WO2014106351A1 - 一种模拟生物神经元和神经突触的单元、装置及方法 - Google Patents

一种模拟生物神经元和神经突触的单元、装置及方法 Download PDF

Info

Publication number
WO2014106351A1
WO2014106351A1 PCT/CN2013/070245 CN2013070245W WO2014106351A1 WO 2014106351 A1 WO2014106351 A1 WO 2014106351A1 CN 2013070245 W CN2013070245 W CN 2013070245W WO 2014106351 A1 WO2014106351 A1 WO 2014106351A1
Authority
WO
WIPO (PCT)
Prior art keywords
pulse signal
biological
material layer
functional material
resistance
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.)
Ceased
Application number
PCT/CN2013/070245
Other languages
English (en)
French (fr)
Inventor
缪向水
李祎
钟应鹏
许磊
孙华军
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huazhong University of Science and Technology
Original Assignee
Huazhong University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Huazhong University of Science and Technology filed Critical Huazhong University of Science and Technology
Publication of WO2014106351A1 publication Critical patent/WO2014106351A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/049Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/06Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/063Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
    • G06N3/065Analogue means
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • G06N3/088Non-supervised learning, e.g. competitive learning

Definitions

  • the invention belongs to the technical field of microelectronic devices, and more particularly to a unit, device and method for simulating biological neurons and synapses.
  • the human brain includes up to ⁇ 10 11 neurons and ⁇ 10 15 synapses, and the connections between neurons and synapses are more chaotic and incomparably complex.
  • This traditional neuromorphic approach is incapable of simulating the human brain, even the mouse brain.
  • IBM uses the "Blue Gene" supercomputer to use 147,456 processor-architecture neural networks to simulate cat cerebral cortical cognition.
  • neuron signal processing can be implemented in nanodevices, the chip size and power consumption of the devices required to simulate the entire brain can be achieved within the achievable range.
  • the Chinese invention name is: Information processing method and apparatus for simulating biological neuron information processing mechanism.
  • the device shown in the patent application file of CN101770560A is based on a CMOS integrated circuit, and a plurality of transistors constitute a neuron, and the integration density is not High, and does not involve synaptic function with learning ability.
  • the Chinese invention name is: a flexible triode that resembles a synaptic structure of a neuron, open
  • the device shown in the patent application file number CN1670963A mimics the structure of neuronal synapses and does not function as a neuron or synapse.
  • an object of the present invention to provide a unit for simulating neurons and synapses capable of simultaneously simulating the function of biological neurons and the function of biological synapses.
  • the present invention provides a unit for simulating biological neurons and synapses, comprising a first electrode layer, a functional material layer connected to the first electrode layer, and a first layer connected to the functional material layer a second electrode layer; the first electrode layer is used to simulate post-synaptic, the second electrode layer is used to simulate pre-synaptic, the material of the functional material layer is a sulfur-based compound, and the conductive material layer is used for conductance Simulating synaptic weights; simulating postsynaptic stimulation by applying a second pulse signal to the first electrode layer, and simulating presynaptic stimulation by applying a first pulse signal to the second electrode layer; The electrical resistance of the functional material layer is used to simulate the excited or resting state of the biological neuron.
  • the first electrode layer is configured to receive an external second pulse signal
  • the second electrode layer is configured to receive an external first pulse signal; when the amplitude of the first pulse signal is different from the first When the difference between the amplitudes of the two pulse signals is positive or negative, the conductance of the functional material layer is changed to realize the simulation of the synaptic weight adjustment function of the biological synapse; when the first pulse signal is When the peak of the signal difference between the second pulse signals is positive or negative, the conductance of the functional material layer is changed to realize the simulation of the pulse time-dependent synaptic plasticity function of the biological synapse; when the functional material layer The resistance change from high resistance state to low resistance state realizes the simulation of biological neuron threshold excitation function; when the resistance of the functional material layer changes from low resistance state to high resistance state, the biological neuron recovers from excited state to rest Simulation of state function; when the resistance of the functional material layer changes from a high resistance state to a low resistance state, an increase in the energy accumulation excitation function of the biological neuron is realized
  • the materials of the first electrode layer and the second electrode layer are both inert conductive metals.
  • first electrode layer, the functional material layer and the second electrode layer structure A sandwich laminate structure, a T-shaped structure, an I-shaped structure or a pyramid-shaped structure.
  • the present invention also provides a device for simulating biological neurons and synapses, comprising a plurality of arrayed synaptic units and a controller coupled to the synaptic unit, wherein the synaptic unit is Unit.
  • the controller is configured to apply a second pulse signal to the first electrode layer, a first pulse signal to the second electrode layer, and control the first pulse signal amplitude and the second
  • the difference between the amplitudes of the pulse signals is positive or negative, and the number of pulses is controlled by controlling the magnitudes of the amplitudes of the first pulse signal and the second pulse signal.
  • the invention also provides a method of simulating biological neurons and synapses, comprising the steps of:
  • Controlling the resistance of the functional material layer to a high resistance state to a low resistance state by controlling the first pulse signal and the second pulse signal and simulating a threshold excitation function of the biological neuron;
  • Controlling the resistance of the functional material layer from a low resistance state to a high resistance state by controlling the first pulse signal and the second pulse signal and simulating a function of recovering a biological neuron from an excited state to a resting state;
  • Varying the number of pulses required to change the resistance of the functional material layer from a high resistance state to a low resistance state by controlling the magnitude of the amplitude of the first pulse signal and the second pulse signal and simulating the energy of the biological neuron Cumulative excitation function;
  • the threshold excitation function of the simulated biological neuron is specifically: when the resistance value of the functional material layer is greater than the first resistance threshold, simulating the resting state function of the biological neuron; when the functional material layer When the resistance of the resistor is less than the second resistance threshold, the excited state function of the biological neuron is simulated.
  • the threshold excitation function step of the simulated biological neuron is specifically: controlling a difference between a magnitude of the first pulse signal and a magnitude of the second pulse signal to be greater than a first voltage threshold and Less than the second voltage threshold, causing the resistance of the functional material layer to change from a high resistance state to a low resistance state, simulating a biological neuron transition from a resting state to an excited state function; by controlling the amplitude of the first pulse signal And a difference between the amplitude of the second pulse signal and the second voltage threshold is such that the resistance of the functional material layer recovers from a low resistance state to a high resistance state, simulating the recovery of the biological neuron from the excited state to the static state Information function.
  • the energy accumulation excitation function step of the simulated biological neuron is specifically: controlling the difference between the amplitude of the first pulse signal and the amplitude of the second pulse signal to be smaller than the first voltage threshold And maintaining a resistance of the functional material layer in a high resistance state; and controlling the first pulse signal and the second pulse by reducing an amplitude of the first pulse signal and a magnitude of the second pulse signal The number of pulses of the signal is increased such that the resistance of the layer of functional material drops from the high resistance state to the second resistance threshold, simulating the energy accumulation excitation function of the biological neuron.
  • the step of synthesizing the synaptic weight adjustment function of the biological synapse is specifically: controlling the difference between the amplitude of the first pulse signal and the amplitude of the second pulse signal to be positive, Decreasing the conductance of the functional material layer, simulating the synaptic weight reduction function of the biological synapse;
  • the conductance of the functional material layer is increased, simulating the synaptic weight rise function of the biological synapse.
  • the step of synthesizing the synaptic weight adjustment function of the biological synapse further comprises: controlling between the amplitude of the first pulse signal and the amplitude of the second pulse signal The amplitude of the positive difference is increased such that the slower the conductance of the functional material layer is reduced, simulating the slower function of the synaptic weight of the biological synapse falling;
  • the pulse time-dependent synaptic plasticity function of the simulated biological synapse comprises:
  • the pulse time-dependent synaptic plasticity function of the simulated biological synapse comprises:
  • the pulse time of the simulated biological synapse is dependent on synaptic plasticity
  • the steps include:
  • the pulse time-dependent synaptic plasticity function of the simulated biological synapse comprises:
  • the unit for simulating biological neurons and synapses provided by the invention can well simulate the threshold excitation function of biological neurons, the function of restoring from an excited state to a resting state, and the function of energy accumulation excitation, and the protrusion of biological nerve synapses.
  • the functions of neurons and synapses are simultaneously realized in a single device, and basic components for low power consumption and small size are provided for constructing a neural network.
  • FIG. 1 is a schematic structural view of a device for simulating biological neurons and synapses according to an embodiment of the present invention
  • Figure 2 (a) is a schematic view showing the structure of a unit for simulating biological neurons and synapses provided in Example 1 of the present invention
  • FIG. 2(b) is a simulated biological neuron stimulated and restored resting according to Embodiment 1 of the present invention.
  • FIG. 2(c), FIG. 2(d) is a simulated biological neuron according to Embodiment 1 of the present invention subjected to multiple thresholds. The squat is stimulated by the test;
  • FIG. 2(e) is a simulated biological synaptic weight adjustment test provided by Embodiment 1 of the present invention
  • FIG. 2(f) and FIG. 2(g) are simulated biological synaptic odd symmetric type I provided by Embodiment 1 of the present invention
  • STDP functional test
  • 2(h) and 2(i) are simulation functional biosynaptic symmetry type II STDP functional tests provided by the first embodiment of the present invention
  • 2(1) and 2(m) are simulated biosynaptic symmetry type II STDP functional tests provided by the first embodiment of the present invention
  • Figure 3 (a) is a schematic view showing the structure of a unit simulating biological neurons and synapses provided in Example 2 of the present invention
  • Figure 3 (b) The simulated biological neuron provided by the embodiment 2 of the present invention is subjected to an excitation and recovery rest test;
  • FIG. 3 (c) Simulated biological synaptic weight adjustment test provided by Example 2 of the present invention
  • Figure 4 (a) The junction of simulated biological neurons and synaptic units provided in Example 3 of the present invention
  • Figure 4 (b) Simulated neuron stimulated and restored resting test provided in Example 3 of the present invention
  • Figure 4 (c) Simulated synaptic weight adjustment test provided in Example 3 of the present invention.
  • the present invention provides a nanodevice based on a sulfur-based compound, which can be applied to artificial neural network circuits or cognitive storage technologies as artificial neurons and artificial synapses.
  • the device acts as an artificial neuron, it can simulate the threshold excitation function of biological neurons; when it is artificial synaptic, it can simulate the synaptic plasticity adjustment function of biological nerve synapses.
  • the sulfur-based compound can reversibly transform between amorphous and crystalline states under the action of externally applied light pulses and electrical pulses.
  • the difference in reflectance and resistivity between the two states is used to store "0".
  • "and" 1" has been widely studied and applied to phase change optical discs and phase change random access memories, and has the advantages of non-volatile, fast operation speed, low power consumption, high reliability, and easy integration with COMS circuits.
  • the energy accumulation threshold phase transition characteristics of amorphous sulfur-based compounds can be used to simulate the threshold excitation characteristics of neurons.
  • the gradual change characteristics of sulfur-based compounds with electrical impulses can be used to simulate synapses. Weight adjustment.
  • the neural component of the present invention can mimic the function of biological neurons, including (1) the neuron can respond to a variety of different inputs; (2) for input exceeding a threshold (0.5 to 3 V), the neuron Can be excited; (3) For inputs that do not exceed the threshold, the neuron can be excited when multiple inputs that do not exceed the threshold accumulate exceed the threshold; (4) the neuron can recover from the excited state to the resting state. It also mimics the functions of biological synapses, including (1) adjusting the weight of synapses; (2) synaptic weights can be changed according to the time difference of the pre- and post-synaptic pulses, ie, the pulse-time-dependent plasticity (STDP) function.
  • STDP pulse-time-dependent plasticity
  • the unit for simulating biological neurons and synapses includes a first electrode a layer, a functional material layer connected to the first electrode layer, a second electrode layer connected to the functional material layer; the first electrode layer is used to simulate presynaptic, and the second electrode layer is used for simulation Post-synaptic, the conductance of the layer of functional material is used to simulate synaptic weights; pre-synaptic stimulation is simulated by applying a second pulse signal to the first electrode layer, by applying a first to the second electrode layer a pulse signal to simulate post-synaptic stimulation; the first electrode layer is for receiving an external second pulse signal, the second electrode layer is for receiving an external first pulse signal; and when the second pulse signal is When the difference between the amplitude and the amplitude of the first pulse signal is positive or negative, the conductance of the functional material layer is changed to achieve a simulation of the synaptic weight adjustment function of the biological synapse; When the peak of the signal difference between the second pulse signal
  • Simulation of (STDP) function simulation of threshold excitation function of biological neurons when the resistance of the functional material layer is high-resistance or low-resistance; when the resistance of the functional material layer changes from high-resistance to low
  • the resistance state or the recovery from the low-resistance state to the high-resistance state realizes the simulation of the recovery of the biological neuron from the excited state to the resting state function; when the resistance of the functional material layer changes from the high-resistance state to the low-resistance state, the conversion rate increases.
  • a simulation of the energy accumulation excitation function of biological neurons is achieved.
  • the materials of the first electrode layer and the second electrode layer are all inert conductive metals, such as platinum (PO, titanium tungsten (TiW), tantalum (Ta), etc.; the material of the functional material layer is sulfur-based a compound such as Ge 2 Sb 2 Te 5 , Sb 2 Te 3 , GeTe, BiTe, AglnSbTe, etc.
  • the unit composed of the first electrode layer, the functional material layer and the second electrode layer may be a sandwich laminate structure, a T-type structure, Type I structure or pyramid structure.
  • the neural component is a two-terminal resistance device, and should have at least one (high resistance state) resting state and one (low resistance state) excited state.
  • the current through the neural component rapidly increases and the neural component is excited.
  • the neural component is capable of maintaining a high electrical resistance at all times, in a resting state, until an electrical signal having a sufficiently large energy exceeding a threshold is applied, the neuron transitioning to an excited state.
  • the neuron is in a resting state, and when an electrical signal whose energy does not exceed a threshold is applied, the neuron remains In the resting state; when several such electrical signals are applied to the neuron, the energy accumulation reaches a threshold, and the neuron changes from a resting state to an excited state, and the resistance decreases remarkably.
  • the neuron is in a resting state.
  • a series of electrical signals having the same energy but not exceeding the threshold are applied, the smaller the energy of the single electrical signal, the electrical signal required for the neuron to transition from the resting state to the excited state. The more the number.
  • the neural component can realize a threshold excitation function of a biological neuron, and an energy accumulation excitation function.
  • the synapse device has a plurality of resistive states whose resistance value changes according to the direction of the current passing through it, the forward current causes its resistance to rise, and the reverse current causes its resistance to decrease. However, when the current is less than a certain threshold, its resistance does not change.
  • Four types of pre- and post-synaptic pulse signals can be designed to achieve four
  • the neurosynaptic device enables the weight adjustment function of the biological synapse and the STDP function.
  • the second electrode of the synaptic device is presynaptic and the first electrode is post-synaptic.
  • the pre-synaptic stimulation signal is presynaptic stimulation
  • the post-synaptic stimulation signal is post-synaptic stimulation.
  • is the time difference between the synaptic and stimuli.
  • the current synaptic stimuli precede the post-synaptic stimuli, ⁇ >0; the current synaptic stimuli are post-posterior, ⁇ 0.
  • Synaptic weight W l/R, R is the resistance of the synaptic device, ⁇ ⁇ is the amount of change in synaptic weight before and after the stimuli.
  • FIG. 1 shows the structure of a device for simulating biological neurons and synapses provided by an embodiment of the present invention, the device comprising a plurality of arrays of synaptic units and a controller connected to the synapse unit,
  • the synaptic unit is the above unit
  • the controller is configured to apply a second pulse signal to the first electrode layer, apply a first pulse signal to the second electrode layer, and control the second pulse signal amplitude and the first pulse signal amplitude
  • the difference between the values is positive or negative, and the peak of the signal difference between the second pulse signal and the first pulse signal is controlled to be positive or negative; and the number of pulses of the second pulse signal and the first pulse signal is controlled.
  • the neural component provided by the present invention includes a first electrode 101, The second electrode 103 and the sulfur-based compound material 102 between the first electrode 101 and the second electrode 103.
  • the first electrode 101 and the sulfur-based compound material 102, the sulfur-based compound material 102, and the second electrode 103 form electrical contact.
  • the first electrode 101 and the second electrode 103 are titanium tungsten (TiW), and the sulfur-based compound material 102 is germanium (Ge 2 Sb 2 Te 5 ).
  • Fig. 2(b) is a diagram showing the function of realizing the threshold excitation function of the neuron and the recovery from the excited state to the resting state according to the test of the present embodiment.
  • the input signal is applied to the second electrode 103, the first electrode 101 is grounded, and the resistor 202 is the resistance between the first electrode 101 and the second electrode 103.
  • the resistance of the neural component includes at least two states, the first being an amorphous high-resistance state, the resistance value being greater than the first resistance threshold (100 kQ), simulating the resting state of the neuron;
  • the crystal is in a low-resistance state, and its resistance value is smaller than the second resistance threshold (10kQ), simulating the excited state of the neuron.
  • the resistance When the input pulse signal exceeds the first voltage threshold (IV) and is lower than the second voltage threshold (2V), such as the pulse signal 201, the resistance changes from a high resistance state to a low resistance state; when the input pulse signal exceeds the second threshold, As with pulse signal 202, the resistance transitions from a low resistance state to a high resistance state.
  • the nerve component is stimulated by a threshold value, transitioning from a resting state (high-resistance state) to an excited state (low-resistance state), and recovering from an excited state (low-resistance state) to a resting state (high-resistance state), Achieve the function of the biological neuron threshold excitation, and restore the resting state.
  • Fig. 2 (c) and Fig. 2 (d) are diagrams showing another test according to the present embodiment, which realizes an energy accumulation excitation function of a neuron.
  • the input signal is applied to the second electrode 103, and the first electrode 101 is grounded, and the resistance is the resistance between the first electrode 101 and the second electrode 103.
  • the neuron component is originally in a resting state (high-impedance state), and its resistance value is greater than the first resistance threshold.
  • the resistance of the neural component does not change significantly and can be maintained in a high resistance state.
  • the resistance of the neuron device rapidly drops to a low resistance state that is less than the second resistance threshold, i.e., the simulated neurons are excited.
  • the number of pulses required to change the neural component from a high impedance state to a low impedance state increases. Achieve energy accumulation and excitation of biological neurons.
  • Figure 2 (e) is a diagram showing another test according to the present embodiment, achieving synaptic weights of synapses Adjustment function.
  • the input signal is applied to the second electrode 103, and the first electrode 101 is grounded, and the resistance is the resistance between the first electrode 101 and the second electrode 103.
  • the resistance representing the synaptic weight has a plurality of resistance values which can be changed with the input signal.
  • the input pulse signal is positive, the resistance increases; when the input pulse signal is negative, the resistance decreases.
  • the larger the amplitude of the positive pulse signal the larger the resistance; the smaller the negative pulse signal, the smaller the resistance.
  • the synaptic weight decreases, and the resistance decreases, that is, the synaptic weight increases. Achieve synaptic weight regulation of biological synapses.
  • FIG. 2 (0 and 2 (g) are diagrams showing another test according to the present embodiment, which implements a synaptic odd-symmetric type I STDP function.
  • the pre-synaptic stimulation signal is applied to the second electrode 103, synapse
  • the post-stimulation signal is applied to the first electrode 101
  • the pre- and post-synaptic signal difference is the signal difference between the second electrode 103 and the first electrode 101.
  • FIG. 2 when ⁇ >0, presynaptic excitation
  • the peak 301 of the post-synaptic stimuli is negative. Under the pre- and post-synaptic stimulation, the electrical resistance of the synaptic device decreases and the synaptic weight increases.
  • Fig. 2 (h) and Fig. 2 (i) are diagrams showing an odd symmetric type II STDP function for realizing synapses according to another test of the present embodiment.
  • the presynaptic stimulation signal is applied to the second electrode 103
  • the post-synaptic stimulation signal is applied to the first electrode 101.
  • the pre- and post-synaptic signal difference is the signal difference between the second electrode 103 and the first electrode 101.
  • ⁇ >0 the peak 303 of the difference between presynaptic and post-synaptic stimuli is positive, and the resistance of the synaptic device rises after the synaptic stimuli. Synaptic weights have dropped.
  • Fig. 2 (j) and Fig. 2 (k) are diagrams showing an even symmetric I-type STDP function for realizing synapses according to another test of the present embodiment.
  • the presynaptic stimuli signal is applied to the second electrode 103
  • the postsynaptic stimuli signal is applied to the first electrode 101.
  • the synaptic signal difference is the signal difference between the second electrode 103 and the first electrode 101.
  • Fig. 2 (1) and Fig. 2 (m) are diagrams showing an even symmetric type II STDP function for realizing synapses according to another test of the present embodiment.
  • the presynaptic stimulation signal is applied to the second electrode 103
  • the post-synaptic stimulation signal is applied to the first electrode 101.
  • the pre- and post-synaptic signal difference is the signal difference between the second electrode 103 and the first electrode 101.
  • the neural component of the present invention includes a first electrode 101 and a second electrode 103.
  • the insulating layer 104 and the first electrode 101 of the sulfur-based compound material 102 and the sulfur-based compound material 102, the sulfur-based compound material 102, and the second electrode 103 are electrically contacted between the first electrode 101 and the second electrode 103.
  • the first electrode 101 and the second electrode 103 are platinum (Pt)
  • the sulfur-based compound material 102 is germanium telluride (GeTe)
  • the insulating layer 104 is made of silicon dioxide (SiO 2 ).
  • the device structure is T-shaped.
  • Fig. 3(b) is a diagram showing the function of realizing the threshold excitation function of the neuron and the recovery from the excited state to the resting state according to the test of the embodiment 2.
  • the input signal is applied to the second electrode 103, the first electrode 101 is grounded, and the resistor 202 is the resistance between the first electrode 101 and the second electrode 103.
  • the resistance of the neural component includes at least two states, the first being an amorphous high-resistance state, the resistance value being greater than the first resistance threshold (110 kQ), simulating the resting state of the neuron;
  • the crystal is in a low-resistance state, and its resistance value is smaller than the second resistance threshold (l lkQ), simulating the excited state of the neuron.
  • the resistance changes from a high resistance state to a low resistance state
  • the second A voltage threshold such as pulse signal 502
  • the nerve component is stimulated by a threshold value, transitioning from a resting state (high-resistance state) to an excited state (low-resistance state), and recovering from an excited state (low-resistance state) to a resting state (high-resistance state), Achieve the function of the biological neuron threshold excitation, and restore the resting state.
  • the energy accumulation excitation function of the neurons can also be realized, and details are not described herein again.
  • Fig. 3(c) is a diagram showing another test according to Embodiment 2, which implements a synaptic weight adjustment function of a synapse.
  • the input signal is applied to the second electrode 103, and the first electrode 101 is grounded, and the resistance is the resistance between the first electrode 101 and the second electrode 103.
  • the resistance representing the synaptic weight has a plurality of resistance values which can be changed with the input signal.
  • the resistance increases; when the input pulse signal is negative, the resistance decreases.
  • the larger the amplitude of the positive pulse signal the resistance
  • the larger the negative pulse signal the smaller the resistance.
  • this embodiment can also implement the STDP function of the synapse, and details are not described herein again.
  • the neural component of the present invention includes a first electrode 101, a second The electrode 103, the insulating layer 104, and the first electrode 101 of the sulfur-based compound material 102 and the sulfur-based compound material 102, the sulfur-based compound material 102, and the second electrode 103 are electrically contacted between the first electrode 101 and the second electrode 103.
  • the first electrode 101 and the second electrode 103 are tantalum (Ta)
  • the sulfur-based compound material 102 is tantalum telluride (Sb 2 Te 3 )
  • the insulating layer 104 is made of silicon dioxide (SiO 2 ).
  • the device structure is type I.
  • Fig. 4(b) is a diagram showing the function of realizing the threshold excitation function of the neuron and the recovery from the excited state to the resting state according to the test of the embodiment 3.
  • the input signal is applied to the second electrode 103, the first electrode 101 is grounded, and the resistor 202 is the resistance between the first electrode 101 and the second electrode 103.
  • the resistance of the neural component includes at least two states, the first being an amorphous high resistance state, the resistance value being greater than the first resistance threshold (90 kQ), simulating the resting state of the neuron;
  • the crystal is in a low-resistance state, and its resistance value is smaller than the second resistance threshold (9kQ), simulating the excited state of the neuron.
  • the resistance changes from the high resistance state to the low resistance state; when the input pulse signal exceeds the second A voltage threshold, such as pulse signal 802, changes from a low resistance state to a high impedance state.
  • the nerve component is stimulated by a threshold value, transitioning from a resting state (high-resistance state) to an excited state (low-resistance state), and recovering from an excited state (low-resistance state) to a resting state (high-resistance state), Achieve the function of the biological neuron threshold excitation, and restore the resting state.
  • the energy accumulation excitation function of the neuron can also be realized in this embodiment, and details are not described herein again.
  • Fig. 4 (c) is a view showing another test according to Embodiment 3, which realizes a synaptic weight adjustment function of a synapse.
  • the input signal is applied to the second electrode 103, and the first electrode 101 is grounded, and the resistance is the resistance between the first electrode 101 and the second electrode 103.
  • the weighted resistor has multiple resistance values that can change with the input signal. When the input pulse signal is positive, the resistance increases; when the input pulse signal is negative, the resistance decreases. The larger the amplitude of the positive pulse signal, the larger the resistance; the smaller the negative pulse signal, the smaller the resistance.
  • the synaptic weight decreases, and the resistance decreases, that is, the synaptic weight increases. Achieve synaptic weight regulation of biological synapses.
  • this embodiment can also implement the STDP function of the synapse, and details are not described herein again.
  • the unit for simulating biological synapses provided by the embodiments of the present invention can also simulate other various STDP functions of the synaptic pulse-dependent synaptic plasticity function, which will not be described in detail herein.
  • the unit, the device and the method provided by the invention can well simulate the threshold excitation function of the biological neuron, the function of recovering from the excited state to the resting state and the energy accumulation excitation function, and the synaptic weight adjustment function of the biological synapse. And pulse time dependent synaptic plasticity function. In order to save space, it is no longer exhaustive.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Neurology (AREA)
  • Semiconductor Memories (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

本发明公开了一种基于硫系化合物的模拟生物神经元和神经突触的单元、装置及方法,该单元包括第一电极层、功能材料层和第二电极层。模拟神经元时,器件接受一个或多个电脉冲剌激,功能材料层的电阻从高阻态转变为低阻态,模拟神经元从一个静息态转变到激发态,实现阈值激发和能量累积激发功能。模拟神经突触时,器件功能材料层的电导能根据输入信号渐变,实现突触权重调节功能;以及根据两端输入信号的时间差改变其突触权重,实现脉冲时间依赖突触可塑性STDP功能。本发明能提供构成人工神经网络的基本元器件。

Description

一种模拟生,经元和神经突触的单元、 装置及^ 【技术领域】
本发明属于微电子器件技术领域, 更具体地, 涉及一种模拟生物神经 元和神经突触的单元、 装置及方法。
【背景技术】
基于冯诺依曼架构的传统计算机中, 处理器与存储器是分立的, 以总 线连接。这样的架构存在所谓的"冯诺依曼瓶颈",难以适应信息呈爆炸式增 长的信息技术飞速发展的当今时代。
相比于冯诺依曼计算机, 人脑神经信息活动具有大规模并行、 分布式 存储与处理、 自组织、 自适应和自学习的特征。 传统的人工神经网络、 神 经形态工程学等领域的研究人员也一直致力于利用非线性电路、 FPGA、 VLSI等手段来模拟神经元电触发、 突触可塑性等神经元突触的基本生物电 特性以及更高级的模式识别、 智能控制等认知功能。 在这些方法中, 仅模 拟一个神经元、 一个突触、 一个学习模块就需要数十个晶体管、 电容、 加 法器。 然而, 人的大脑中包括了多达〜 1011个神经元以及〜 1015个突触, 神经 元、 突触之间的连接更是混沌的、 无比复杂的。 这种传统的神经形态工程 办法对于模拟人类大脑, 即使是小鼠大脑都是无能为力的, IBM利用"蓝色 基因"超级计算机使用了 147456个处理器架构神经元网络来模拟猫的大脑 皮层认知功能。 如果能在纳米器件中实现神经元的信号处理, 那么模拟整 个大脑所需器件集成起来的芯片尺寸、 功耗才能在可实现范围之内。
构建神经网络涉及神经元和神经突触设计和制备。 中国发明名称为: 模拟生物神经元信息处理机制的信息处理方法及装置, 公开号为 CN101770560A的专利申请文件中所示的装置是基于 CMOS集成电路, 用 多个晶体管构成一个神经元, 集成密度不高, 而且不涉及具有学习能力的 神经突触功能。 中国发明名称为: 仿神经元突触结构的柔性三极管, 公开 号为 CN1670963A的专利申请文件中所示的装置模拟神经元突触的结构, 并没有实现神经元、 神经突触的功能。
【发明内容】
针对现有技术的缺陷, 本发明的目的在于提供能同时模拟生物神经元 的功能和生物神经突触的功能的一种模拟神经元和神经突触的单元。
为实现上述目的, 本发明提供了一种模拟生物神经元和神经突触的单 元, 包括第一电极层、 与所述第一电极层连接的功能材料层, 与所述功能 材料层连接的第二电极层; 所述第一电极层用于模拟突触后, 所述第二电 极层用于模拟突触前, 所述功能材料层的材料为硫系化合物, 所述功能材 料层的电导用于模拟突触权重; 通过给所述第一电极层施加第二脉冲信号 来模拟突触后剌激, 通过给所述第二电极层施加第一脉冲信号来模拟突触 前剌激; 所述功能材料层的电阻用于模拟生物神经元的激发态或静息态。
更进一步地, 所述第一电极层用于接收外部的第二脉冲信号, 所述第 二电极层用于接收外部的第一脉冲信号; 当所述第一脉冲信号的幅值与所 述第二脉冲信号的幅值之间的差值为正或负时, 所述功能材料层的电导发 生改变实现了生物神经突触的突触权重调节功能的模拟; 当所述第一脉冲 信号与所述第二脉冲信号之间的信号差峰值为正或负时, 所述功能材料层 的电导发生改变实现了生物神经突触的脉冲时间依赖突触可塑性功能的模 拟; 当所述功能材料层的电阻从高阻态转换为低阻态实现了生物神经元阈 值激发功能的模拟; 当所述功能材料层的电阻从低阻态转变为高阻态实现 了生物神经元从激发态恢复到静息态功能的模拟; 当所述功能材料层的电 阻从高阻态转变为低阻态的转变率增加实现了生物神经元的能量累积激发 功能的模拟。
更进一步地, 所述第一电极层和所述第二电极层的材料均为惰性导电 金属。
更进一步地, 所述第一电极层、 所述功能材料层和所述第二电极层构 成三明治叠层结构、 T型结构、 I型结构或金字塔型结构。
本发明还提供了一种模拟生物神经元和神经突触的装置, 包括多个阵 列排布的神经突触单元以及与所述神经突触单元连接的控制器, 所述神经 突触单元为上述的单元。
更进一步地, 所述控制器用于给所述第一电极层施加第二脉冲信号, 给所述第二电极层施加第一脉冲信号, 并控制所述第一脉冲信号幅值与所 述第二脉冲信号幅值之间的差值为正或负, 并通过控制所述第一脉冲信号 和所述第二脉冲信号的幅值大小来控制脉冲数目。
本发明还提供了一种模拟生物神经元和神经突触的方法, 包括下述步 骤:
在第一电极层上施加第二脉冲信号, 在第二电极层上施加第一脉冲信 号;
通过控制所述第一脉冲信号与所述第二脉冲信号使得所述功能材料层 的电阻为高阻态转换为低阻态并模拟了生物神经元的阈值激发功能;
通过控制所述第一脉冲信号与所述第二脉冲信号使得所述功能材料层 的电阻从低阻态恢复为高阻态并模拟了生物神经元从激发态恢复到静息态 的功能;
通过控制所述第一脉冲信号与所述第二脉冲信号的幅值大小来改变所 述功能材料层的电阻从高阻态转变为低阻态所需的脉冲数目并模拟了生物 神经元的能量累积激发功能;
通过控制所述第一脉冲信号的幅值与所述第二脉冲信号的幅值之间的 差值的正或负来调节所述功能材料层的电导的变化并模拟生物神经突触的 突触权重调节功能;
通过控制所述第一脉冲信号与所述第二脉冲信号之间的信号差峰值为 正或负来调节所述功能材料层的电导的变化并模拟生物神经突触的脉冲时 间依赖突触可塑性功能。 更进一步地, 所述模拟生物神经元的阈值激发功能具体为: 当所述功 能材料层的电阻的阻值大于第一电阻阈值模拟了生物神经元的静息态功 能; 当所述功能材料层的电阻的阻值小于第二电阻阈值时, 模拟了生物神 经元的激发态功能。
更进一步地, 所述模拟生物神经元的阈值激发功能步骤具体为: 通过控制所述第一脉冲信号的幅值与所述第二脉冲信号的幅值之间的 差值大于第一电压阈值且小于第二电压阈值, 使得所述功能材料层的电阻 从高阻态转变为低阻态, 模拟了生物神经元从静息态转变成激发态功能; 通过控制所述第一脉冲信号的幅值与所述第二脉冲信号的幅值之间的 差值大于第二电压阈值, 使得所述功能材料层的电阻从低阻态恢复为高阻 态, 模拟了生物神经元从激发态恢复到静息态功能。
更进一步地, 所述模拟生物神经元的能量累积激发功能步骤具体为: 通过控制所述第一脉冲信号的幅值与所述第二脉冲信号的幅值之间的 差值小于第一电压阈值, 使得所述功能材料层的电阻保持高阻态; 并通过 减小所述第一脉冲信号的幅值与所述第二脉冲信号的幅值控制所述第一脉 冲信号与所述第二脉冲信号的脉冲数目增加, 使得所述功能材料层的电阻 从所述高阻态下降至所述第二电阻阈值, 模拟了生物神经元的能量累积激 发功能。
更进一步地, 所述模拟生物神经突触的突触权重调节功能步骤具体为: 通过控制所述第一脉冲信号的幅值与所述第二脉冲信号的幅值之间的 差值为正, 使得所述功能材料层的电导减小, 模拟了生物神经突触的突触 权重下降功能;
通过控制所述第一脉冲信号与所述第二脉冲信号之间的差值为负, 使 得所述功能材料层的电导增大, 模拟了生物神经突触的突触权重上升功能。
更进一步地, 所述模拟生物神经突触的突触权重调节功能步骤还包括: 通过控制所述第一脉冲信号的幅值与所述第二脉冲信号的幅值之间的 正差值的幅值增强, 使得所述功能材料层的电导减小得越慢, 模拟了生物 神经突触的突触权重下降得越慢的功能;
通过控制所述第一脉冲信号与所述第二脉冲信号之间的负差值的幅值 增强, 使得所述功能材料层的电导的增大得越快, 模拟了生物神经突触的 突触权重上升得越快的功能。
更进一步地, 所述模拟生物神经突触的脉冲时间依赖突触可塑性功能 步骤包括:
控制所述第一脉冲信号与所述第二脉冲信号的时间差大于零并调整所 述第一脉冲信号和所述第二脉冲信号的形状, 使得所述第一脉冲信号与所 述第二脉冲信号之间的信号差峰值为负, 所述功能材料层的电导增大, 模 拟了生物神经突触的突触权重增大的功能;
控制所述第一脉冲信号与所述第二脉冲信号的时间差小于零并调整所 述第一脉冲信号和所述第二脉冲信号的形状, 使得所述第一脉冲信号与所 述第二脉冲信号之间的信号差峰值为正, 所述功能材料层的电导减小, 模 拟了生物神经突触的突触权重较小的功能。
更进一步地, 所述模拟生物神经突触的脉冲时间依赖突触可塑性功能 步骤包括:
控制所述第一脉冲信号与所述第二脉冲信号的时间差大于零并调整所 述第一脉冲信号和所述第二脉冲信号的形状, 使得所述第一脉冲信号与所 述第二脉冲信号之间的信号差峰值为正, 所述功能材料层的电导减小, 模 拟了生物神经突触的突触权重减小的功能;
控制所述第一脉冲信号与所述第二脉冲信号的时间差小于零并调整所 述第一脉冲信号和所述第二脉冲信号的形状, 使得所述第一脉冲信号与所 述第二脉冲信号之间的信号差峰值为负, 所述功能材料层的电导增大, 模 拟了生物神经突触的突触权重增大的功能。
更进一步地, 所述模拟生物神经突触的脉冲时间依赖突触可塑性功能 步骤包括:
控制所述第一脉冲信号与所述第二脉冲信号的时间差的绝对值小于所 述第二脉冲信号宽度的四分之一并调整所述第一脉冲信号和所述第二脉冲 信号的形状, 使得所述第一脉冲信号与所述第二脉冲信号之间的信号差峰 值为负, 所述功能材料层的电导增大, 模拟了生物神经突触的突触权重增 大的功能;
控制所述第一脉冲信号与所述第二脉冲信号的时间差的绝对值大于等 于所述第二脉冲信号宽度的四分之一并调整所述第一脉冲信号和所述第二 脉冲信号的形状, 使得所述第一脉冲信号与所述第二脉冲信号之间的信号 差峰值为正, 所述功能材料层的电导减小, 模拟了生物神经突触的突触权 重减小的功能。
更进一步地, 所述模拟生物神经突触的脉冲时间依赖突触可塑性功能 步骤包括:
控制所述第一脉冲信号与所述第二脉冲信号的时间差的绝对值小于所 述第二脉冲信号宽度的二分之一并调整所述第一脉冲信号和所述第二脉冲 信号的形状, 使得所述第一脉冲信号与所述第二脉冲信号之间的信号差峰 值大于所述第一脉冲信号的峰值, 所述功能材料层的电导减小, 模拟了生 物神经突触的突触权重减小的功能;
控制所述第一脉冲信号与所述第二脉冲信号的时间差的绝对值大于等 于所述第二脉冲信号宽度的二分之一并调整所述第一脉冲信号和所述第二 脉冲信号的形状, 使得所述第一脉冲信号与所述第二脉冲信号之间的信号 差峰值为小于等于所述第一脉冲信号的峰值, 所述功能材料层的电导不变, 模拟了生物神经突触的突触权重不变的功能。
本发明提供的模拟生物神经元和神经突触的单元能很好的模拟生物神 经元的阈值激发功能、 从激发态恢复到静息态的功能和能量累积激发功能, 以及生物神经突触的突触权重调节功能和脉冲时间依赖突触可塑性功能; 且在单一器件内同时实现神经元和神经突触的功能, 为构建神经网络提供 功耗低、 尺寸小的基本元件。
【附图说明】
图 1 是本发明实施例提供的模拟生物神经元和神经突触的装置的结构 示意图;
图 2 (a) 是本发明实施例 1提供的模拟生物神经元和神经突触的单元 的结构示意图;
图 2 (b) 是本发明实施例 1提供的模拟生物神经元受激发和恢复静息 图 2 (c) 、 图 2 (d) 是本发明实施例 1提供的模拟生物神经元受多个 阈值下剌激受激发测试;
图 2 (e) 是本发明实施例 1提供的模拟生物神经突触权重调节测试; 图 2 (f) 、 图 2 (g) 是本发明实施例 1提供的模拟生物神经突触奇对 称 I型 STDP功能测试;
图 2 (h) 、 图 2 (i) 是本发明实施例 1提供的模拟生物神经突触奇对 称 II型 STDP功能测试;
图 2 (j ) 、 图 2 (k) 是本发明实施例 1提供的模拟生物神经突触偶对 称 I STDP功能测试;
图 2 (1) 、 图 2 (m) 是本发明实施例 1提供的模拟生物神经突触偶对 称 II型 STDP功能测试;
图 3 (a) 本发明实施例 2提供的模拟生物神经元和神经突触的单元的 结构示意图;
图 3 (b) 本发明实施例 2提供的模拟生物神经元受激发和恢复静息测 试;
图 3 (c) 本发明实施例 2提供的模拟生物神经突触权重调节测试; 图 4 (a) 本发明实施例 3提供的模拟生物神经元和神经突触单元的结 构示意图;
图 4 (b ) 本发明实施例 3提供的模拟神经元受激发和恢复静息测试; 图 4 ( c ) 本发明实施例 3提供的模拟神经突触权重调节测试。
【具体实肺式】
为了使本发明的目的、 技术方案及优点更加清楚明白, 以下结合附图 及实施例, 对本发明进行进一步详细说明。 应当理解, 此处所描述的具体 实施例仅仅用以解释本发明, 并不用于限定本发明。
本发明提供了一种基于硫系化合物的纳米器件, 能够作为人工神经元 以及人工神经突触应用于人工神经网络电路或认知存储技术。 该器件作为 人工神经元作用时, 能模拟实现生物神经元的阈值激发功能; 作为人工神 经突触作用时, 能模拟实现生物神经突触的突触可塑性调节功能。
硫系化合物作为一种存储介质, 能够在外界施加的光脉冲、 电脉冲作 用下, 在非晶态与晶态之间可逆转变, 两态间的反射率差异和电阻率差异 用于存储" 0"和 "1",已被广泛成熟地研究于应用于相变光盘和相变随机存储 器, 具有非挥发性、 操作速度快、 功耗小、 可靠性高和易于与 COMS电路 集成等优点。 非晶态的硫系化合物所具有的能量累积阈值相变特性可以用 来模拟神经元的阈值激发特性类似; 硫系化合物随电脉冲剌激的阻值渐变 特性又可以用来模拟神经突触的权重调节。
本发明的神经元器件能模拟生物神经元的功能, 包括 (1)该神经元能对 多种不同的输入作出相应的响应; (2)对于超过阈值 (0.5〜3V) 的输入, 该 神经元能受激发; (3)对于不超过阈值的输入, 当多个不超过阈值的输入累 积超过阈值时, 该神经元能受激发; (4)该神经元能从激发态恢复到静息态。 同时还能模拟生物神经突触的功能, 包括 (1)能调节突触的权重; (2)突触权 重可根据突触前后脉冲的时间差改变,即实现脉冲时间依赖的可塑性 (STDP) 功能。
本发明实施例提供的模拟生物神经元和神经突触的单元包括第一电极 层、 与所述第一电极层连接的功能材料层, 与所述功能材料层连接的第二 电极层; 所述第一电极层用于模拟突触前, 所述第二电极层用于模拟突触 后, 所述功能材料层的电导用于模拟突触权重; 通过给所述第一电极层施 加第二脉冲信号来模拟突触前剌激, 通过给所述第二电极层施加第一脉冲 信号来模拟突触后剌激; 所述第一电极层用于接收外部的第二脉冲信号, 所述第二电极层用于接收外部的第一脉冲信号; 当所述第二脉冲信号的幅 值与所述第一脉冲信号的幅值之间的差值为正或负时, 所述功能材料层的 电导发生改变实现了生物神经突触的突触权重调节功能的模拟; 当所述第 二脉冲信号与所述第一脉冲信号之间的信号差峰值为正或负时, 所述功能 材料层的电导发生改变实现了生物神经突触的脉冲时间依赖突触可塑性
( STDP) 功能的模拟; 当所述功能材料层的电阻为高阻态或低阻态实现了 生物神经元的阈值激发功能的模拟; 当所述功能材料层的电阻从高阻态转 变为低阻态或从低阻态恢复为高阻态实现了生物神经元从激发态恢复到静 息态功能的模拟; 当所述功能材料层的电阻从高阻态转变为低阻态的转变 率增加实现了生物神经元的能量累积激发功能的模拟。
作为本发明的一个实施例, 第一电极层和第二电极层的材料均为惰性 导电金属, 如铂 (PO、 钛钨 (TiW) 和钽 (Ta) 等; 功能材料层的材料为 硫系化合物, 如 Ge2Sb2Te5、 Sb2Te3、 GeTe、 BiTe和 AglnSbTe等。 由第一 电极层、 功能材料层和第二电极层构成的单元可以为三明治叠层结构、 T 型结构、 I型结构或金字塔型结构。
在本发明实施例中, 该神经元器件为一两端电阻器件, 应该至少具有 一个 (高电阻态) 静息态和一个 (低电阻态) 激发态。 当该神经元从静息 态转变为激发态时, 通过该神经元器件的电流迅速增大, 该神经元器件被 激发。 该神经元器件能一直保持高电阻, 处于静息态, 直到被施加一个能 量足够大、 超过阈值的电信号输入, 该神经元转变到激发态。 该神经元处 于静息态, 当被施加一个能量不超过阈值的电信号时, 该神经元仍然保持 在静息态; 当若干个如此的电信号作用于该神经元后, 能量积累达到阈值, 该神经元从静息态转变为激发态, 电阻明显下降。 该神经元处于在静息态, 当被施加一系列能量相同, 但均不超过阈值的电信号时, 单个电信号能量 越小, 该神经元从静息态转变到激发态所需的电信号个数越多。 该神经元 器件能实现生物神经元的阈值激发功能, 和能量累积激发功能。 该神经突 触器件具有多个阻态, 其电阻值根据通过它的电流的方向而改变, 正向电 流使其电阻上升, 反向电流使其电阻下降。 但当电流小于一定阈值时, 其 电阻不发生变化。 可以通过设计突触前后脉冲信号, 实现四种
STDP(spike-timing dependent plasticity剌激时间依赖突触可塑性)功能。该神 经突触器件能实现生物神经突触的权重调节功能和 STDP功能。
在本发明实施例中, 神经突触器件的第二电极是突触前, 第一电极是 突触后。 施加在突触前的剌激信号为突触前剌激, 施加在突触后的剌激信 号为突触后剌激。 Δΐ为前后突触剌激的时间差, 当前突触剌激先于后突触 剌激, Δΐ>0; 当前突触剌激后于后突出, ΔΚ0。 突触权重 W=l/R, R为神 经突触器件的电阻, Δ ¥为剌激作用前后, 突触权重的改变量。
图 1示出了本发明实施例提供的模拟生物神经元和神经突触的装置的 结构, 该装置包括多个阵列排布的神经突触单元以及与所述神经突触单元 连接的控制器, 神经突触单元为上述的单元, 控制器用于给第一电极层施 加第二脉冲信号, 给第二电极层施加第一脉冲信号, 并控制第二脉冲信号 幅值与所述第一脉冲信号幅值之间的差值为正或负, 控制第二脉冲信号与 所述第一脉冲信号之间的信号差峰值为正或负; 并控制第二脉冲信号和第 一脉冲信号的脉冲数目。
为了更进一步地说明本发明实施例提供的模拟生物神经元和神经突触 的单元, 现以具体实例并结合附图详述如下:
图 2 (a) 示出了本发明实施例 1提供的模拟生物神经元和神经突触的 单元的结构; 参考图 2 (a) , 本发明提供的神经元器件包括第一电极 101、 第二电极 103和第一电极 101和第二电极 103之间硫系化合物材料 102。第 一电极 101和硫系化合物材料 102、硫系化合物材料 102和第二电极 103形 成电接触。 其中第一电极 101和第二电极 103为钛钨 (TiW), 硫系化合物 材料 102为锗锑碲 (Ge2Sb2Te5)。
图 2 (b) 是示出根据本实施例的测试, 实现神经元的阈值激发功能, 以及从激发态恢复到静息态的功能。其中输入信号施加在第二电极 103上, 第一电极 101接地, 电阻 202为第一电极 101与第二电极 103之间的电阻。 参考图 2 (b), 神经元器件的电阻至少包括两个态, 第一个是非晶态高阻 态, 其电阻值大于第一电阻阈值 (100kQ), 模拟神经元的静息态; 第一个 是晶态低阻态,其电阻值小于第二电阻阈值(10kQ),模拟神经元的激发态。 当输入的脉冲信号超过第一电压阈值 (IV) 而低于第二电压阈值 (2V), 如脉冲信号 201, 电阻从高阻态转变成低阻态; 当输入的脉冲信号超过第二 阈值, 如脉冲信号 202, 电阻从低阻态转变成高阻态。 神经元器件受到超过 阈值的剌激, 从静息态 (高阻态) 转变成激发态 (低阻态), 并且能从激发 态 (低阻态) 恢复到静息态 (高阻态), 实现生物神经元阈值激发功能, 和 恢复静息态的功能。
图 2 (c)和图 2 (d) 是示出根据本实施例的另一测试, 实现神经元的 能量积累激发功能。 其中输入信号施加在第二电极 103 上, 第一电极 101 接地, 电阻为第一电极 101与第二电极 103之间的电阻。 参考图 2 (c), 神 经元器件原本处于静息态 (高阻态), 其电阻值大于第一电阻阈值。 当输入 的脉冲信号小于第一电压阈值时, 神经元器件的电阻不会明显变化, 能维 持在高阻态。 当小于第一电压阈值的脉冲信号达到一定数目后, 神经元器 件的电阻迅速下降至小于第二电阻阈值的低阻态, 即能模拟神经元被激发。 参考图 2 (d), 当输入的脉冲信号进一步减小时, 使神经元器件从高阻态转 变成低阻态所需的脉冲数目增加。 实现生物神经元的能量累积激发功能。
图 2 (e) 是示出根据本实施例的另一测试, 实现神经突触的突触权重 调节功能。 其中输入信号施加在第二电极 103上, 第一电极 101接地, 电 阻为第一电极 101与第二电极 103之间的电阻。 参考图 2 (e), 表示神经突 触权重的电阻有多个电阻值, 能随输入信号而改变。 当输入脉冲信号为正, 电阻增大; 当输入脉冲信号为负, 电阻减小。 正脉冲信号幅值越大, 电阻 越大; 负脉冲信号越小, 电阻越小。 电阻增大即突触权重减小, 电阻减小 即突触权重增大。 实现生物神经突触的突触权重调节功能。
图 2 (0 和图 2 (g) 是示出根据本实施例的另一测试, 实现神经突触 奇对称 I型 STDP功能。 其中突触前剌激信号施加在第二电极 103上, 突 触后剌激信号施加在第一电极 101上, 突触前后信号差为第二电极 103与 第一电极 101之间的信号差。 参考图 2 (0, 当 Δΐ>0, 突触前剌激与突触后 剌激之差的峰值 301 为负, 在此组突触前后剌激作用下, 神经突触器件的 电阻下降, 突触权重上升。 当 ΔΚ0, 突触前剌激与突触后剌激之差的峰值 302为正, 在此组突触前后剌激作用下, 神经突触器件的电阻上升, 突触权 重下降。 参考图图 2 (g), 在突触前后剌激的共同作用下, 当 Δΐ>0时, 器 件电阻下降, 突触权重上升, AW>0, 并且 AW随 Δΐ指数衰减; 当 ΔΚ0时, 器件电阻上升, 突触权重下降, Δ ¥<0, 并且 AW随 Δΐ指数衰减。 实现生 物神经突触奇对称 I型 STDP功能。
图 2 (h) 和图 2 (i) 是示出根据本实施例的另一测试, 实现神经突触 的奇对称 II型 STDP功能。 其中突触前剌激信号施加在第二电极 103上, 突触后剌激信号施加在第一电极 101 上, 突触前后信号差为第二电极 103 与第一电极 101之间的信号差。 参考图 2 (h), 当 Δΐ>0, 突触前剌激与突 触后剌激之差的峰值 303 为正, 在此组突触前后剌激作用下, 神经突触器 件的电阻上升, 突触权重下降。 当 ΔΚ0, 突触前剌激与突触后剌激之差的 峰值 304为负, 在此组突触前后剌激作用下, 神经突触器件的电阻下降, 突触权重上升。 参考图图 2 (i), 在突触前后剌激的共同作用下, 当 Δΐ>0 时, 器件电阻上升, 突触权重下降, Δ ¥<0, 并且 AW随 Δΐ指数衰减; 当 ΔΚ0时, 器件电阻下降, 突触权重上升, Δ ¥>0, 并且 AW随 Δΐ指数衰减。 实现生物神经突触的奇对称 II型 STDP功能。
图 2 (j ) 和图 2 (k) 是示出根据本实施例的另一测试, 实现神经突触 的偶对称 I型 STDP功能。 其中突触前剌激信号施加在第二电极 103上, 突触后剌激信号施加在第一电极 101 上, 突触前后信号差为第二电极 103 与第一电极 101之间的信号差。 参考图 2 (j ), 当 Δΐ较大 (Δΐ的绝对值大 于突触后剌激脉冲宽度的一半), 突触前剌激与突触后剌激之差的峰值 305 为正, 在此组突触前后剌激作用下, 神经突触器件的电阻上升, 突触权重 下降。 当 Δΐ较小 (Δΐ的绝对值小于等于突触后剌激脉冲宽度的一半), 突 触前剌激与突触后剌激之差的峰值 306为正, 在此组突触前后剌激作用下, 神经突触器件的电阻下降, 突触权重上升。 参考图图 2 (k), 在突触前后剌 激的共同作用下, 当 Δΐ较大时时, 器件电阻上升, 突触权重下降, AW<0; 当 Δΐ较小时, 器件电阻下降, 突触权重上升, Δ ¥>0。 并且 AW随 Δΐ呈垂 直平移正态分布。 实现生物神经突触偶对称 I型 STDP功能。
图 2 (1)和图 2 (m)是示出根据本实施例的另一测试, 实现神经突触 的偶对称 II型 STDP功能。 其中突触前剌激信号施加在第二电极 103上, 突触后剌激信号施加在第一电极 101 上, 突触前后信号差为第二电极 103 与第一电极 101之间的信号差。 参考图 2 (1), 当 Δΐ较大 (Δΐ的绝对值大 于突触后剌激脉冲宽度的四分之一), 突触前剌激与突触后剌激之差的峰值 307小于突触前剌激脉冲峰值的 1.5倍, 在此组突触前后剌激作用下, 神经 突触器件的电阻不变, 突触权重不变。 当 Δΐ较小 (Δΐ的绝对值小于等于突 触后剌激脉冲宽度的四分之一), 突触前剌激与突触后剌激之差的峰值 308 大于突触前剌激脉冲峰值的 1.5倍, 在此组突触前后剌激作用下, 神经突触 器件的电阻上升, 突触权重下降。 参考图图 2 (m), 在突触前后剌激的共 同作用下, 当 Δΐ较大时时, 器件电阻不变, 突触权重不变; 当 Δΐ较小时, 器件电阻上升, 突触权重下降, Δ ¥<0。 并且 AW随 Δΐ呈负向正态分布。 实现生物神经突触偶对称 II型 STDP功能。
图 3 (a) 示出了本发明实施例 2提供的模拟生物神经元和神经突触的 单元的结构; 参考图 3 (a), 本发明的神经元器件包括第一电极 101、 第二 电极 103、绝缘层 104和第一电极 101和第二电极 103之间硫系化合物材料 102第一电极 101和硫系化合物材料 102、硫系化合物材料 102和第二电极 103形成电接触。 其中第一电极 101和第二电极 103为铂 (Pt), 硫系化合 物材料 102为碲化锗 (GeTe), 绝缘层 104材料为二氧化硅 (Si02)。 器件 结构呈 T型。
图 3 (b) 是示出根据实施例 2的测试, 实现神经元的阈值激发功能, 以及从激发态恢复到静息态的功能。其中输入信号施加在第二电极 103上, 第一电极 101接地, 电阻 202为第一电极 101与第二电极 103之间的电阻。 参考图 3 (b),神经元器件的电阻至少包括两个态,第一个是非晶态高阻态, 其电阻值大于第一电阻阈值 (110kQ), 模拟神经元的静息态; 第一个是晶 态低阻态, 其电阻值小于第二电阻阈值(l lkQ), 模拟神经元的激发态。 当 输入的脉冲信号超过第一电压阈值 (1.2V) 而低于第二电压阈值 (2.5V), 如脉冲信号 501, 电阻从高阻态转变成低阻态; 当输入的脉冲信号超过第二 电压阈值, 如脉冲信号 502, 电阻从低阻态转变成高阻态。 神经元器件受到 超过阈值的剌激, 从静息态 (高阻态) 转变成激发态 (低阻态), 并且能从 激发态(低阻态)恢复到静息态(高阻态), 实现生物神经元阈值激发功能, 和恢复静息态的功能。 同理, 本实施例 2也能实现神经元的能量积累激发 功能, 在此不再赘述。
图 3 (c) 是示出根据实施例 2的另一测试, 实现神经突触的突触权重 调节功能。 其中输入信号施加在第二电极 103上, 第一电极 101接地, 电 阻为第一电极 101与第二电极 103之间的电阻。 参考图 3 (c), 表示神经突 触权重的电阻有多个电阻值, 能随输入信号而改变。 当输入脉冲信号为正, 电阻增大; 当输入脉冲信号为负, 电阻减小。 正脉冲信号幅值越大, 电阻 越大; 负脉冲信号越小, 电阻越小。 电阻增大即突触权重减小, 电阻减小 即突触权重增大。 实现生物神经突触的突触权重调节功能。 同理, 本实施 例也能实现神经突触的 STDP功能, 在此不再赘述。
图 4 (a) 示出了本发明实施例 3提供的模拟生物神经元和神经突触的 的单元的结构; 参考图 4 (a) , 本发明的神经元器件包括第一电极 101、 第 二电极 103、绝缘层 104和第一电极 101和第二电极 103之间硫系化合物材 料 102第一电极 101和硫系化合物材料 102、硫系化合物材料 102和第二电 极 103形成电接触。 其中第一电极 101和第二电极 103为钽 (Ta), 硫系化 合物材料 102为碲化锑 (Sb2Te3), 绝缘层 104材料为二氧化硅 (Si02)。 器 件结构呈 I型。
图 4 (b) 是示出根据实施例 3的测试, 实现神经元的阈值激发功能, 以及从激发态恢复到静息态的功能。其中输入信号施加在第二电极 103上, 第一电极 101接地, 电阻 202为第一电极 101与第二电极 103之间的电阻。 参考图 4 (b),神经元器件的电阻至少包括两个态,第一个是非晶态高阻态, 其电阻值大于第一电阻阈值(90kQ), 模拟神经元的静息态; 第一个是晶态 低阻态, 其电阻值小于第二电阻阈值 (9kQ), 模拟神经元的激发态。 当输 入的脉冲信号超过第一电压阈值 (0.8V) 而低于第二电压阈值 (1.5V), 如 脉冲信号 801, 电阻从高阻态转变成低阻态; 当输入的脉冲信号超过第二电 压阈值, 如脉冲信号 802, 电阻从低阻态转变成高阻态。神经元器件受到超 过阈值的剌激, 从静息态 (高阻态) 转变成激发态 (低阻态), 并且能从激 发态 (低阻态) 恢复到静息态 (高阻态), 实现生物神经元阈值激发功能, 和恢复静息态的功能。 同理, 本实施例也能实现神经元的能量积累激发功 能, 在此不再赘述。
图 4 (c) 是示出根据实施例 3的另一测试, 实现神经突触的突触权重 调节功能。 其中输入信号施加在第二电极 103上, 第一电极 101接地, 电 阻为第一电极 101与第二电极 103之间的电阻。 参考图 4 (c), 表示神经突 触权重的电阻有多个电阻值, 能随输入信号而改变。 当输入脉冲信号为正, 电阻增大; 当输入脉冲信号为负, 电阻减小。 正脉冲信号幅值越大, 电阻 越大; 负脉冲信号越小, 电阻越小。 电阻增大即突触权重减小, 电阻减小 即突触权重增大。 实现生物神经突触的突触权重调节功能。 同理, 本实施 例也能实现神经突触的 STDP功能, 在此不再赘述。
本发明实施例提供的模拟生物神经突触的单元还可以模拟神经突触的 脉冲时间依赖突触可塑性功能的其它多种 STDP功能, 在此不再一一详述。
本发明提供的单元、 装置和方法均能很好的模拟生物神经元的阈值激 发功能、 从激发态恢复到静息态的功能和能量累积激发功能, 以及生物神 经突触的突触权重调节功能和脉冲时间依赖突触可塑性功能。 为了节省篇 幅, 在此不再 穷举。
本领域的技术人员容易理解, 以上所述仅为本发明的较佳实施例而已, 并不用以限制本发明, 凡在本发明的精神和原则之内所作的任何修改、 等 同替换和改进等, 均应包含在本发明的保护范围之内。

Claims

权 利 要 求
1、 一种模拟生物神经元和神经突触的单元, 其特征在于, 包括第一电 极层、 与所述第一电极层连接的功能材料层, 与所述功能材料层连接的第 二电极层;
所述第一电极层用于模拟突触后, 所述第二电极层用于模拟突触前, 所述功能材料层的材料为硫系化合物, 所述功能材料层的电导用于模拟突 触权重; 通过给所述第一电极层施加第二脉冲信号来模拟突触后剌激, 通 过给所述第二电极层施加第一脉冲信号来模拟突触前剌激; 所述功能材料 层的电阻用于模拟生物神经元的激发态或静息态。
2、 如权利要求 1所述的单元, 其特征在于, 所述第一电极层用于接收 外部的第二脉冲信号, 所述第二电极层用于接收外部的第一脉冲信号; 当 所述第一脉冲信号的幅值与所述第二脉冲信号的幅值之间的差值为正或负 时, 所述功能材料层的电导发生改变实现了生物神经突触的突触权重调节 功能的模拟; 当所述第一脉冲信号与所述第二脉冲信号之间的信号差峰值 为正或负时, 所述功能材料层的电导发生改变实现了生物神经突触的脉冲 时间依赖突触可塑性功能的模拟; 当所述功能材料层的电阻从高阻态转换 为低阻态实现了生物神经元阈值激发功能的模拟; 当所述功能材料层的电 阻从低阻态转变为高阻态实现了生物神经元从激发态恢复到静息态功能的 模拟; 当所述功能材料层的电阻从高阻态转变为低阻态的转变率增加实现 了生物神经元的能量累积激发功能的模拟。
3、 如权利要求 1所述的单元, 其特征在于, 所述第一电极层和所述第 二电极层的材料均为惰性导电金属。
4、 如权利要求 1所述的单元, 其特征在于, 所述第一电极层、 所述功 能材料层和所述第二电极层构成三明治叠层结构、 T型结构、 I型结构或金 字塔型结构。
5、 一种模拟生物神经元和神经突触的装置, 包括多个阵列排布的模拟 生物神经元和神经突触的单元以及与其连接的控制器, 其特征在于, 所述 模拟生物神经元和神经突触的单元为权利要求 1-4任一项所述的单元。
6、 如权利要求 5所述的装置, 其特征在于, 所述控制器用于给所述第 一电极层施加第二脉冲信号, 给所述第二电极层施加第一脉冲信号, 并控 制所述第一脉冲信号幅值与所述第二脉冲信号幅值之间的差值为正或负, 并通过控制所述第一脉冲信号和所述第二脉冲信号的幅值大小来控制脉冲 数目。
7、 一种模拟生物神经元和神经突触的方法, 其特征在于, 包括下述步 骤:
在第一电极层上施加第二脉冲信号, 在第二电极层上施加第一脉冲信 号;
通过控制所述第一脉冲信号与所述第二脉冲信号使得所述功能材料层 的电阻为高阻态转换为低阻态并模拟了生物神经元的阈值激发功能;
通过控制所述第一脉冲信号与所述第二脉冲信号使得所述功能材料层 的电阻从低阻态恢复为高阻态并模拟了生物神经元从激发态恢复到静息态 的功能;
通过控制所述第一脉冲信号与所述第二脉冲信号的幅值大小来改变所 述功能材料层的电阻从高阻态转变为低阻态所需的脉冲数目并模拟了生物 神经元的能量累积激发功能;
通过控制所述第一脉冲信号的幅值与所述第二脉冲信号的幅值之间的 差值的正或负来调节所述功能材料层的电导的变化并模拟生物神经突触的 突触权重调节功能;
通过控制所述第一脉冲信号与所述第二脉冲信号之间的信号差峰值为 正或负来调节所述功能材料层的电导的变化并模拟生物神经突触的脉冲时 间依赖突触可塑性功能。
8、 如权利要求 7所述的方法, 其特征在于, 所述模拟生物神经元的阈 值激发功能具体为: 当所述功能材料层的电阻的阻值大于第一电阻阈值模 拟了生物神经元的静息态功能; 当所述功能材料层的电阻的阻值小于第二 电阻阈值时, 模拟了生物神经元的激发态功能。
9、 如权利要求 7或 8所述的方法, 其特征在于, 所述模拟生物神经元 的阈值激发功能步骤具体为:
通过控制所述第一脉冲信号的幅值与所述第二脉冲信号的幅值之间的 差值大于第一电压阈值且小于第二电压阈值, 使得所述功能材料层的电阻 从高阻态转变为低阻态, 模拟了生物神经元从静息态转变成激发态功能; 通过控制所述第一脉冲信号的幅值与所述第二脉冲信号的幅值之间的 差值大于第二电压阈值, 使得所述功能材料层的电阻从低阻态恢复为高阻 态, 模拟了生物神经元从激发态恢复到静息态功能。
10、 如权利要求 7所述的方法, 其特征在于, 所述模拟生物神经元的 能量累积激发功能步骤具体为:
通过控制所述第一脉冲信号的幅值与所述第二脉冲信号的幅值之间的 差值小于第一电压阈值, 使得所述功能材料层的电阻保持高阻态; 并通过 减小所述第一脉冲信号的幅值与所述第二脉冲信号的幅值控制所述第一脉 冲信号与所述第二脉冲信号的脉冲数目增加, 使得所述功能材料层的电阻 从所述高阻态下降至所述第二电阻阈值, 模拟了生物神经元的能量累积激 发功能。
11、 如权利要求 7所述的方法, 其特征在于, 所述模拟生物神经突触 的突触权重调节功能步骤具体为:
通过控制所述第一脉冲信号的幅值与所述第二脉冲信号的幅值之间的 差值为正, 使得所述功能材料层的电导减小, 模拟了生物神经突触的突触 权重下降功能;
通过控制所述第一脉冲信号与所述第二脉冲信号之间的差值为负, 使 得所述功能材料层的电导增大, 模拟了生物神经突触的突触权重上升功能。
12、 如权利要求 11所述的方法, 其特征在于, 所述模拟生物神经突触 的突触权重调节功能步骤还包括:
通过控制所述第一脉冲信号的幅值与所述第二脉冲信号的幅值之间的 正差值的幅值增强, 使得所述功能材料层的电导减小得越慢, 模拟了生物 神经突触的突触权重下降得越慢的功能;
通过控制所述第一脉冲信号与所述第二脉冲信号之间的负差值的幅值 增强, 使得所述功能材料层的电导的增大得越快, 模拟了生物神经突触的 突触权重上升得越快的功能。
13、 如权利要求 7所述的方法, 其特征在于, 所述模拟生物神经突触 的脉冲时间依赖突触可塑性功能步骤包括:
控制所述第一脉冲信号与所述第二脉冲信号的时间差大于零并调整所 述第一脉冲信号和所述第二脉冲信号的形状, 使得所述第一脉冲信号与所 述第二脉冲信号之间的信号差峰值为负, 所述功能材料层的电导增大, 模 拟了生物神经突触的突触权重增大的功能;
控制所述第一脉冲信号与所述第二脉冲信号的时间差小于零并调整所 述第一脉冲信号和所述第二脉冲信号的形状, 使得所述第一脉冲信号与所 述第二脉冲信号之间的信号差峰值为正, 所述功能材料层的电导减小, 模 拟了生物神经突触的突触权重较小的功能。
14、 如权利要求 7所述的方法, 其特征在于, 所述模拟生物神经突触 的脉冲时间依赖突触可塑性功能步骤包括:
控制所述第一脉冲信号与所述第二脉冲信号的时间差大于零并调整所 述第一脉冲信号和所述第二脉冲信号的形状, 使得所述第一脉冲信号与所 述第二脉冲信号之间的信号差峰值为正, 所述功能材料层的电导减小, 模 拟了生物神经突触的突触权重减小的功能;
控制所述第一脉冲信号与所述第二脉冲信号的时间差小于零并调整所 述第一脉冲信号和所述第二脉冲信号的形状, 使得所述第一脉冲信号与所 述第二脉冲信号之间的信号差峰值为负, 所述功能材料层的电导增大, 模 拟了生物神经突触的突触权重增大的功能。
15、 如权利要求 7所述的方法, 其特征在于, 所述模拟生物神经突触 的脉冲时间依赖突触可塑性功能步骤包括:
控制所述第一脉冲信号与所述第二脉冲信号的时间差的绝对值小于所 述第二脉冲信号宽度的四分之一并调整所述第一脉冲信号和所述第二脉冲 信号的形状, 使得所述第一脉冲信号与所述第二脉冲信号之间的信号差峰 值为负, 所述功能材料层的电导增大, 模拟了生物神经突触的突触权重增 大的功能;
控制所述第一脉冲信号与所述第二脉冲信号的时间差的绝对值大于等 于所述第二脉冲信号宽度的四分之一并调整所述第一脉冲信号和所述第二 脉冲信号的形状, 使得所述第一脉冲信号与所述第二脉冲信号之间的信号 差峰值为正, 所述功能材料层的电导减小, 模拟了生物神经突触的突触权 重减小的功能。
16、 如权利要求 7所述的方法, 其特征在于, 所述模拟生物神经突触 的脉冲时间依赖突触可塑性功能步骤包括:
控制所述第一脉冲信号与所述第二脉冲信号的时间差的绝对值小于所 述第二脉冲信号宽度的二分之一并调整所述第一脉冲信号和所述第二脉冲 信号的形状, 使得所述第一脉冲信号与所述第二脉冲信号之间的信号差峰 值大于所述第一脉冲信号的峰值, 所述功能材料层的电导减小, 模拟了生 物神经突触的突触权重减小的功能;
控制所述第一脉冲信号与所述第二脉冲信号的时间差的绝对值大于等 于所述第二脉冲信号宽度的二分之一并调整所述第一脉冲信号和所述第二 脉冲信号的形状, 使得所述第一脉冲信号与所述第二脉冲信号之间的信号 差峰值为小于等于所述第一脉冲信号的峰值, 所述功能材料层的电导不变, 拟了生物神经突触的突触权重不变的功能 (
PCT/CN2013/070245 2013-01-04 2013-01-09 一种模拟生物神经元和神经突触的单元、装置及方法 Ceased WO2014106351A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201310001461.6A CN103078054B (zh) 2013-01-04 2013-01-04 一种模拟生物神经元和神经突触的单元、装置及方法
CN201310001461.6 2013-01-04

Publications (1)

Publication Number Publication Date
WO2014106351A1 true WO2014106351A1 (zh) 2014-07-10

Family

ID=48154522

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2013/070245 Ceased WO2014106351A1 (zh) 2013-01-04 2013-01-09 一种模拟生物神经元和神经突触的单元、装置及方法

Country Status (2)

Country Link
CN (1) CN103078054B (zh)
WO (1) WO2014106351A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106137193A (zh) * 2016-07-29 2016-11-23 普罗朗生物技术(无锡)有限公司 人体生物电导多值模拟器
CN110033084A (zh) * 2018-01-11 2019-07-19 华邦电子股份有限公司 突触系统与突触方法
US11270194B2 (en) 2017-07-26 2022-03-08 International Business Machines Corporation System and method for constructing synaptic weights for artificial neural networks from signed analog conductance-pairs of varying significance
CN115585901A (zh) * 2021-07-05 2023-01-10 中国科学院微电子研究所 柔性多模态融合人工感受电路、系统、器件及其制备方法

Families Citing this family (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103078055B (zh) * 2013-01-04 2015-06-03 华中科技大学 一种模拟生物神经突触的单元、装置及方法
US9418331B2 (en) * 2013-10-28 2016-08-16 Qualcomm Incorporated Methods and apparatus for tagging classes using supervised learning
CN103530690A (zh) * 2013-10-31 2014-01-22 中国科学院上海微系统与信息技术研究所 一种神经元器件及神经网络
CN104899640B (zh) * 2014-07-21 2019-09-10 徐志强 神经网络的模拟装置及方法
CN104518088A (zh) * 2014-11-25 2015-04-15 中国科学院宁波材料技术与工程研究所 一种生物神经突触仿生电子器件的制备方法及其产品
CN104888346B (zh) * 2014-12-21 2020-10-13 徐志强 对昏迷大脑进行神经刺激的方法及装置
US10217046B2 (en) * 2015-06-29 2019-02-26 International Business Machines Corporation Neuromorphic processing devices
CN105046325B (zh) * 2015-07-06 2017-12-15 电子科技大学 一种基于类mos发光器件模拟生物神经网络的电路
CN107122828B (zh) * 2017-05-09 2020-05-05 清华大学 电路结构及其驱动方法、神经网络
TWI625681B (zh) * 2017-05-11 2018-06-01 國立交通大學 神經網路處理系統
CN107909146B (zh) * 2017-11-13 2021-09-17 中国科学院微电子研究所 基于易失性阈值转变器件的神经元电路
CN108549932B (zh) * 2018-03-22 2021-09-14 清华大学 一种基于液态金属的类脑拟神经器件
CN108985447B (zh) * 2018-06-15 2020-10-16 华中科技大学 一种硬件脉冲神经网络系统
CN109325596B (zh) * 2018-09-17 2024-03-26 中国传媒大学 一种基于钙浓度的神经突触可塑性计算方法
CN109460819B (zh) * 2018-10-25 2022-04-29 清华大学 一种用于模拟生物体光突触的方法及器件
KR102251765B1 (ko) * 2018-12-19 2021-05-14 한양대학교 산학협력단 뉴런 및 이를 포함하는 뉴로모픽 시스템
US11347999B2 (en) * 2019-05-22 2022-05-31 International Business Machines Corporation Closed loop programming of phase-change memory
TWI709090B (zh) 2019-08-30 2020-11-01 阿證科技股份有限公司 類神經人工智慧決策網路核心系統及其資訊處理方法
CN111630529B (zh) * 2019-11-15 2023-09-01 北京时代全芯存储技术股份有限公司 类神经电路以及运作方法
CN113517391B (zh) * 2020-09-24 2024-02-02 中国科学院微电子研究所 全忆阻器基的神经形态计算芯片及制备方法
CN113241369B (zh) * 2021-04-29 2022-05-06 齐鲁工业大学 一种支持孤波传导的模拟神经纤维的pn微米线
US20230056865A1 (en) * 2021-08-19 2023-02-23 Samsung Electronics Co., Ltd. Method and apparatus with artificial network generation and/or implementation corresponding to a natural neural network
CN113723602B (zh) * 2021-09-01 2024-11-15 哈尔滨工程大学 一种光纤结构的神经突触器件
CN116952409A (zh) * 2023-07-26 2023-10-27 华中科技大学 一种基于硫系阈值转变器件的温度感知人工神经元系统
CN117613100A (zh) * 2024-01-19 2024-02-27 西交利物浦大学 基于金属有机框架的具有滤波特性的神经元及其制备方法

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6141241A (en) * 1998-06-23 2000-10-31 Energy Conversion Devices, Inc. Universal memory element with systems employing same and apparatus and method for reading, writing and programming same
US20040153426A1 (en) * 2002-03-12 2004-08-05 Alex Nugent Physical neural network liquid state machine utilizing nanotechnology
CN102185101A (zh) * 2010-01-19 2011-09-14 索尼公司 存储元件和存储装置

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4248187B2 (ja) * 2002-03-27 2009-04-02 シャープ株式会社 集積回路装置及びニューロ素子
US7958071B2 (en) * 2007-04-19 2011-06-07 Hewlett-Packard Development Company, L.P. Computational nodes and computational-node networks that include dynamical-nanodevice connections
US9092736B2 (en) * 2010-07-07 2015-07-28 Qualcomm Incorporated Communication and synapse training method and hardware for biologically inspired networks
US9269042B2 (en) * 2010-09-30 2016-02-23 International Business Machines Corporation Producing spike-timing dependent plasticity in a neuromorphic network utilizing phase change synaptic devices
CN102456157B (zh) * 2010-10-20 2015-08-26 北京大学 神经元器件和神经网络
CN102496385B (zh) * 2011-12-26 2014-04-16 电子科技大学 一种脉冲时序活动性转换电路

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6141241A (en) * 1998-06-23 2000-10-31 Energy Conversion Devices, Inc. Universal memory element with systems employing same and apparatus and method for reading, writing and programming same
US20040153426A1 (en) * 2002-03-12 2004-08-05 Alex Nugent Physical neural network liquid state machine utilizing nanotechnology
CN102185101A (zh) * 2010-01-19 2011-09-14 索尼公司 存储元件和存储装置

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106137193A (zh) * 2016-07-29 2016-11-23 普罗朗生物技术(无锡)有限公司 人体生物电导多值模拟器
US11270194B2 (en) 2017-07-26 2022-03-08 International Business Machines Corporation System and method for constructing synaptic weights for artificial neural networks from signed analog conductance-pairs of varying significance
CN110033084A (zh) * 2018-01-11 2019-07-19 华邦电子股份有限公司 突触系统与突触方法
CN110033084B (zh) * 2018-01-11 2023-04-07 华邦电子股份有限公司 突触系统与突触方法
CN115585901A (zh) * 2021-07-05 2023-01-10 中国科学院微电子研究所 柔性多模态融合人工感受电路、系统、器件及其制备方法

Also Published As

Publication number Publication date
CN103078054B (zh) 2015-06-03
CN103078054A (zh) 2013-05-01

Similar Documents

Publication Publication Date Title
WO2014106351A1 (zh) 一种模拟生物神经元和神经突触的单元、装置及方法
Suri et al. Phase change memory as synapse for ultra-dense neuromorphic systems: Application to complex visual pattern extraction
Jackson et al. Nanoscale electronic synapses using phase change devices
CN103078055B (zh) 一种模拟生物神经突触的单元、装置及方法
CN106981567B (zh) 一种基于光电耦合忆阻器的人工突触器件及其调制方法
Hasan et al. Enabling back propagation training of memristor crossbar neuromorphic processors
CN102456157B (zh) 神经元器件和神经网络
Li et al. Associative learning with temporal contiguity in a memristive circuit for large‐scale neuromorphic networks
US11017292B2 (en) Artificial neuron apparatus
An et al. Realizing behavior level associative memory learning through three-dimensional memristor-based neuromorphic circuits
Nandakumar et al. Building brain-inspired computing systems: Examining the role of nanoscale devices
CN107122828B (zh) 电路结构及其驱动方法、神经网络
Cobley et al. A self-resetting spiking phase-change neuron
Zarudnyi et al. Spike-timing dependent plasticity in unipolar silicon oxide RRAM devices
Boybat et al. Stochastic weight updates in phase-change memory-based synapses and their influence on artificial neural networks
CN103530690A (zh) 一种神经元器件及神经网络
Chen et al. Optimal tuning of memristor conductance variation in spiking neural networks for online unsupervised learning
Mahalanabis et al. Demonstration of spike timing dependent plasticity in CBRAM devices with silicon neurons
Zayer et al. Modeling framework and comparison of memristive devices and associated STDP learning windows for neuromorphic applications
Suri et al. Addition of HfO2 interface layer for improved synaptic performance of phase change memory (PCM) devices
Yu Orientation classification by a winner-take-all network with oxide RRAM based synaptic devices
Garbin et al. Probabilistic neuromorphic system using binary phase-change memory (PCM) synapses: Detailed power consumption analysis
Jiang et al. Circuit design of RRAM-based neuromorphic hardware systems for classification and modified Hebbian learning
Elhamdaoui et al. Synapse design based on memristor
Wu et al. Enabling bio-plausible multi-level stdp using cmos neurons with dendrites and bistable rrams

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 13869923

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 13869923

Country of ref document: EP

Kind code of ref document: A1