behavior similar to the behavior of a ferromagnetic material having a unipolar polarization cycle.
- the polarization cycle of the first ferroelectric tunnel junction is similar to the union of two unipolar polarization cycles, one for the negative V potential differences and the other for the positive V potential differences.
- FIG. 4 An example of an antiferroelectric polarization cycle Ca having a hysteretic transition to a positive or negative voltage ferroelectric state is shown in FIG. 4.
- FIG. 4 An example of a cycle Cr3 describing the variation of the electrical resistance R during the antiferroelectric polarization cycle Ca is also shown in Figure 4.
- the ferroelectric material MF is composed of atoms of a set of elements.
- the ferroelectric material MF is BiFe0 3 .
- the set of elements is formed of oxygen, bismuth and iron.
- the antiferroelectric material MA comprises atoms of each element of the set of elements.
- the antiferroelectric material comprises bismuth atoms, oxygen atoms and iron atoms.
- the antiferroelectric material MA further comprises atoms of at least one additional element that does not belong to the set of elements.
- the additional element is a rare earth.
- the rare earth is, for example, a rare earth of valence 3.
- Samarium, neodymium, lanthanum, gadolinium and dysprosium are examples of rare earths of valence 3.
- each barrier layer 50, 65 is made of the ferroelectric material as in the first example.
- a single training step is performed, during which the second electrodes 45 and the fourth electrodes 60 are made simultaneously.
- the manufacturing method further comprises an insertion step.
- the insertion step is, for example, carried out between the deposition step 1 10 and the training step 120.
- the ferroelectric material MF composing the first barrier layers 50 is converted into antiferroelectric material MA.
- atoms of the additional element are inserted in each first layer barrier 35 of the first set for transforming the ferroelectric material MF of each first barrier layer 35 into the antiferroelectric material MA.
- the atoms of the additional element are, for example, inserted by implantation.
- Implantation also known as ion implantation
- ions of an element are accelerated and projected at high speed onto a material to insert ions into the material.
- the atoms of the additional element are inserted by ion diffusion.
- Ionic diffusion is a technique in which the material to be modified is brought into contact with a source material containing atoms of the additional element, the atoms of the additional element migrating from the source material into the material to be modified by diffusion.
- the antiferroelectric material MA is composed of atoms of the same elements as the ferroelectric material MF, but in different proportions. The proportions of the atoms are then modified during the insertion step, that is to say that the additional element is an element of the set of elements.
- At least one neuron 20 comprises an electroresistive component 70.
- the electroresistive component 70 is electrically connected to an electrode 40, 45 of the corresponding first stack 35.
- the electroresistive component 70 is configured to receive the electrical current C.
- the electroresistive component 70 is configured to be traversed by the electric current C.
- the electroresistive component 70 is interposed between a synapse 25 and the first electrode 40. In a variant, the electroresistive component 70 is interposed between a synapse 25 and the second electrode 45.
- the electroresistive component 70 has a variable electrical resistance Rv.
- the variable electrical resistance Rv can be modified by a user of the neuromimetic network 10.
- the electroresistive component 70 is, for example, a memristor.
- a ferroelectric memristor is an example of a memristor.
- FIG. 8 An electrical diagram equivalent to the neuron 20 is shown in FIG. 8.
- the series resistor Rs is considered to be a component of the variable electrical resistance Rv and has therefore not been represented in FIG. 8.
- the electroresistive component 70 is a transistor.
- the electroresistive component 70 is a transistor whose electrical resistance variable Rv is likely to be modified according to a gate voltage Vg applied to the transistor.
- a graph 230 representing the variation of the frequency F as a function of the variable electrical resistance Rv and the DC voltage Vdc applied to the assembly formed of the electroresistive component 70 and the first stack 35 connected in series. As shown in graph 230, the frequency F depends on the variable electrical resistance.
- the neuromimetic network 10 is then adaptable to a wide range of uses.
- the modulation of the series resistance Rs makes it possible to modify the frequency of the pulses emitted by the neurons.
- neurons comprising such an electroresistive component 70 form an input layer of the neuromimetic network 10, i.e., all of the neurons 20 receiving the EC input electric current (s).
- the electroresistive components 70 make it possible to code for each neuron 20 of the input layer an analog input information in the form of a pulse frequency.
- each of the first, second and third conductive materials M1, M2, M3 may be selected from the group consisting of metals, conductive oxides and semiconductor oxides.
- Each metal is, for example, selected from the group consisting of: Pt, Pd, Au, Co, W, Al, Ir, Cu, Ni and Cs.
- Each conductive oxide is, for example, selected from the group consisting of: SrRu0 3 LaNi0 3, (Ca, Ce) Mn0 3, (La, Sr) Mn0 3, La 0, 5SR 0.5 CoO 3, IR0 2 and Ru0 2 .
- Each semiconductor oxide is, for example, selected from the group consisting of: ZnO, Sn0 2 , In 2 0 3 and ITO.
- the semiconductor oxide is, for example, doped.
- Each antiferroelectric material is, for example, selected from the group consisting of: PbZr0 3 , Pb (In, Nb) O 3 , R ⁇ _ C B3 C Zg0 3 , PbZr ⁇ Sn x Os, PbHf0 3 , Pb (In 1 / 2 Nb 1/2 ) 0 3 , Pbo 89 Nbo o 2 [(Zr 57 Sn O 43 ) o 94 TiO 6 ] o 98 0 3 , (Bi, Y) FeO 3 , (Bi, Sc) FeO 3 , solid type solutions (Bi, RE) Fe0 3 where RE is a rare earth valence 3 NaNb0 3 AgNb0 3 SmFe0 3 Sri_ x Ca x Ti0 3, Pb (ybi / 2 Nbi / 2) 0 3, Pb (ybi / 2 Tai / 2) 0 3, Pb (Coi / 2 W 1/2) 0
- Each ferroelectric material is, for example, selected from the group consisting of: PbMgi / 3 Nb 2/3 0 3 , (PbMgi / 3 Nb 2/3 0 3 ) i- x (PbTiO 3) x , PbZni / 3 Nb 2 / 3 0 3 , (PbZni / 3 Nb 2/3 0 3 ) i- x (PbTi0 3) x, PbSco5Nbo5O3 (PbSco5Nbo 503) x i- (PbSco5Tao50 3) x, x Zr x BaTii- 0 3 Bai- x Sr x Ti0 3, BaTii- x Sn x 0 3, BaTi0 3, PbZn- x Ti x OS and BiFe0 3 .
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Neurology (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Hardware Design (AREA)
- Microelectronics & Electronic Packaging (AREA)
- Power Engineering (AREA)
- Semiconductor Memories (AREA)
Abstract
Description
Claims
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| FR1701263A FR3074337B1 (fr) | 2017-11-30 | 2017-11-30 | Reseau neuromimetique et procede de fabrication associe |
| PCT/EP2018/083087 WO2019106127A1 (fr) | 2017-11-30 | 2018-11-30 | Réseau neuromimétique et procédé de fabrication associé |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| EP3718054A1 true EP3718054A1 (fr) | 2020-10-07 |
Family
ID=61873347
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| EP18807648.3A Pending EP3718054A1 (fr) | 2017-11-30 | 2018-11-30 | Réseau neuromimétique et procédé de fabrication associé |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US11551749B2 (fr) |
| EP (1) | EP3718054A1 (fr) |
| CN (1) | CN111712838B (fr) |
| FR (1) | FR3074337B1 (fr) |
| WO (1) | WO2019106127A1 (fr) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP7429431B2 (ja) * | 2020-02-27 | 2024-02-08 | 国立研究開発法人産業技術総合研究所 | 情報処理装置および情報処理装置の駆動方法 |
| CN111598229B (zh) * | 2020-05-11 | 2024-10-15 | 中国科学院微电子研究所 | 一种神经元电路、基于神经网络的集成电路和电子设备 |
| US12207565B2 (en) | 2022-01-25 | 2025-01-21 | Eagle Technology, Llc | Multi-level multiferroic memory device and related methods |
| US12274073B2 (en) | 2022-01-25 | 2025-04-08 | Eagle Technology, Llc | Multiferroic memory with piezoelectric layers and related methods |
| US12329040B2 (en) * | 2022-01-25 | 2025-06-10 | Eagle Technology, Llc | Multiferroic tunnel junction memory device and related methods |
| CN115146770B (zh) * | 2022-07-22 | 2025-04-18 | 北京大学深圳研究生院 | 一种基于铁电隧穿结的电子突触电路及神经网络电路 |
Family Cites Families (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP3193302B2 (ja) * | 1996-06-26 | 2001-07-30 | ティーディーケイ株式会社 | 膜構造体、電子デバイス、記録媒体および強誘電体薄膜の製造方法 |
| FR2945147B1 (fr) * | 2009-04-30 | 2012-03-30 | Thales Sa | Dispositif memristor a resistance ajustable grace au deplacement d'une paroi magnetique par transfert de spin et utilisation dudit memristor dans un reseau de neurones |
| FR2946788B1 (fr) * | 2009-06-11 | 2016-11-11 | Thales Sa | Dispositif a resistance ajustable. |
| US8311965B2 (en) * | 2009-11-18 | 2012-11-13 | International Business Machines Corporation | Area efficient neuromorphic circuits using field effect transistors (FET) and variable resistance material |
| CN103246904B (zh) * | 2013-05-24 | 2016-04-06 | 北京大学 | 基于阻变忆阻器的时间关联学习神经元电路及其实现方法 |
| US20140365413A1 (en) * | 2013-06-06 | 2014-12-11 | Qualcomm Incorporated | Efficient implementation of neural population diversity in neural system |
| CN103346256A (zh) * | 2013-07-03 | 2013-10-09 | 南京大学 | 一种基于铁电隧道结的忆阻器 |
| JP5659361B1 (ja) * | 2013-07-04 | 2015-01-28 | パナソニックIpマネジメント株式会社 | ニューラルネットワーク回路、およびその学習方法 |
| CN105373829B (zh) * | 2014-09-02 | 2018-05-04 | 北京大学 | 一种全连接神经网络结构 |
| CN105160401B (zh) * | 2015-08-27 | 2017-08-11 | 电子科技大学 | 一种基于忆阻器阵列的wta神经网络及其应用 |
| WO2017039611A1 (fr) * | 2015-08-31 | 2017-03-09 | Hewlett Packard Enterprise Development Lp | Empilements de matériaux pour memristances unipolaires à faible courant |
| US10650308B2 (en) * | 2015-09-23 | 2020-05-12 | Politecnico Di Milano | Electronic neuromorphic system, synaptic circuit with resistive switching memory and method of performing spike-timing dependent plasticity |
| US10679121B2 (en) * | 2015-12-30 | 2020-06-09 | SK Hynix Inc. | Synapse and a neuromorphic device including the same |
| CN105720194B (zh) * | 2016-03-14 | 2018-05-15 | 广东工业大学 | 用于交叉点内存阵列用的电阻开关与选择器共存器件及制备方法和应用 |
| US9767408B1 (en) * | 2016-09-16 | 2017-09-19 | International Business Machines Corporation | Multi-memristive synapse with clock-arbitrated weight update |
| JP2019179499A (ja) * | 2018-03-30 | 2019-10-17 | ソニー株式会社 | 半導体装置及び積和演算装置 |
| US10453528B1 (en) * | 2018-06-14 | 2019-10-22 | International Business Machines Corporation | Controlling aggregate signal amplitude from device arrays by segmentation and time-gating |
-
2017
- 2017-11-30 FR FR1701263A patent/FR3074337B1/fr active Active
-
2018
- 2018-11-30 WO PCT/EP2018/083087 patent/WO2019106127A1/fr not_active Ceased
- 2018-11-30 EP EP18807648.3A patent/EP3718054A1/fr active Pending
- 2018-11-30 CN CN201880077656.XA patent/CN111712838B/zh active Active
- 2018-11-30 US US16/768,549 patent/US11551749B2/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| US11551749B2 (en) | 2023-01-10 |
| US20200294581A1 (en) | 2020-09-17 |
| CN111712838A (zh) | 2020-09-25 |
| WO2019106127A1 (fr) | 2019-06-06 |
| FR3074337A1 (fr) | 2019-05-31 |
| CN111712838B (zh) | 2024-05-24 |
| FR3074337B1 (fr) | 2021-04-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3718054A1 (fr) | Réseau neuromimétique et procédé de fabrication associé | |
| JP7584899B2 (ja) | メモリスタ、及びそれを含むニューロモーフィック装置 | |
| US9978940B2 (en) | Memristor and method of production thereof | |
| KR101811108B1 (ko) | 부도체-도체 전이현상을 이용한 뉴런 소자를 포함한 고집적 뉴로모픽 시스템 및 고집적 뉴로모픽 회로 | |
| US10147897B2 (en) | Synaptic transistor based on metal nano-sheet and method of manufacturing the same | |
| JP2019179499A (ja) | 半導体装置及び積和演算装置 | |
| US12229652B2 (en) | Two-terminal metastable mixed-conductor memristive devices | |
| Liang et al. | Transparent HfO x-based memristor with robust flexibility and synapse characteristics by interfacial control of oxygen vacancies movement | |
| US20190131384A1 (en) | Ferroelectric tunnel junction structure and method of fabricating the same | |
| KR20200144035A (ko) | 멤리스터 및 이를 포함하는 뉴로모픽 장치 | |
| Zrinski et al. | Composite memristors by nanoscale modification of Hf/Ta anodic oxides | |
| Lee et al. | Energy Efficient Hybrid Reservoir Computing Using Hf0. 5Zr0. 5O2 Ferroelectric Thin‐Film Transistors With an Integrated Optically and Electrically Synaptic Functions | |
| CN111244270A (zh) | 一种模拟生物神经元功能的电子器件及方法 | |
| FR2772508A1 (fr) | Dispositif de memoire ferroelectrique et son procede de pilotage | |
| KR20230162975A (ko) | 재구성 가능한 인메모리 컴퓨팅을 위한 필드 프로그래밍 가능 페로 다이오드들 | |
| KR20240010269A (ko) | 시냅스 소자 및 이의 제조 방법 | |
| WO2010142762A1 (fr) | Dispositif ferroélectrique à résistance ajustable | |
| US20140374693A1 (en) | Varied multilayer memristive device | |
| EP4195207B1 (fr) | Mémoires non volatiles à technologies mixtes oxram/feram | |
| CN118350427A (zh) | 一种多功能可重构神经元电路 | |
| KR102820396B1 (ko) | 시간 커널 소자, 시간 커널 컴퓨팅 시스템 및 이들의 동작 방법 | |
| EP4100885B1 (fr) | Dispositif de classification, ensemble de classification et procédé de classification associé | |
| Shi et al. | Coplanar Floating-Gate Antiferroelectric Transistor with Multifunctionality for All-in-One Analog Reservoir Computing | |
| CN115511064A (zh) | 神经电路、检测电路、检测系统及电路制备方法 | |
| WO2024208723A1 (fr) | Circuit memoire |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: UNKNOWN |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |
|
| PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
|
| 17P | Request for examination filed |
Effective date: 20200529 |
|
| AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
|
| AX | Request for extension of the european patent |
Extension state: BA ME |
|
| RIN1 | Information on inventor provided before grant (corrected) |
Inventor name: GARCIA, VINCENT Inventor name: LOCATELLI, NICOLAS Inventor name: BIBES, MANUEL Inventor name: GROLLIER, JULIE |
|
| DAV | Request for validation of the european patent (deleted) | ||
| DAX | Request for extension of the european patent (deleted) | ||
| RAP3 | Party data changed (applicant data changed or rights of an application transferred) |
Owner name: UNIVERSITE PARIS-SACLAY Owner name: CENTRE NATIONAL DE LA RECHERCHE SCIENTIFIQUE Owner name: THALES |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: EXAMINATION IS IN PROGRESS |
|
| 17Q | First examination report despatched |
Effective date: 20230707 |