CN113496273A - Magnetic tunnel junction and neuron nonlinear response device - Google Patents
Magnetic tunnel junction and neuron nonlinear response device Download PDFInfo
- Publication number
- CN113496273A CN113496273A CN202010262733.8A CN202010262733A CN113496273A CN 113496273 A CN113496273 A CN 113496273A CN 202010262733 A CN202010262733 A CN 202010262733A CN 113496273 A CN113496273 A CN 113496273A
- Authority
- CN
- China
- Prior art keywords
- magnetic
- layer
- tunnel junction
- bottom electrode
- current
- 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.)
- Granted
Links
- 230000005291 magnetic effect Effects 0.000 title claims abstract description 167
- 230000004044 response Effects 0.000 title claims abstract description 32
- 210000002569 neuron Anatomy 0.000 title claims abstract description 19
- 230000005290 antiferromagnetic effect Effects 0.000 claims abstract description 44
- 230000008878 coupling Effects 0.000 claims abstract description 27
- 238000010168 coupling process Methods 0.000 claims abstract description 27
- 238000005859 coupling reaction Methods 0.000 claims abstract description 27
- 229910001385 heavy metal Inorganic materials 0.000 claims abstract description 11
- 230000004888 barrier function Effects 0.000 claims abstract description 10
- 125000006850 spacer group Chemical group 0.000 claims abstract description 9
- 210000000225 synapse Anatomy 0.000 claims description 35
- 230000009467 reduction Effects 0.000 claims description 22
- 230000000946 synaptic effect Effects 0.000 claims description 20
- 239000000463 material Substances 0.000 claims description 10
- 230000005281 excited state Effects 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 8
- 230000010354 integration Effects 0.000 claims description 7
- 229910052697 platinum Inorganic materials 0.000 claims description 6
- 229910052715 tantalum Inorganic materials 0.000 claims description 6
- 238000000034 method Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- GWEVSGVZZGPLCZ-UHFFFAOYSA-N titanium dioxide Inorganic materials O=[Ti]=O GWEVSGVZZGPLCZ-UHFFFAOYSA-N 0.000 claims description 4
- 238000012546 transfer Methods 0.000 claims description 4
- 229910003321 CoFe Inorganic materials 0.000 claims description 3
- 229910019236 CoFeB Inorganic materials 0.000 claims description 3
- 229910015136 FeMn Inorganic materials 0.000 claims description 3
- 229910005335 FePt Inorganic materials 0.000 claims description 3
- 229910001030 Iron–nickel alloy Inorganic materials 0.000 claims description 3
- 229910019041 PtMn Inorganic materials 0.000 claims description 3
- PNEYBMLMFCGWSK-UHFFFAOYSA-N aluminium oxide Inorganic materials [O-2].[O-2].[O-2].[Al+3].[Al+3] PNEYBMLMFCGWSK-UHFFFAOYSA-N 0.000 claims description 3
- 229910052804 chromium Inorganic materials 0.000 claims description 3
- 229910052593 corundum Inorganic materials 0.000 claims description 3
- 229910052737 gold Inorganic materials 0.000 claims description 3
- GNRSAWUEBMWBQH-UHFFFAOYSA-N nickel(II) oxide Inorganic materials [Ni]=O GNRSAWUEBMWBQH-UHFFFAOYSA-N 0.000 claims description 3
- 230000007306 turnover Effects 0.000 claims description 3
- 229910001845 yogo sapphire Inorganic materials 0.000 claims description 3
- 229910052719 titanium Inorganic materials 0.000 claims description 2
- 230000001537 neural effect Effects 0.000 claims 1
- 238000005265 energy consumption Methods 0.000 abstract description 5
- 230000008901 benefit Effects 0.000 abstract description 2
- 230000000694 effects Effects 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 3
- 238000013529 biological neural network Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000005355 Hall effect Effects 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 210000003050 axon Anatomy 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000011217 control strategy Methods 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000005389 magnetism Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000005062 synaptic transmission Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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; CALCULATING OR 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
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Theoretical Computer Science (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Neurology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Microelectronics & Electronic Packaging (AREA)
- Hall/Mr Elements (AREA)
Abstract
The invention discloses a magnetic tunnel junction, which comprises a top electrode, a bottom electrode and a tunnel junction, wherein the bottom electrode is a heavy metal layer with two ports having spin Hall angles, and the tunnel junction comprises a magnetic free layer, an insulating barrier layer, a magnetic pinning layer, an antiferromagnetic coupling spacer layer, an antiferromagnetic coupling magnetic layer and an antiferromagnetic layer which are sequentially arranged on the bottom electrode; the magnetic free layer, the magnetic pinning layer and the anti-ferromagnetic coupling magnetic layer are all in-plane magnetic moments, the easy axis is along the width direction of the bottom electrode, the magnetic moment directions of the magnetic pinning layer and the anti-ferromagnetic coupling magnetic layer are opposite, and the direction of the anti-ferromagnetic sequence of the anti-ferromagnetic layer is the easy axis direction of the anti-ferromagnetic coupling magnetic layer. The invention also provides a nonlinear response device comprising the magnetic tunnel junction neuron. The magnetic tunnel junction and the corresponding nonlinear response device provided by the invention have the advantages that the nonlinear response characteristic of the device is excellent, the energy consumption is low, and the stability is high.
Description
Technical Field
The invention belongs to the technical field of magnetic elements, and further relates to a magnetic tunnel junction and a neuron nonlinear response device comprising the same.
Background
The von Neumann system adopted by a modern computer system is a system with separated operation and storage, the data transmission speed between an operation core (cpu) and a cache region seriously restricts the increase of the overall calculation rate, and the cache region can bring about the great increase of energy consumption in the data storage process, thereby restricting the improvement of the calculation efficiency. The system of transportation and storage separation is increasingly unable to meet the demand of people for high-performance computation, especially the realization of the recognition function.
After the introduction of the brain-like artificial neural network, the brain-like artificial neural network has recently been regarded as a powerful competitor of von neumann architecture. The biological brain exists in a biological neural network formed by hundreds of millions of neuron cells which are connected with each other through synapses, and after the neuron cells collect input signals transmitted by the synapses, the neuron cells make nonlinear response according to the strength of the signals to determine output signals of the neuron cells and transmit the output signals along axons. The nonlinear response characteristic of the neuron can enable the neural network to have excellent recognition capability, and meanwhile, due to the weight storage function of synapses, the neural network integrating transportation and storage can realize the recognition function which is difficult to realize by a traditional computer with low energy consumption and high efficiency. The electronic element is used for simulating the structure and the characteristics of a biological neural network to establish an artificial neural network, and two devices, namely an artificial synapse and an artificial neuron, exist.
At present, a system formed by a CMOS technology is mainly applied to an artificial neuron nonlinear response system or a traditional computer is directly utilized for simulation, and a traditional switch transistor is still used as an operation core component, so that the formed nonlinear response system is very complicated and cannot be separated from the characteristics of high energy consumption and low efficiency. This limits the development of artificial neural networks to greater integration, high efficiency, and small volume.
Disclosure of Invention
In view of the defects in the prior art, the invention aims to provide a magnetic tunnel junction and a corresponding nonlinear response device so as to realize the simplicity and high efficiency of a neuron nonlinear response system, so that the nonlinear response characteristic of the device is excellent, the energy consumption is low, and the stability is high.
In order to achieve the purpose, the invention adopts the following technical scheme:
a magnetic tunnel junction comprises a top electrode, a bottom electrode and a tunnel junction arranged between the top electrode and the bottom electrode, wherein the bottom electrode is a heavy metal layer with two ports having spin Hall angles, and the tunnel junction comprises a magnetic free layer, an insulating barrier layer, a magnetic pinning layer, an anti-ferromagnetic coupling spacer layer, an anti-ferromagnetic coupling magnetic layer and an anti-ferromagnetic layer which are sequentially arranged on the bottom electrode;
the magnetic free layer, the magnetic pinning layer and the anti-ferromagnetic coupling magnetic layer are all in-plane magnetic moments, the easy axis is along the width direction of the bottom electrode, the magnetic moment directions of the magnetic pinning layer and the anti-ferromagnetic coupling magnetic layer are opposite, and the direction of the anti-ferromagnetic sequence of the anti-ferromagnetic layer is the easy axis direction of the anti-ferromagnetic coupling magnetic layer.
In particular, the top electrode and the tunnel junction have the same cross-sectional shape, and the projections of the top electrode and the tunnel junction in the plane of the bottom electrode are located entirely within the bottom electrode.
Wherein the cross-sectional shape is circular or elliptical; when the cross section is circular, the radius ranges from 50nm to 500 nm; when the cross section is in an oval shape, the value ranges of the major axis and the minor axis are 50 nm-500 nm.
Wherein, the bottom electrode is a strip structure, and the width of the bottom electrode is 500 nm-5 μm.
The bottom electrode is made of Pt or Ta, the top electrode is made of Pt, Ta, Cr, Au or Ti, the magnetic free layer, the magnetic pinning layer and the antiferromagnetic coupling magnetic layer are made of CoFeB, NiFe, CoFe, Co or FePt independently, and the insulating barrier layer is made of MgO and Al2O3Or TiO2The material of the antiferromagnetic coupling spacer layer is Ru, and the material of the antiferromagnetic layer is IrMn, PtMn, FeMn, NiO or NiCoO.
The bottom electrode is 1 nm-10 nm thick, the insulating barrier layer is 0.5 nm-5 nm thick, the antiferromagnetic coupling spacer layer is 0.8 nm-0.9 nm thick, the antiferromagnetic layer is 10 nm-50 nm thick, the magnetic free layer, the magnetic pinning layer and the antiferromagnetic coupling magnetic layer are 1 nm-10 nm thick respectively, and the top electrode is 10 nm-100 nm thick.
The invention also provides a neuron nonlinear response device which comprises a synapse system and the magnetic tunnel junction, wherein the synapse system comprises a synapse array, synapse signals output by the synapse array are connected to one end of the bottom electrode, and the other end of the bottom electrode is grounded; the top electrode is configured to connect a reduction current and a read current to obtain an output voltage.
Specifically, the spin orbit torque generated by the synaptic signal pulse current flowing through the bottom electrode acts on the magnetic free layer, and the probability of the magnetic moment flip of the magnetic free layer has a similar sigmoid function relationship with the magnitude of the pulse current at a certain pulse length of the synaptic signal input current, which is specifically as follows (1):
wherein w is a constant of intrinsic properties of the magnetic free layer and the heavy metal layer bottom electrode and is related to the pulse time Deltat, T is temperature, ISYNThe current is synaptic current, and I is the current magnitude when the turnover probability is 50%; w and I are both functions of the pulse time Δ t; specifically, the following formulae (2) and (3):
specifically, the initial state of the magnetic tunnel junction is a state in which magnetic moments of the magnetic free layer and the magnetic pinned layer are parallel, and corresponds to a relatively low output voltage; the excited state of the magnetic tunnel junction is a magnetic moment antiparallel state of the magnetic free layer and the magnetic pinning layer, and corresponds to a relatively high output voltage;
a synapse signal pulse current input by the synapse array generates a spin orbit torque to act on the magnetic free layer, so that the magnetic torque of the magnetic free layer is turned over, and the magnetic tunnel junction is in an excited state; after the synaptic signal pulse current is finished, reducing current is introduced into the magnetic tunnel to generate spin transfer torque to act on the magnetic free layer, and the magnetic free layer is reduced to an initial state; after the magnetic free layer is reduced, the reduction current is cut off;
and the reading current port finishes the emission of one output voltage pulse in the process that the magnetic tunnel junction is overturned from the initial state to the excited state and then is restored to the initial state.
Wherein the synapse system further comprises a signal integration element connected between the magnetic tunnel junction and the array of synapses for integrating a signal of each synapse of the array of synapses into a total synapse current pulse connected to one end of the bottom electrode.
According to the magnetic tunnel junction and the corresponding nonlinear response device provided by the embodiment of the invention, the writing, reading and recovering of the nonlinear response device based on the tunnel junction are separated by utilizing the STT (Spin Transfer Torque) effect and the SOT (Spin Orbit Torque) effect, so that the circuit is simplified, the nonlinear response of the turnover probability and the pulse current is realized, and the nonlinear response of the tunnel junction is easier to utilize.
Drawings
FIG. 1 is a schematic diagram of a magnetic tunnel junction according to an embodiment of the present invention;
FIG. 2 is a schematic cross-sectional view taken along line A-A of the device of FIG. 1;
fig. 3 and 4 are schematic structural diagrams of a neuron nonlinear response device provided by an embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention are described in detail below with reference to the accompanying drawings. Examples of these preferred embodiments are illustrated in the accompanying drawings. The embodiments of the invention shown in the drawings and described in accordance with the drawings are exemplary only, and the invention is not limited to these embodiments.
It should be noted that, in order to avoid obscuring the present invention with unnecessary details, only the structures and/or processing steps closely related to the scheme according to the present invention are shown in the drawings, and other details not so relevant to the present invention are omitted.
Embodiments of the present invention first provide a magnetic tunnel junction, and referring to fig. 1 and 2, the magnetic tunnel junction 30 includes a top electrode 12 and a bottom electrode 11 and a tunnel junction 20 disposed between the top electrode 12 and the bottom electrode 11. The bottom electrode 11 is a heavy metal layer with two ports having spin hall angles, and the tunnel junction 20 includes a magnetic free layer 21, an insulating barrier layer 22, a magnetic pinning layer 23, an antiferromagnetic coupling spacer layer 24, an antiferromagnetic coupling magnetic layer 25, and an antiferromagnetic layer 26 sequentially disposed on the bottom electrode 11.
The bottom electrode layer 11 preferably has a thickness of 1nm to 10nm, and the material is preferably Pt or Ta.
Among them, the top electrode layer 12 is preferably 10nm to 100nm thick, and the material is preferably Pt, Ta, Cr, Au, or Ti.
Among them, the preferred thicknesses of the magnetic free layer 21, the magnetic pinned layer 23, and the antiferromagnetic coupling magnetic layer 25 are 1nm to 10nm, respectively, all have in-plane magnetic moments, and the easy axis is along the width direction (e.g., Y direction in fig. 1) of the heavy metal bottom electrode 11. Preferred materials for the magnetic free layer 21, the magnetic pinned layer 23, and the antiferromagnetically-coupled magnetic layer 25 are each independently selected from CoFeB, NiFe, CoFe, Co, or FePt. Wherein the magnetic moments of the magnetic pinning layer 23 and the anti-ferromagnetically coupled magnetic layer 25 are opposite in direction, and the magnetic pinning layer 23 is fixed by the anti-ferromagnetic layer 26 under the working condition of the device without significant change, the magnetic moment of the magnetic free layer 21 can be switched under the action of a magnetic field or current.
The thickness of the insulating barrier layer 22 is preferably 0.5nm to 5nm, and the material is preferably MgO or Al2O3Or TiO2. The insulating barrier layer 22 allows electrons to be in the magnetic free layer21 and the magnetic pinned layer 23.
Wherein the preferred thickness of the antiferromagnetically-coupled spacer layer 24 is 0.8nm to 0.9nm, more preferably 0.85nm, and the preferred material is Ru.
Among them, the preferred thickness of the antiferromagnetic layer 26 is 10nm to 50nm, the direction of antiferromagnetic order is the easy axis direction of the antiferromagnetic coupling magnetic layer 25, and the preferred material is IrMn, PtMn, FeMn, NiO, NiCoO or other combinations.
Further, referring to fig. 1, the area of the heavy metal layer bottom electrode 11 having the spin hall angle is larger than that of the other layers, and the shape is a long strip, and two electrode ports are led out from two ends of the long strip, and the width is preferably 500nm to 5 μm.
Wherein the respective layer structures of the top electrode 12 and the tunnel junction 20 have the same cross-sectional shape, and the projections of the top electrode 12 and the tunnel junction 20 in the plane of the bottom electrode 11 are completely located within the bottom electrode 11. Preferably, the cross-sectional shape is circular or elliptical; when the cross section is circular, the radius ranges from 50nm to 500 nm; when the cross section is in an oval shape, the value ranges of the major axis and the minor axis are 50 nm-500 nm.
Embodiments of the present invention also provide a neuron nonlinear response device, junction and fig. 1 to 4, comprising a synapse system 60 and a magnetic tunnel junction 30 as described above. The synapse system 60 comprises a synapse array 61 of the neuron and a synapse signal integration element 62, the synapse signal output by the synapse array 61 is connected to one end of a bottom electrode 11 of the magnetic tunnel junction 30, and the other end of the bottom electrode 11 is grounded. The top electrode 12 of the magnetic tunnel junction 30 is configured to connect a reduction current (I)th)40 and obtaining the output voltage (V)out) Read current (I)0)50, specifically, the top electrode 12 is connected to two inputs, one is a read current 50 and the other is a reduction current 40.
The preferred implementation of the synapse array 61 is a multi-level memristor. The signal integration element 62 is used to integrate the signal of each synapse into a total synaptic current pulse (I)SYN)63. It should be noted that if the synaptic transmission of the pulse signal is a current pulse, the magnitude of the current pulse and the magnetism of the current pulseThe working range of the tunnel junction 30 is better and the signal integration element 62 may be omitted. The preferred duration of the total synaptic current pulse is between 0.1ns and 10 ns.
Specifically, the initial state of the magnetic tunnel junction 30 is a state in which the magnetic moments of the magnetic free layer 21 and the magnetic pinned layer 23 are parallel, corresponding to a relatively low output voltage; the excited state of the magnetic tunnel junction 30 is the antiparallel state of the magnetic moments of the magnetic free layer 21 and the magnetic pinned layer 23, corresponding to a relatively high output voltage.
The read current 50 is a constant small current I0Outputs a voltage V according to the resistance state of the tunnel junction when it flows through the tunnel junction 20out. When synaptic current ISYNThe magnetic tunnel junction 30 is excited from the parallel state to the antiparallel state, and the output voltage VoutWill increase accordingly.
The reduction current 40 is a current that reduces the tunnel junction from an antiparallel state to a parallel state by the spin torque (STT) effect, which can ensure a certain magnetic moment flip of the magnetic free layer 21 without a current-dependent probability of flip. The direction of the reduction current 40 is from the ground to the reduction current port.
The reduction current 40 is only passed to the magnetic tunnel junction 30 when the synaptic current 63 is not passed and the tunnel junction 30 is in an antiparallel state, and the magnetic tunnel junction 30 stops being passed after switching from the antiparallel state to a parallel state. The switching state of the reduction current 40 may be controlled by the waveform generator or by the circuit.
The output voltage of the tunnel junction changes from a higher positive value to a negative value when the reduction current 40 flows in, and returns to a lower positive value when the reduction current 40 is turned off, corresponding to the emission of a voltage pulse. The probability of the emission of a voltage pulse depends on the magnitude of the total synaptic current 63.
In this embodiment, as shown in fig. 4, the turn-on and turn-off of the reduction current 40 can be controlled by an NMOS transistor, and the gate voltage V of the MOS transistorcontrolIs inputted by the control circuit.
After a fixed turn-on time of the reduction current 40, when the tunnel junction is in an antiparallel state and the total synaptic current 63 does not flow in, the control circuit inputs a high gate voltage to turn on the reduction current 40, and after the fixed turn-on time, the control circuit inputs a low gate voltage to turn off the reduction current 40, and the tunnel junction 30 returns to the initial state within the flow-in time. The preferred passage time of the reduction current 40 is 2ns to 10 ns.
The magnitude of the reducing current 40 can be controlled to be inversely related to the turn-on time, and the magnitude of the reducing current 40 at a certain turn-on time needs to be sufficient to cause the magnetic moment of the free layer 21 to be flipped deterministically. Another control strategy is to fix the magnitude of a certain reduction current 40 and control the circuit to turn off the reduction current 40 after the magnetic moments are reduced to a parallel state. The two strategies have corresponding influences on the complexity and the power consumption of the system and need to be selected according to the system.
Combining the above synaptic signal pulse current I input by the synaptic arraySYNA spin orbit torque is generated to act on the magnetic free layer 21, so that the magnetic moment of the magnetic free layer 21 is reversed, and the magnetic tunnel 30 junction is in an excited state; synaptic signal pulse current ISYNAfter the end, reducing the current IthIntroducing a magnetic tunnel junction to generate spin transfer torque to act on the magnetic free layer 21, and reducing the magnetic free layer 21 to an initial state; after the magnetic free layer 21 is reduced, the reduction current is turned off. Wherein, the process of the magnetic tunnel junction 30 being turned over from the initial state to the excited state and then being reduced to the initial state, the read current I0The port finishes the primary output voltage VoutAnd (4) transmitting the pulse.
The magnetic tunnel and the nonlinear response device based on the magnetic tunnel can be prepared by a micro-nano processing technology and an integrated circuit technology.
The embodiments of the present invention will be further described in conjunction with the principle of tunnel junction probability inversion.
The current flowing through the bottom electrode of the thin heavy metal layer with strong spin-orbit coupling can generate spin current perpendicular to the current due to the interface effect and the spin hall effect, and when the spin current acts on the magnetic free layer 21, the magnetic moment of the magnetic free layer can generate fixed precession related to the current magnitude. Meanwhile, at the temperature T of not absolute zero degree, the magnetic moment of the magnetic free layer 21 deviates from the equilibrium position by small random precession caused by thermal disturbance. The effect of random switching of the magnetic moment of the magnetic free layer 21 around the current threshold is achieved by the random precession generated by thermal disturbance, which is a determining factor, and the fixed precession generated by the spin current, wherein the probability of random switching under the conditions of the Neille temperature of the antiferromagnetic layer 26 and the fixed pulse time Deltat has a sigmoid function-like relation with the magnitude of the current flowing through the heavy metal layer. It can be approximated within the tolerance of the error as expressed by the following equation (1):
wherein w is a constant intrinsic to the free layer 21 and the heavy metal layer 11 and is also related to Δ T, T is temperature, typically 300K, ISYNI is the current magnitude at which the probability of flipping is 50% for the total synaptic current. w and I are both functions of the pulse time Δ t, as shown in equations (2) and (3) below.
w is in direct proportion to delta t, and I is in inverse proportion to delta t.
As can be seen from the junctions and equations (1), (2) and (3), the nonlinear response characteristic of the entire tunnel junction is directly determined by w and Δ t. In order to obtain a good non-linear response sigmoid curve, which is not too steep or too slow, a suitable Δ t needs to be selected on the basis of the optimization w. Preferably, the pulse ISYNThe duration of (A) is in the order of nanoseconds (ns) to optimize the probability of flip of the free layer magnetic moment and ISYNNon-linear relationship of magnitude.
Output aspect, output voltage V of tunnel junctionoutIs determined by the read current I0And (5) controlling.
In terms of power consumption, the power consumption of a nonlinear response device is determined by the synaptic current ISYNAnd reduction current IthThe control is carried out by controlling the temperature of the air conditioner,the overall power consumption can be reduced by optimizing w and the duration of the two currents, both of which are in the range of 10 muA to 5mA, so that the power consumption is much smaller than that of the conventional CMOS system.
In summary, the magnetic tunnel junction and the corresponding nonlinear response device provided by the embodiments of the present invention utilize the STT effect and the SOT effect to separate the writing, reading, and recovery of the nonlinear response device based on the tunnel junction, thereby simplifying the circuit, realizing the nonlinear response of the switching probability and the magnitude of the pulse current, and simultaneously making the nonlinear response of the tunnel junction easier to use.
The foregoing is directed to embodiments of the present application and it is noted that numerous modifications and adaptations may be made by those skilled in the art without departing from the principles of the present application and are intended to be within the scope of the present application.
Claims (10)
1. A magnetic tunnel junction is characterized by comprising a top electrode, a bottom electrode and a tunnel junction arranged between the top electrode and the bottom electrode, wherein the bottom electrode is a heavy metal layer with two ports having spin Hall angles, and the tunnel junction comprises a magnetic free layer, an insulating barrier layer, a magnetic pinning layer, an antiferromagnetic coupling spacer layer, an antiferromagnetic coupling magnetic layer and an antiferromagnetic layer which are sequentially arranged on the bottom electrode;
the magnetic free layer, the magnetic pinning layer and the anti-ferromagnetic coupling magnetic layer are all in-plane magnetic moments, the easy axis is along the width direction of the bottom electrode, the magnetic moment directions of the magnetic pinning layer and the anti-ferromagnetic coupling magnetic layer are opposite, and the direction of the anti-ferromagnetic sequence of the anti-ferromagnetic layer is the easy axis direction of the anti-ferromagnetic coupling magnetic layer.
2. The magnetic tunnel junction of claim 1 wherein the top electrode and the tunnel junction have the same cross-sectional shape and the projections of the top electrode and the tunnel junction in the plane of the bottom electrode are located entirely within the bottom electrode.
3. The magnetic tunnel junction of claim 2 wherein the cross-sectional shape is circular or elliptical; when the cross section is circular, the radius ranges from 50nm to 500 nm; when the cross section is in an oval shape, the value ranges of the major axis and the minor axis are 50 nm-500 nm.
4. The magnetic tunnel junction of claim 3 wherein the bottom electrode is an elongated structure having a width of 500nm to 5 μm.
5. The MTJ of any of claims 1-4, wherein the bottom electrode is made of Pt or Ta, the top electrode is made of Pt, Ta, Cr, Au or Ti, the magnetic free layer, the magnetic pinned layer and the antiferromagnetically-coupled magnetic layer are each independently selected from CoFeB, NiFe, CoFe, Co or FePt, and the insulating barrier layer is made of MgO, Al2O3Or TiO2The material of the antiferromagnetic coupling spacer layer is Ru, and the material of the antiferromagnetic layer is IrMn, PtMn, FeMn, NiO or NiCoO.
6. The magnetic tunnel junction of claim 5 wherein the bottom electrode is 1nm to 10nm thick, the insulating barrier layer is 0.5nm to 5nm thick, the antiferromagnetic coupling spacer layer is 0.8nm to 0.9nm thick, the antiferromagnetic layer is 10nm to 50nm thick, the magnetic free layer, the magnetic pinned layer and the antiferromagnetic coupling magnetic layer are 1nm to 10nm thick, respectively, and the top electrode is 10nm to 100nm thick.
7. A neuron nonlinear response device, comprising a synapse system and a magnetic tunnel junction according to any one of claims 1-6, wherein the synapse system comprises a synapse array, a synapse signal output by the synapse array is connected to one end of a bottom electrode, and the other end of the bottom electrode is grounded; the top electrode is configured to connect a reduction current and a read current to obtain an output voltage.
8. The nonlinear response device in neuron of claim 7, wherein the magnetic free layer is acted on by spin orbit torque generated by synaptic signal pulse current flowing through the bottom electrode, and the probability of magnetic moment flip of the magnetic free layer has a sigmoid function relation with the magnitude of the pulse current at a certain pulse length of the synaptic signal input current, which is specifically expressed by the following formula (1):
wherein w is a constant of intrinsic properties of the magnetic free layer and the heavy metal layer bottom electrode and is related to the pulse time Deltat, T is temperature, ISYNThe current is synaptic current, and I is the current magnitude when the turnover probability is 50%; w and I are both functions of the pulse time Δ t; specifically, the following formulae (2) and (3):
w＝w(Δt)；
I*＝I*(w)；
9. the neuron nonlinear response device of claim 8, wherein the initial state of the magnetic tunnel junction is a state in which magnetic moments of the magnetic free layer and the magnetic pinned layer are parallel, corresponding to a relatively low output voltage; the excited state of the magnetic tunnel junction is a magnetic moment antiparallel state of the magnetic free layer and the magnetic pinning layer, and corresponds to a relatively high output voltage;
a synapse signal pulse current input by the synapse array generates a spin orbit torque to act on the magnetic free layer, so that the magnetic torque of the magnetic free layer is turned over, and the magnetic tunnel junction is in an excited state; after the synaptic signal pulse current is finished, reducing current is introduced into the magnetic tunnel to generate spin transfer torque to act on the magnetic free layer, and the magnetic free layer is reduced to an initial state; after the magnetic free layer is reduced, the reduction current is cut off;
and the reading current port finishes the emission of one output voltage pulse in the process that the magnetic tunnel junction is overturned from the initial state to the excited state and then is restored to the initial state.
10. The neuronal nonlinear response device according to any of claims 7-9, wherein the synapse system further comprises a signal integration element connected between the magnetic tunnel junction and the synapse array, the signal integration element configured to integrate a signal of each synapse of the synapse array into a total synaptic current pulse connected to one end of the bottom electrode.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010262733.8A CN113496273B (en) | 2020-04-03 | 2020-04-03 | Magnetic tunnel junction and neuron nonlinear response device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010262733.8A CN113496273B (en) | 2020-04-03 | 2020-04-03 | Magnetic tunnel junction and neuron nonlinear response device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113496273A true CN113496273A (en) | 2021-10-12 |
| CN113496273B CN113496273B (en) | 2024-05-14 |
Family
ID=77995189
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010262733.8A Active CN113496273B (en) | 2020-04-03 | 2020-04-03 | Magnetic tunnel junction and neuron nonlinear response device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113496273B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140114893A1 (en) * | 2011-05-31 | 2014-04-24 | International Business Machines Corporation | Low-power event-driven neural computing architecture in neural networks |
| CN103917992A (en) * | 2011-11-09 | 2014-07-09 | 高通股份有限公司 | Method and apparatus for using memory in probabilistic manner to store synaptic weights of neural network |
| US20170330070A1 (en) * | 2016-02-28 | 2017-11-16 | Purdue Research Foundation | Spin orbit torque based electronic neuron |
| CN107871815A (en) * | 2016-09-28 | 2018-04-03 | 中国科学院苏州纳米技术与纳米仿生研究所 | Memristor based on MTJ |
| CN110095141A (en) * | 2018-01-31 | 2019-08-06 | 中国科学院苏州纳米技术与纳米仿生研究所 | Radar type spinning microwave detector and its preparation method and application based on magnetic tunnel junction |
| US20190272870A1 (en) * | 2018-03-02 | 2019-09-05 | Sandisk Technologies Llc | Probabilistic neuron circuits |
-
2020
- 2020-04-03 CN CN202010262733.8A patent/CN113496273B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140114893A1 (en) * | 2011-05-31 | 2014-04-24 | International Business Machines Corporation | Low-power event-driven neural computing architecture in neural networks |
| CN103917992A (en) * | 2011-11-09 | 2014-07-09 | 高通股份有限公司 | Method and apparatus for using memory in probabilistic manner to store synaptic weights of neural network |
| US20170330070A1 (en) * | 2016-02-28 | 2017-11-16 | Purdue Research Foundation | Spin orbit torque based electronic neuron |
| CN107871815A (en) * | 2016-09-28 | 2018-04-03 | 中国科学院苏州纳米技术与纳米仿生研究所 | Memristor based on MTJ |
| CN110095141A (en) * | 2018-01-31 | 2019-08-06 | 中国科学院苏州纳米技术与纳米仿生研究所 | Radar type spinning microwave detector and its preparation method and application based on magnetic tunnel junction |
| US20190272870A1 (en) * | 2018-03-02 | 2019-09-05 | Sandisk Technologies Llc | Probabilistic neuron circuits |
Non-Patent Citations (3)
| Title |
|---|
| SAMUEL N等: "A Probabilistic Synapse With Strained MTJs for Spiking Neural Networks", 《IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS ( VOLUME: 31, ISSUE: 4, APRIL 2020) 20190618》 * |
| 申见昕;尚大山;孙阳;: "基于磁电耦合效应的基本电路元件和非易失性存储器", 物理学报, no. 12 * |
| 蔡佳林: "基于自旋电子器件的神经形态器件研究", 《中国优秀博士学位论文全文数据库》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113496273B (en) | 2024-05-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8199565B2 (en) | Magnetic random access memory with dual spin torque reference layers | |
| US11790968B2 (en) | Spintronic device, SOT-MRAM storage cell, storage array and in-memory computing circuit | |
| US8023315B2 (en) | Magnetoresistive effect element and magnetic random access memory | |
| US8023316B2 (en) | Magnetic tunnel junction with compensation element | |
| US9577182B2 (en) | Magnetoresistance effect element and magnetic memory | |
| EP2320425B1 (en) | Selection device for a spin transfer torque magnetoresistive random access memory | |
| US7759750B2 (en) | Magnetic memory cell and random access memory | |
| US20050174837A1 (en) | Logic-in-memory circuit using magnetoresistive element | |
| US20090034326A1 (en) | Methods and apparatus for thermally assisted programming of a magnetic memory device | |
| Peng et al. | Magnetic tunnel junctions for spintronics: principles and applications | |
| Jaiswal et al. | Energy-efficient memory using magneto-electric switching of ferromagnets | |
| Everitt et al. | Size dependence of switching thresholds for pseudo spin valve MRAM cells | |
| US6873542B2 (en) | Antiferromagnetically coupled bi-layer sensor for magnetic random access memory | |
| US7358846B2 (en) | Magnetic spin valve with a magnetoelectric element | |
| KR20180045328A (en) | Spin Neuron Device and Method of operating the same | |
| CN113496273B (en) | Magnetic tunnel junction and neuron nonlinear response device | |
| Zheng et al. | Micromagnetic principles in pseudo spin valve memory element design | |
| US20230394291A1 (en) | Neuron device based on spin orbit torque | |
| CN112864314A (en) | Magneto-resistor device, method for changing resistance state of magneto-resistor device and synapse learning module | |
| WO2020215403A1 (en) | Multifunctional self-spinning electronic logic gate device | |
| WO2011099476A1 (en) | Ferromagnetic tunnel junction element | |
| WO2022155828A1 (en) | Three-state spintronic device, memory cell, memory array, and read-write circuit | |
| WO2022036623A1 (en) | Non-external magnetic field oriented spin flip sot-mram and array | |
| CN113497180A (en) | Spin memristor synapse device based on sigermin |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |