Notice of Pre-AIA or AIA Status
This Non-Final communication is in response to Application No. 17/869,940 filed 07/21/2022. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Amendment
The amendments filed 12/15/2025 which provides amendments to claims 10, 19, 46-47 and canceled claim 21. Claims 1-20 and 22-48 are pending. The amendment to the claims has overcome the 101 rejection.
Response to Arguments
Applicant’s arguments with respect to 35 U.S.C § 102 and 103 filed 12/15/2025 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-18, 44-45 and 47 rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 1, 44, and 47 state “controlling another firing time difference between the AP of the first biological neuron and the PSP of the second biological neuron to be the adjusted firing time difference”. It is unclear how one would control the firing time of biological neurons. Claims 2-18, and 45, which are dependent one claims 1 and 44 respectively, are also rejected for the same reason.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-17, 19-20, 22-31, and 33-48 are rejected under 35 U.S.C. 103 as being unpatentable over Serb (NPL: “Memristive synapses connect brain and silicon spiking neurons” (2020)) in view of Saxena (NPL: “Towards Neuromorphic Learning Machines Using Emerging Memory Devices with Brain-Like Energy Efficiency” (2018))
Regarding claim 1, Serb teaches:
performing an adjusting, based on a firing time difference between an action potential (AP) of a first biological neuron and a post-synaptic potential (PSP) of a second biological neuron, of a first conductance value of a first memory corresponding to the first and second biological neurons; (Page 1 2nd paragraph “The memristor MR1 stores synaptic weights as resistive states. The thin film capacitive microelectrode13 CME delivers stimuli to the biological neuron (BN) that are adjusted by the memristive weights (Fig. 1b). Thus, in analogy with a native synapse, ABsyn operates by injecting in the BN an excitatory current, which reflects a plasticity-dependent synaptic strength. To emulate plasticity, the memristor MR1 is operated as a two-terminal device through a control system that receives pre- and post-synaptic depolarisations from one silicon neuron (ANpre) and one biological neuron (BN), respectively.”).
adjusting the firing time difference based on the PSP of the second biological neuron; (Figure 2. “Firing frequency is modulated in four phases, targeting the induction of plasticity as per the sequence: none/LTP/none/LTD, using the chosen plasticity rule. (b) BN firing response to ABsyn inputs. After LTP induction, the original 10 Hz pacemaker stimulation becomes capable of eliciting BN action potentials, thus reflecting the increase of postsynaptic potential amplitudes to above threshold. Firing persists until the commencement of the depotentiation/depression phase. (c) MR2 weight evolution. Data points denote resistance values for the intended LTP (red), LTD (blue) or no polarity change (black) phases. The right vertical axis indicates the corresponding weight. X-axis common to all panels.”)
Serb does not teach:
A method with artificial neural network generation, the method comprising:
and performing another adjusting, by controlling another firing time difference between the AP of the first biological neuron and the PSP of the second biological neuron to be the adjusted firing time difference, of a second conductance value of the first memory corresponding to the first and second biological neurons.
However Saxena does:
A method with artificial neural network generation, the method comprising: ((Abstract) “In this article, we review the challenges involved and present a pathway to realize large-scale mixed-signal NeuSoCs, from device arrays and circuits to spike-based deep learning algorithms with ‘brain-like’ energy-efficiency”)
and performing another adjusting, by controlling another firing time difference between the AP of the first biological neuron and the PSP of the second biological neuron to be the adjusted firing time difference, of a second conductance value of the first memory corresponding to the first and second biological neurons. (Section 3.3 second paragraph: “During the training phase, i.e., when the signal T = 1, the voltage spikes with positive pulse and negative tail are propagated in the forward (pre spikes) as well as the backward direction (post spikes). This enables learning by adjusting the synaptic weights using STDP based program or erase mechanism seen in Figure 2.” Training implies an iterative method meaning the adjusting happens multiple times.)
Serb and Saxena are considered analogous art to the claimed invention because they are in the same field of endeavor being spiking neural networks. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the biological neuron and silicon spiking neuron connect of Serb with the spiking neural network method of Saxena. One would want to do this to better replicate the natural neural networks of biology.
Regarding claim 2, Saxena in view of Serb teaches claim 1 as outlined above. Saxena further teaches:
resetting a conductance of the first memory after the adjusting of the first conductance value and before the other adjusting of the second conductance value. (They describe erase as resetting in 3.5 (4) Polarity: "Most RRAMs are employed with bipolar switching where program (Set) and erase (Reset) operations require positive and negative voltage polarity to be applied across the device.")
Regarding claim 3, Saxena in view of Serb teaches claim 2 as outlined above. Saxena further teaches:
the second conductance value is same as the first conductance value upon the resetting of the conductance of the first memory. (They describe erase as resetting in 3.5 (4) Polarity: "Most RRAMs are employed with bipolar switching where program (Set) and erase (Reset) operations require positive and negative voltage polarity to be applied across the device.") Resetting implies they are the same)
Regarding claim 4, Saxena in view of Serb teaches claim 1 as outlined above. Serb further teaches:
the performing of the adjusting of the first conductance value includes performing the adjusting of the first conductance value when the firing time difference is less than or equal to a threshold value. (Figure 2. “Firing frequency is modulated in four phases, targeting the induction of plasticity as per the sequence: none/LTP/none/LTD, using the chosen plasticity rule. (b) BN firing response to ABsyn inputs. After LTP induction, the original 10 Hz pacemaker stimulation becomes capable of eliciting BN action potentials, thus reflecting the increase of postsynaptic potential amplitudes to above threshold. Firing persists until the commencement of the depotentiation/depression phase. (c) MR2 weight evolution. Data points denote resistance values for the intended LTP (red), LTD (blue) or no polarity change (black) phases. The right vertical axis indicates the corresponding weight. X-axis common to all panels.”)).
Regarding claim 5, Saxena in view of Serb teaches claim 1 as outlined above. Saxena further teaches:
comprising converting the PSP of the second biological neuron into a pulse signal, wherein the firing time difference between the AP of the first biological neuron and PSP of the second biological neuron is a time difference between the AP of the first biological neuron and a firing time of the pulse signal. (3,2 2nd Paragraph: "This scheme effectively converts the time overlap (∆t) of pre and post spikes into program or erase voltage pulses" and Serb teaches the biological neuron as outlined above. )
Regarding claim 6, Saxena in view of Serb teaches claim 5 as outlined above. Saxena further teaches:
the first memory is a memory corresponding to a first cross point of a crossbar memory, and (3.1 "CMOS neurons and RRAM synapses are organized in a crossbar network to realize a single-level of neural interconnections as shown in Figure 1. In this architecture, each input neuron is connected to another output neuron through a two-terminal RRAM to form a crossbar, or cross-point, array.")
wherein performing of the adjusting of the first conductance value and the performing of the other adjusting of the second conductance value respectively include connecting the AP of the first biological neuron to a first electrode of the crossbar memory and connecting the PSP of the second biological neuron to a second electrode of the crossbar memory, where the first electrode and the second electrode overlap at the first cross point. (Section 3.2 first paragraph "STDP states that the synaptic weight w is updated according to the relative timing of the pre- and post-synaptic neuron firing… As illustrated in Figure 2a, a spike pair with the pre-synaptic spike arrives before the post-synaptic spike results in increasing the synaptic strength, or long-term potentiation (LTP); a pre-synaptic spike after a post-synaptic spike results in decreasing the synaptic strength, or long-term depression (LTD). Changes in the synaptic weight plotted as a function of the relative arrival timing of the post-synaptic spike with respect to the pre-synaptic spike is called the STDP learning function or learning window.” And 3.1 "CMOS neurons and RRAM synapses are organized in a crossbar network to realize a single-level of neural interconnections as shown in Figure 1. In this architecture, each input neuron is connected to another output neuron through a two-terminal RRAM to form a crossbar, or cross-point, array." And Serb teaches the connecting biological neurons to electrodes as outlined above).
Regarding claim 7, Saxena in view of Serb teaches claim 6 as outlined above. Saxena further teaches:
the overlapping of the first electrode and the second electrode at the first cross point includes the first electrode connecting to a first terminal of a resistive memory and the second electrode connecting to a second terminal of the resistive memory at the cross point. (3.1 "CMOS neurons and RRAM synapses are organized in a crossbar network to realize a single-levelof neural interconnections as shown in Figure 1. In this architecture, each input neuron is connected to another output neuron through a two-terminal RRAM to form a crossbar, or cross-point, array")
Regarding claim 8, Saxena in view of Serb teaches claim 1 as outlined above. Saxena further teaches:
the first memory is a memory corresponding to a cross point of a crossbar memory, where first electrodes of the crossbar memory overlap second electrodes of the crossbar memory at respective cross points of the crossbar memory. (3.1 "CMOS neurons and RRAM synapses are organized in a crossbar network to realize a single-level of neural interconnections as shown in Figure 1. In this architecture, each input neuron is connected to another output neuron through a two-terminal RRAM to form a crossbar, or cross-point, array")
Regarding claim 9, Saxena in view of Serb teaches claim 8 as outlined above. Serb further teaches:
the performing of the adjusting of the first conductance value of the first memory includes providing signals from first biological neurons to respective first electrodes of the crossbar memory, and providing signals from second biological neurons to respective second electrodes of the crossbar memory, and (Figure 1. “The two synaptors, ABsyn and BAsyn, connect the ‘presynaptic’ silicon neuron (ANpre) to the brain neuron (BN), and BN to the ‘postsynaptic’ silicon neuron, ANpost. The two memristors, MR1 and MR2, emulate plasticity in the two synaptors, whereas electronics-to-BN and BN-to-electronics signal transmission are mediated by the CME and the patch-clamp electrode.”)
Saxena further teaches:
wherein connection relationships between pairs of the first biological neurons and the second biological neurons are indicated by first memories of cross points of the crossbar memory with adjusted first conductance values. (You can see this relationship in Figure 1 specifically in (b) & (d) and Serb teaches the biological neurons as outlined above.)
Regarding claim 10, Saxena in view of Serb teaches claim 9 as outlined above. Serb further teaches:
the first biological neurons and the second biological neurons include at least one common biological neuron. (Figure. 1 shows this connection).
Regarding claim 11, Saxena in view of Serb teaches claim 9 as outlined above. Saxena further teaches:
the performing of the other adjusting of the second conductance value includes performing the other adjusting for only second conductance values of the first memories corresponding to the cross points of the crossbar memory with the adjusted first conductance values. (Figure 1 shows the crossbar and Section 3.3 second paragraph: “During the training phase, i.e., when the signal T = 1, the voltage spikes with positive pulse and negative tail are propagated in the forward (pre spikes) as well as the backward direction (post spikes). This enables learning by adjusting the synaptic weights using STDP based program or erase mechanism seen in Figure 2.” Training implies an iterative method meaning the adjusting happens multiple times.)
Regarding claim 12, Saxena in view of Serb teaches claim 8 as outlined above. Serb further teaches:
signals of first biological neurons are provided to the first electrodes and signals of second biological neurons are provided to the second electrodes, and (Discussion 2nd paragraph “Biological-to-electronic (in BAsyn) and electronic-to-biological (in ABsyn) signal transmission was realized through microelectrodes. For BAsyn, a patch-clamp microelectrode in whole-cell configuration recorded the spikes of the biological neuron. This invasive solution was preferred to non-invasive extracellular microelectrodes as it gave us the opportunity to capture subthreshold responses of BN to ABsyn activity in a very clean manner throughout the reported di-synaptor circuit experiment.”))
wherein the performing of the adjusting of the first conductance value further comprises adjusting first conductance values of first memories, of different cross points of the crossbar memory that respectively correspond to different pairs of the first and second biological neurons, corresponding to first pairs of the first and second biological neurons that have firing time differences less than or equal to a threshold value, representing that the each of the first pairs have pre-/post- synaptic relationships. (Page 1 2nd paragraph “The memristor MR1 stores synaptic weights as resistive states. The thin film capacitive microelectrode13 CME delivers stimuli to the biological neuron (BN) that are adjusted by the memristive weights (Fig. 1b). Thus, in analogy with a native synapse, ABsyn operates by injecting in the BN an excitatory current, which reflects a plasticity-dependent synaptic strength. To emulate plasticity, the memristor MR1 is operated as a two-terminal device through a control system that receives pre- and post-synaptic depolarisations from one silicon neuron (ANpre) and one biological neuron (BN), respectively.”).
Regarding claim 13, Saxena in view of Serb teaches claim 12 as outlined above. Saxena further teaches:
the performing of the other adjusting of the second conductance value includes performing the other adjusting of only the first memories corresponding to the first pairs. (Section 3.3 second paragraph: “During the training phase, i.e., when the signal T = 1, the voltage spikes with positive pulse and negative tail are propagated in the forward (pre spikes) as well as the backward direction (post spikes). This enables learning by adjusting the synaptic weights using STDP based program or erase mechanism seen in Figure 2.” Training implies an iterative method meaning the adjusting happens multiple times. Figure 1 and 2 shows the memories and pairs)
Regarding claim 14, Saxena in view of Serb teaches claim 1 as outlined above. Saxena further teaches:
the performing of the other adjusting of the second conductance value includes selectively performing the other adjusting based on determinations of which of first memories, corresponding to different pairs of first and second biological neurons, were adjusted in the performing of the adjusting of the first conductance value selectively for each of the different pairs of the first and second biological neurons. (3.2 Paragraph 2 Figure 1-2 "In pair-wise STDP learning, spikes sent from pre- and post-synaptic have their voltage amplitudes below the program and erase switching thresholds (V+th and V−th ) of a bipolar RRAM device. RRAM switching events may occur only if this spike pair overlaps and creates a net potential (Vnet) greater than the switching threshold, as illustrated in Figure 2b,c. Here, for Vnet > V+th , RRAM is incrementally programmed (conductance is increased) causing long-term potentiation (LTP) in the synapse. On the other hand, for the case Vnet < V−th , the RRAM is incrementally erased (conductance is decreased) and long-term depression (LTD) occurs in the synapse. In case of no temporal overlap, the pre-synaptic pulse is integrated in the neuron and thus should have a net positive area and smaller amplitude than the program or erase thresholds. This in turn sets a constraint for the voltage spikes that V−th < Vspk(t) < V+th must always be ensured to avoid disturbing the RRAM state. This scheme effectively converts the time overlap (∆t) of pre and post spikes into program or erase voltage pulses”).
Regarding claim 15, Saxena in view of Serb teaches claim 1 as outlined above. Serb further teaches:
the adjusting of the firing time difference comprises adjusting a firing time corresponding to the PSP of the second biological neuron based on an amplitude of the PSP of the second biological neuron. (Figure 2. “Firing frequency is modulated in four phases, targeting the induction of plasticity as per the sequence: none/LTP/none/LTD, using the chosen plasticity rule. (b) BN firing response to ABsyn inputs. After LTP induction, the original 10 Hz pacemaker stimulation becomes capable of eliciting BN action potentials, thus reflecting the increase of postsynaptic potential amplitudes to above threshold. Firing persists until the commencement of the depotentiation/depression phase. (c) MR2 weight evolution. Data points denote resistance values for the intended LTP (red), LTD (blue) or no polarity change (black) phases. The right vertical axis indicates the corresponding weight. X-axis common to all panels.”)
Regarding claim 16, Saxena in view of Serb teaches claim 1 as outlined above. Serb further teaches:
obtaining the AP of the first biological neuron and the PSP of the second biological neuron in real-time. (Discussion 3rd paragraph “TiOx memristors were at the core of plasticity emulation in both synaptors. Plasticity was implemented by pulse programming one memristor (per synapse) to store synaptic weights, which were computed and updated in real-time by a plasticity algorithm based on a BCM-inspired model.”).
Regarding claim 17, Saxena in view of Serb teaches claim 1 as outlined above. Saxena further teaches:
selectively adjusting another conductance value of the first memory based on a counted number of pairs, of PSPs of the second biological neuron and APs of the first biological neuron, having respective firing times within a first threshold, within a predetermined time window, wherein the adjusting of the conductance value is performed a number of times equal to the counted number, resulting in the first conductance value. (3.2 Paragraph 2 Figure 1-2 "In pair-wise STDP learning, spikes sent from pre- and post-synaptic have their voltage amplitudes below the program and erase switching thresholds (V+th and V−th ) of a bipolar RRAM device. RRAM switching events may occur only if this spike pair overlaps and creates a net potential (Vnet) greater than the switching threshold, as illustrated in Figure 2b,c. Here, for Vnet > V+th , RRAM is incrementally programmed (conductance is increased) causing long-term potentiation (LTP) in the synapse. On the other hand, for the case Vnet < V−th , the RRAM is incrementally erased (conductance is decreased) and long-term depression (LTD) occurs in the synapse. In case of no temporal overlap, the pre-synaptic pulse is integrated in the neuron and thus should have a net positive area and smaller amplitude than the program or erase thresholds. This in turn sets a constraint for the voltage spikes that V−th < Vspk(t) < V+th must always be ensured to avoid disturbing the RRAM state. This scheme effectively converts the time overlap (∆t) of pre and post spikes into program or erase voltage pulses” and Serb teaches the biological neurons as outlined above).
Regarding claim 19, Serb teaches:
generating a conductance value of a first memory corresponding to first and second biological neurons based on a counted number of pairs, of post-synaptic potential (PSPs) of the second biological neuron and action potentials (APs) of the first biological neuron, having respective firing times within a first threshold, within a predetermined time window. (Page 1 2nd paragraph “The memristor MR1 stores synaptic weights as resistive states. The thin film capacitive microelectrode13 CME delivers stimuli to the biological neuron (BN) that are adjusted by the memristive weights (Fig. 1b). Thus, in analogy with a native synapse, ABsyn operates by injecting in the BN an excitatory current, which reflects a plasticity-dependent synaptic strength. To emulate plasticity, the memristor MR1 is operated as a two-terminal device through a control system that receives pre- and post-synaptic depolarisations from one silicon neuron (ANpre) and one biological neuron (BN), respectively.”).
adjusting the conductance value whenever a firing time difference between an AP of the first biological neuron and a firing time of a PSP of the second biological neuron is less than or equal to the first threshold. (Figure 2. “Firing frequency is modulated in four phases, targeting the induction of plasticity as per the sequence: none/LTP/none/LTD, using the chosen plasticity rule. (b) BN firing response to ABsyn inputs. After LTP induction, the original 10 Hz pacemaker stimulation becomes capable of eliciting BN action potentials, thus reflecting the increase of postsynaptic potential amplitudes to above threshold. Firing persists until the commencement of the depotentiation/depression phase. (c) MR2 weight evolution. Data points denote resistance values for the intended LTP (red), LTD (blue) or no polarity change (black) phases. The right vertical axis indicates the corresponding weight. X-axis common to all panels.”)
Serb does not teach:
A method with artificial neural network generation, the method comprising:
However, Saxena does
A method with artificial neural network generation, the method comprising: ((Abstract) “In this article, we review the challenges involved and present a pathway to realize large-scale mixed-signal NeuSoCs, from device arrays and circuits to spike-based deep learning algorithms with ‘brain-like’ energy-efficiency”)
Serb and Saxena are considered analogous art to the claimed invention because they are in the same field of endeavor being spiking neural networks. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the biological neuron and silicon spiking neuron connect of Serb with the spiking neural network method of Saxena. One would want to do this to better replicate the natural neural networks of biology.
Regarding claim 20, Saxena in view of Serb teaches claim 19 as outlined above. Saxena further teaches:
counting the number of pairs, of the PSPs of the second biological neuron and the APs of the first biological neuron, that have the respective firing times within the first threshold, wherein the determining of the conductance value of the first memory further comprises determining the conductance value of a cross point memory, of a crossbar memory, corresponding to the first and second biological neurons in response to the first and second biological neurons being determined to have a pre-/post- synaptic relationship based on the counted number. (. (Section 3.2 first paragraph "STDP states that the synaptic weight w is updated according to the relative timing of the pre- and post-synaptic neuron firing” And 3.2 Paragraph 2 Figure 1-2 "In pair-wise STDP learning, spikes sent from pre- and post-synaptic have their voltage amplitudes below the program and erase switching thresholds (V+th and V−th ) of a bipolar RRAM device. RRAM switching events may occur only if this spike pair overlaps and creates a net potential (Vnet) greater than the switching threshold, as illustrated in Figure 2b,c. Here, for Vnet > V+th , RRAM is incrementally programmed (conductance is increased) causing long-term potentiation (LTP) in the synapse. On the other hand, for the case Vnet < V−th , the RRAM is incrementally erased (conductance is decreased) and long-term depression (LTD) occurs in the synapse. In case of no temporal overlap, the pre-synaptic pulse is integrated in the neuron and thus should have a net positive area and smaller amplitude than the program or erase thresholds. This in turn sets a constraint for the voltage spikes that V−th < Vspk(t) < V+th must always be ensured to avoid disturbing the RRAM state. This scheme effectively converts the time overlap (∆t) of pre and post spikes into program or erase voltage pulses” And figures 1 and 2 show the crossbar memory and Serb teaches the biological neurons as outlined above).
Regarding claim 22, Saxena in view of Serb teaches claim 19 as outlined above. Saxena further teaches:
converting a PSP of the second biological neuron into a pulse signal, wherein the determining of the conductance value of the first memory further comprises determining the conductance value of a cross point memory of the first memory based on a timing difference between an arrival time of an AP of the first biological neuron and an arrival time of the pulse signal of the second biological neuron (3,2 2nd Paragraph: "This scheme effectively converts the time overlap (∆t) of pre and post spikes into program or erase voltage pulses" and Serb teaches the biological neuron as outlined above.)
Regarding claim 23, Saxena in view of Serb teaches claim 19 as outlined above. Serb further teaches:
obtaining the APs of the first biological neuron and the PSPs of the second biological neuron in real-time. (Discussion 3rd paragraph “TiOx memristors were at the core of plasticity emulation in both synaptors. Plasticity was implemented by pulse programming one memristor (per synapse) to store synaptic weights, which were computed and updated in real-time by a plasticity algorithm based on a BCM-inspired model.”).
Regarding claim 24, Serb teaches:
selectively adjusting a conductance value of a first memory corresponding to first and second biological neurons based on a counted number of pairs, of post-synaptic potential (PSPs) of the second biological neuron and action potentials (APs) of the first biological neuron, having respective firing times within a first threshold, within a predetermined time window. (Page 1 2nd paragraph “The memristor MR1 stores synaptic weights as resistive states. The thin film capacitive microelectrode13 CME delivers stimuli to the biological neuron (BN) that are adjusted by the memristive weights (Fig. 1b). Thus, in analogy with a native synapse, ABsyn operates by injecting in the BN an excitatory current, which reflects a plasticity-dependent synaptic strength. To emulate plasticity, the memristor MR1 is operated as a two-terminal device through a control system that receives pre- and post-synaptic depolarisations from one silicon neuron (ANpre) and one biological neuron (BN), respectively.”).
Serb does not teach:
A method with artificial neural network generation, the method comprising:
However, Saxena does
A method with artificial neural network generation, the method comprising: ((Abstract) “In this article, we review the challenges involved and present a pathway to realize large-scale mixed-signal NeuSoCs, from device arrays and circuits to spike-based deep learning algorithms with ‘brain-like’ energy-efficiency”)
Serb and Saxena are considered analogous art to the claimed invention because they are in the same field of endeavor being spiking neural networks. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the biological neuron and silicon spiking neuron connect of Serb with the spiking neural network method of Saxena. One would want to do this to better replicate the natural neural networks of biology.
Regarding claim 25, Saxena in view of Serb teaches claim 24 as outlined above. Serb further teaches:
the adjusting of the conductance value is performed a determined number of times. (Discussion 3rd paragraph “TiOx memristors were at the core of plasticity emulation in both synaptors. Plasticity was implemented by pulse programming one memristor (per synapse) to store synaptic weights, which were computed and updated in real-time by a plasticity algorithm based on a BCM-inspired model. BAsyn weights were converted to current injections into the silicon neuron; ABsyn weights, instead, were transformed in depolarising voltage stimuli delivered through the capacitive microelectrode to the biological neuron. Thus, by making an analogy between ABsyn and an excitatory glutamatergic synapse, transmembrane currents induced by capacitive stimulation corresponded to currents through glutamate AMPA receptors; the resistive states of the memristor were changes of AMPA conductance driven by long-term plasticity; the plasticity algorithm was collectively representing the molecular mechanisms leading to changes of AMPA conductance (e.g., NMDA-dependent mechanisms).”)
Regarding claim 26, Saxena in view of Serb teaches claim 25 as outlined above. Serb further teaches:
the selective adjusting includes selecting the adjusting of the conductance value in response to the counted number being greater than or equal to a threshold. (Figure 2. “Firing frequency is modulated in four phases, targeting the induction of plasticity as per the sequence: none/LTP/none/LTD, using the chosen plasticity rule. (b) BN firing response to ABsyn inputs. After LTP induction, the original 10 Hz pacemaker stimulation becomes capable of eliciting BN action potentials, thus reflecting the increase of postsynaptic potential amplitudes to above threshold. Firing persists until the commencement of the depotentiation/depression phase. (c) MR2 weight evolution. Data points denote resistance values for the intended LTP (red), LTD (blue) or no polarity change (black) phases. The right vertical axis indicates the corresponding weight. X-axis common to all panels.”)).
Regarding claim 27, Saxena in view of Serb teaches claim 26 as outlined above. Saxena further teaches:
the determined number of times is equal to the counted number. (Section 3.2 first paragraph “RRAM switching events may occur only if this spike pair overlaps and creates a net potential (Vnet) greater than the switching threshold, as illustrated in Figure 2b,c.”).
Regarding claim 28, Saxena in view of Serb teaches claim 25 as outlined above. Saxena further teaches:
performing another adjusting, based on a firing time difference between an AP of the first biological neuron and a PSP of the second biological neuron, of a first conductance value of the first memory, wherein the first conductance value is a resultant conductance value of the adjusting of the conductance value the determined number of times. (Section 3.3 second paragraph: “During the training phase, i.e., when the signal T = 1, the voltage spikes with positive pulse and negative tail are propagated in the forward (pre spikes) as well as the backward direction (post spikes). This enables learning by adjusting the synaptic weights using STDP based program or erase mechanism seen in Figure 2.” Training implies an iterative method meaning the adjusting happens multiple times and Serb teaches the biological neuron as outlined above.)
Regarding claim 29, Saxena in view of Serb teaches claim 24 as outlined above. Saxena further teaches:
the first memory is a memory corresponding to a cross point of a crossbar memory, where first electrodes of the crossbar memory overlap second electrodes of the crossbar memory at respective cross points of the crossbar memory. (3.1 "CMOS neurons and RRAM synapses are organized in a crossbar network to realize a single-level of neural interconnections as shown in Figure 1. In this architecture, each input neuron is connected to another output neuron through a two-terminal RRAM to form a crossbar, or cross-point, array.")
Regarding claim 30, Saxena in view of Serb teaches claim 29 as outlined above. Serb further teaches:
the selective adjusting includes providing signals from first biological neurons to respective first electrodes of the crossbar memory, and providing signals from second biological neurons to respective second electrodes of the crossbar memory, and (Discussion 2nd paragraph “Biological-to-electronic (in BAsyn) and electronic-to-biological (in ABsyn) signal transmission was realized through microelectrodes. For BAsyn, a patch-clamp microelectrode in whole-cell configuration recorded the spikes of the biological neuron. This invasive solution was preferred to non-invasive extracellular microelectrodes as it gave us the opportunity to capture subthreshold responses of BN to ABsyn activity in a very clean manner throughout the reported di-synaptor circuit experiment.”))
Saxena further teaches:
wherein connection relationships and strengths between pairs of the first biological neurons and the second biological neurons are indicated by corresponding first memories of cross points of the crossbar memory that have adjusted conductance values resulting from the selective adjusting. (You can see this relationship in Figure 1 specifically in (b) & (d) and Serb teaches the biological neurons as outlined above.)
Regarding claim 31, Saxena in view of Serb teaches claim 24 as outlined above. Saxena further teaches:
counting the number of pairs, of the PSPs of the second biological neuron and the APs of the first biological neurons, that have the respective firing times within the first threshold, and (3.2 Paragraph 2 Figure 1-2 "In pair-wise STDP learning, spikes sent from pre- and post-synaptic have their voltage amplitudes below the program and erase switching thresholds (V+th and V−th ) of a bipolar RRAM device. RRAM switching events may occur only if this spike pair overlaps and creates a net potential (Vnet) greater than the switching threshold, as illustrated in Figure 2b,c. Here, for Vnet > V+th , RRAM is incrementally programmed (conductance is increased) causing long-term potentiation (LTP) in the synapse. On the other hand, for the case Vnet < V−th , the RRAM is incrementally erased (conductance is decreased) and long-term depression (LTD) occurs in the synapse” and Serb teaches the biological neuron as outlined above.)
determining that the first and second biological neurons have a pre-/post- synaptic relationship in response to the counted number being greater than or equal to the second threshold value. (Figure 2(c) shows the potential being less that a threshold "For V_net < V^−_th, LTD occurs and the RRAM conductance is decreased (erase operation)." And Serb teaches the biological neuron)
Regarding claim 33, Serb teaches:
An electronic device comprising: (Discussion 2nd paragraph “Synaptors were implemented by relying on two separate physical electronic components: one for signal trans mission and one for plasticity. Biological-to-electronic (in BAsyn) and electronic-to-biological (in ABsyn) signal transmission was realized through microelectrodes.)
a processor configured to: (Methods “The central part of the artificial side of the bio-hybrid system is formed by a reconfigurable on-line learning spiking neuromorphic processor (ROLLS)28, which contains neuromorphic CMOS circuits emulating short-term plasticity (STP) properties of synapses29 and long-term plasticity (LTP) ones30. In addition, this processor comprises mixed signal analogue-digital circuits which implement a model of the adaptive exponential integrate-and-fire neuron31.”)
control a provision of an action potential (AP) of a first biological neuron to the first electrode, and a provision of a post-synaptic potential (PSP) of a second biological neuron to the second electrode, to selectively adjust, based on a firing time difference between the AP of the first biological neuron and the PSP of the second biological neuron, a first conductance value of the first memory; (Page 1 2nd paragraph “The memristor MR1 stores synaptic weights as resistive states. The thin film capacitive microelectrode13 CME delivers stimuli to the biological neuron (BN) that are adjusted by the memristive weights (Fig. 1b). Thus, in analogy with a native synapse, ABsyn operates by injecting in the BN an excitatory current, which reflects a plasticity-dependent synaptic strength. To emulate plasticity, the memristor MR1 is operated as a two-terminal device through a control system that receives pre- and post-synaptic depolarisations from one silicon neuron (ANpre) and one biological neuron (BN), respectively.”).
adjust the firing time difference based on the PSP of a second biological neuron; (Figure 2. “Firing frequency is modulated in four phases, targeting the induction of plasticity as per the sequence: none/LTP/none/LTD, using the chosen plasticity rule. (b) BN firing response to ABsyn inputs. After LTP induction, the original 10 Hz pacemaker stimulation becomes capable of eliciting BN action potentials, thus reflecting the increase of postsynaptic potential amplitudes to above threshold. Firing persists until the commencement of the depotentiation/depression phase. (c) MR2 weight evolution. Data points denote resistance values for the intended LTP (red), LTD (blue) or no polarity change (black) phases. The right vertical axis indicates the corresponding weight. X-axis common to all panels.”)
Serb does not teach:
a crossbar memory having first electrodes, including a first electrode, that overlap second electrodes, including a second electrode, of the crossbar memory at respective cross points of the crossbar memory, with the first electrode and the second electrode overlapping to respectively connect to a first memory at a first cross point of the crossbar memory; and
and control another provision of the AP of the first biological neuron to the first electrode and the PSP of the second biological neuron to the second electrode, having the adjusted firing time difference between the AP of the first biological neuron and the PSP of the second biological neuron, to selectively adjust a second conductance value of the first memory
However Saxena does:
a crossbar memory having first electrodes, including a first electrode, that overlap second electrodes, including a second electrode, of the crossbar memory at respective cross points of the crossbar memory, with the first electrode and the second electrode overlapping to respectively connect to a first memory at a first cross point of the crossbar memory; and (3.1 "CMOS neurons and RRAM synapses are organized in a crossbar network to realize a single-level of neural interconnections as shown in Figure 1. In this architecture, each input neuron is connected to another output neuron through a two-terminal RRAM to form a crossbar, or cross-point, array.")
and control another provision of the AP of the first biological neuron to the first electrode and the PSP of the second biological neuron to the second electrode, having the adjusted firing time difference between the AP of the first biological neuron and the PSP of the second biological neuron, to selectively adjust a second conductance value of the first memory. (Section 3.3 second paragraph: “During the training phase, i.e., when the signal T = 1, the voltage spikes with positive pulse and negative tail are propagated in the forward (pre spikes) as well as the backward direction (post spikes). This enables learning by adjusting the synaptic weights using STDP based program or erase mechanism seen in Figure 2.” Training implies an iterative method meaning the adjusting happens multiple times.)
Serb and Saxena are considered analogous art to the claimed invention because they are in the same field of endeavor being spiking neural networks. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the biological neuron and silicon spiking neuron connect of Serb with the spiking neural network method of Saxena. One would want to do this to better replicate the natural neural networks of biology.
Regarding claim 34, Saxena in view of Serb teaches claim 33 as outlined above. Saxena further teaches:
the processor is further configured to control a resetting of a conductance of the first memory after the first conductance value is adjusted and before the second conductance value is adjusted. (They describe erase as resetting in 3.5 (4) Polarity: "Most RRAMs are employed with bipolar switching where program (Set) and erase (Reset) operations require positive and negative voltage polarity to be applied across the device.")
Regarding claim 35, Saxena in view of Serb teaches claim 34 as outlined above. Saxena further teaches:
the second conductance value is same as the first conductance value upon the resetting of the conductance of the first memory. (They describe erase as resetting in 3.5 (4) Polarity: "Most RRAMs are employed with bipolar switching where program (Set) and erase (Reset) operations require positive and negative voltage polarity to be applied across the device.") Resetting implies they are the same)
Regarding claim 36, Saxena in view of Serb teaches claim 33 as outlined above. Serb further teaches:
the first conductance value is adjusted in response to the firing time difference being less than or equal to a threshold value. (Figure 2. “Firing frequency is modulated in four phases, targeting the induction of plasticity as per the sequence: none/LTP/none/LTD, using the chosen plasticity rule. (b) BN firing response to ABsyn inputs. After LTP induction, the original 10 Hz pacemaker stimulation becomes capable of eliciting BN action potentials, thus reflecting the increase of postsynaptic potential amplitudes to above threshold. Firing persists until the commencement of the depotentiation/depression phase. (c) MR2 weight evolution. Data points denote resistance values for the intended LTP (red), LTD (blue) or no polarity change (black) phases. The right vertical axis indicates the corresponding weight. X-axis common to all panels.”)).
Regarding claim 37, Saxena in view of Serb teaches claim 33 as outlined above. Saxena further teaches:
convert the PSP of the second biological neuron into a pulse signal; and (3,2 2nd Paragraph: "This scheme effectively converts the time overlap (∆t) of pre and post spikes into program or erase voltage pulses" and Serb teaches the biological neurons as outlined above.)
perform the control of the provision of the PSP of the second biological neuron to the second electrode by providing the pulse to the second electrode, and perform the control of the other provision of the PSP of the second biological neuron to the second electrode by providing the pulse to the second electrode, wherein the firing time difference between the AP of the first biological neuron and the PSP of the second biological neuron is a time difference between the AP of the first biological neuron and a firing time of the pulse signal, and (3,2 2nd Paragraph: "This scheme effectively converts the time overlap (∆t) of pre and post spikes into program or erase voltage pulses" and Serb teaches the biological neurons as outlined above.)
wherein the adjusted firing time difference is between the AP of the first biological neuron and the pulse signal corresponding to the adjusted firing time difference. (3,2 2nd Paragraph: "This scheme effectively converts the time overlap (∆t) of pre and post spikes into program or erase voltage pulses" and Serb teaches the biological neurons as outlined above. )
Regarding claim 38, Saxena in view of Serb teaches claim 33 as outlined above. Serb further teaches:
the adjustment of the firing time difference, the processor is configured to adjust a firing time corresponding to the PSP of the second biological neuron based on an amplitude of the PSP of the second biological neuron (Figure 2. “Firing frequency is modulated in four phases, targeting the induction of plasticity as per the sequence: none/LTP/none/LTD, using the chosen plasticity rule. (b) BN firing response to ABsyn inputs. After LTP induction, the original 10 Hz pacemaker stimulation becomes capable of eliciting BN action potentials, thus reflecting the increase of postsynaptic potential amplitudes to above threshold. Firing persists until the commencement of the depotentiation/depression phase. (c) MR2 weight evolution. Data points denote resistance values for the intended LTP (red), LTD (blue) or no polarity change (black) phases. The right vertical axis indicates the corresponding weight. X-axis common to all panels.”)
Regarding claim 39, Serb teaches:
An electronic device comprising: (Discussion 2nd paragraph “Synaptors were implemented by relying on two separate physical electronic components: one for signal trans mission and one for plasticity. Biological-to-electronic (in BAsyn) and electronic-to-biological (in ABsyn) signal transmission was realized through microelectrodes.)
a processor configured to control a provision of action potentials (APs) of a first biological neuron to the first electrode, and a provision of post-synaptic potentials (PSPs) of a second biological neuron to the second electrode, to selectively adjust a conductance value of the first memory based on a counted number of pairs, of the PSPs of the second biological neuron and the APs of the first biological neuron, having respective firing times within a first threshold, within a predetermined time window. (Page 1 2nd paragraph “The memristor MR1 stores synaptic weights as resistive states. The thin film capacitive microelectrode13 CME delivers stimuli to the biological neuron (BN) that are adjusted by the memristive weights (Fig. 1b). Thus, in analogy with a native synapse, ABsyn operates by injecting in the BN an excitatory current, which reflects a plasticity-dependent synaptic strength. To emulate plasticity, the memristor MR1 is operated as a two-terminal device through a control system that receives pre- and post-synaptic depolarisations from one silicon neuron (ANpre) and one biological neuron (BN), respectively.”).
Serb does not teach:
a crossbar memory having first electrodes, including a first electrode, that overlap second electrodes, including a second electrode, of the crossbar memory at respective cross points of the crossbar memory, with the first electrode and the second electrode overlapping to respectively connect to a first memory corresponding to a first cross point of the crossbar memory; and
However, Saxena does:
a crossbar memory having first electrodes, including a first electrode, that overlap second electrodes, including a second electrode, of the crossbar memory at respective cross points of the crossbar memory, with the first electrode and the second electrode overlapping to respectively connect to a first memory corresponding to a first cross point of the crossbar memory; and (3.1 "CMOS neurons and RRAM synapses are organized in a crossbar network to realize a single-level of neural interconnections as shown in Figure 1. In this architecture, each input neuron is connected to another output neuron through a two-terminal RRAM to form a crossbar, or cross-point, array.")
Serb and Saxena are considered analogous art to the claimed invention because they are in the same field of endeavor being spiking neural networks. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the biological neuron and silicon spiking neuron connect of Serb with the spiking neural network method of Saxena. One would want to do this to better replicate the natural neural networks of biology.
Regarding claim 40, Saxena in view of Serb teaches claim 39 as outlined above. Saxena further teaches:
the control of the provision of the APs of the first biological neuron to the first electrode and the provision of the PSPs of the second biological neuron, includes providing respective APs of the first biological neural and respective PSPs of the second biological neuron a number of times equal to the counted number of pairs to adjust the conductance value of the first memory based the counted number of pairs. (3.2 Paragraph 2 Figure 1-2 "In pair-wise STDP learning, spikes sent from pre- and post-synaptic have their voltage amplitudes below the program and erase switching thresholds (V+th and V−th ) of a bipolar RRAM device. RRAM switching events may occur only if this spike pair overlaps and creates a net potential (Vnet) greater than the switching threshold, as illustrated in Figure 2b,c. Here, for Vnet > V+th , RRAM is incrementally programmed (conductance is increased) causing long-term potentiation (LTP) in the synapse. On the other hand, for the case Vnet < V−th , the RRAM is incrementally erased (conductance is decreased) and long-term depression (LTD) occurs in the synapse. In case of no temporal overlap, the pre-synaptic pulse is integrated in the neuron and thus should have a net positive area and smaller amplitude than the program or erase thresholds. This in turn sets a constraint for the voltage spikes that V−th < Vspk(t) < V+th must always be ensured to avoid disturbing the RRAM state. This scheme effectively converts the time overlap (∆t) of pre and post spikes into program or erase voltage pulses” and Serb teaches the biological neurons as outlined above.).
Regarding claim 41, Saxena in view of Serb teaches claim 39 as outlined above. Saxena further teaches:
the processor is further configured to count the number of pairs, of the PSPs of the second biological neuron and the APs of the first biological neuron, that have the respective firing times within the first threshold, and in response to the counted number being greater than or equal to a threshold, perform the provision of the APs of the fist biological neuron and the provision of the PSPs of the second biological neuron. (3.2 Paragraph 2 Figure 1-2 "In pair-wise STDP learning, spikes sent from pre- and post-synaptic have their voltage amplitudes below the program and erase switching thresholds (V+th and V−th ) of a bipolar RRAM device. RRAM switching events may occur only if this spike pair overlaps and creates a net potential (Vnet) greater than the switching threshold, as illustrated in Figure 2b,c. Here, for Vnet > V+th , RRAM is incrementally programmed (conductance is increased) causing long-term potentiation (LTP) in the synapse. On the other hand, for the case Vnet < V−th , the RRAM is incrementally erased (conductance is decreased) and long-term depression (LTD) occurs in the synapse. In case of no temporal overlap, the pre-synaptic pulse is integrated in the neuron and thus should have a net positive area and smaller amplitude than the program or erase thresholds. This in turn sets a constraint for the voltage spikes that V−th < Vspk(t) < V+th must always be ensured to avoid disturbing the RRAM state. This scheme effectively converts the time overlap (∆t) of pre and post spikes into program or erase voltage pulses” and Serb teaches the biological neurons as outlined above).
Regarding claim 42, Saxena in view of Serb teaches claim 39 as outlined above. Saxena further teaches:
the processor is further configured to convert a PSP of the second biological neuron into a pulse signal, and (3,2 2nd Paragraph: "This scheme effectively converts the time overlap (∆t) of pre and post spikes into program or erase voltage pulses" and Serb teaches the biological neurons as outlined above)
determine the conductance value of a cross point memory of the crossbar memory based on a timing difference between an arrival time of an AP of the first biological neuron and an arrival time of the pulse signal of the second biological neuron. (Section 3.2 first paragraph "STDP states that the synaptic weight w is updated according to the relative timing of the pre- and post-synaptic neuron firing… As illustrated in Figure 2a, a spike pair with the pre-synaptic spike arrives before the post-synaptic spike results in increasing the synaptic strength, or long-term potentiation (LTP); a pre-synaptic spike after a post-synaptic spike results in decreasing the synaptic strength, or long-term depression (LTD). Changes in the synaptic weight plotted as a function of the relative arrival timing of the post-synaptic spike with respect to the pre-synaptic spike is called the STDP learning function or learning window.” and Serb teaches the biological neurons as outlined above).
Regarding claim 43, Saxena in view of Serb teaches claim 39 as outlined above. Serb further teaches:
the processor is further configured to obtain the APs of the first biological neuron and the PSPs of the second biological neuron in real-time. (Discussion 3rd paragraph “TiOx memristors were at the core of plasticity emulation in both synaptors. Plasticity was implemented by pulse programming one memristor (per synapse) to store synaptic weights, which were computed and updated in real-time by a plasticity algorithm based on a BCM-inspired model.”).
Regarding claim 44, Serb teaches:
obtaining action potentials (APs) of a first biological neuron of a natural neural network and post-synaptic potentials (PSPs) of a second biological neuron of the natural neural network; (Figure 1. “MR1 resistive states are translated into weighted voltage stimuli. These are delivered to BN through the CME capacitance (CCME) causing EPSP-like depolarisations, in turn leading to action potential firing (Supplementary Fig. 1).”)
adjusting, based on a firing time difference between an AP of the first biological neuron and a PSP of the second biological neuron, a conductance value of the first cross point memory; (Figure 2. “Firing frequency is modulated in four phases, targeting the induction of plasticity as per the sequence: none/LTP/none/LTD, using the chosen plasticity rule. (b) BN firing response to ABsyn inputs. After LTP induction, the original 10 Hz pacemaker stimulation becomes capable of eliciting BN action potentials, thus reflecting the increase of postsynaptic potential amplitudes to above threshold. Firing persists until the commencement of the depotentiation/depression phase. (c) MR2 weight evolution. Data points denote resistance values for the intended LTP (red), LTD (blue) or no polarity change (black) phases. The right vertical axis indicates the corresponding weight. X-axis common to all panels.”)
adjusting the firing time difference based on the PSP of the second biological neuron; and (Section 3.2 first paragraph “Changes in the synaptic weight plotted as a function of the relative arrival timing of the post-synaptic spike with respect to the pre-synaptic spike is called the STDP learning function or learning window. Furthermore, during the inference mode, only the pre-spikes with the positive rectangular pulse are used for carrying the feedforward inputs through the SNN. The post-spikes and the negative tails are activated during the training mode only to enable on-chip learning. This not only saves energy but also avoids undesirable changes to the synaptic weights).
Serb does not teach:
A method with artificial neural network generation, the method comprising:
determining a first conductance value of a first cross point memory corresponding to the first and second biological neurons, as a first mode; determining a second conductance value of a second cross point memory corresponding to the first and second biological neurons, as a second mode; and determining a final conductance value of a final cross point memory based on at least one of the first conductance value and the second conductance value, wherein the determining of the first conductance value comprises:
controlling another firing time difference between the AP of the first biological neuron and the PSP of the second biological neuron to be the adjusted firing time difference, and based on the controlled other firing time difference, adjusting another conductance value of the first cross point memory, to be the first conductance value, and
wherein the determining of the second conductance value comprises counting a number of pairs, of the PSPs of the second biological neuron and the APs of the first biological neuron, that have respective firing times within a first threshold, and selectively adjusting a conductance value of the second cross point memory based on the counted number of pairs, to be the second conductance value.
However Saxena does:
A method with artificial neural network generation, the method comprising: ((Abstract) “In this article, we review the challenges involved and present a pathway to realize large-scale mixed-signal NeuSoCs, from device arrays and circuits to spike-based deep learning algorithms with ‘brain-like’ energy-efficiency”)
determining a first conductance value of a first cross point memory corresponding to the first and second biological neurons, as a first mode; determining a second conductance value of a second cross point memory corresponding to the first and second biological neurons, as a second mode; and determining a final conductance value of a final cross point memory based on at least one of the first conductance value and the second conductance value, wherein the determining of the first conductance value comprises: (Introduction “"Inspired by biological nervous systems, artificial neural networks (ANNs) were developed that have achieved remarkable success in a few specific applications" and " The discovery of spike-timing-dependent-plasticity (STDP) local learning rule [5,6] and mathematical analysis of spike-based winner-take-all (WTA) motifs have opened new avenues in spike-based neural network research” And Section 3.2 first paragraph "STDP states that the synaptic weight w is updated according to the relative timing of the pre- and post-synaptic neuron firing… As illustrated in Figure 2a, a spike pair with the pre-synaptic spike arrives before the post-synaptic spike results in increasing the synaptic strength, or long-term potentiation (LTP); a pre-synaptic spike after a post-synaptic spike results in decreasing the synaptic strength, or long-term depression (LTD). Changes in the synaptic weight plotted as a function of the relative arrival timing of the post-synaptic spike with respect to the pre-synaptic spike is called the STDP learning function or learning window.” And Section 3.3 second paragraph: “During the training phase, i.e., when the signal T = 1, the voltage spikes with positive pulse and negative tail are propagated in the forward (pre spikes) as well as the backward direction (post spikes). This enables learning by adjusting the synaptic weights using STDP based program or erase mechanism seen in Figure 2.” Training implies an iterative method meaning the adjusting happens multiple times).
controlling another firing time difference between the AP of the first biological neuron and the PSP of the second biological neuron to be the adjusted firing time difference, and based on the controlled other firing time difference, adjusting another conductance value of the first cross point memory, to be the first conductance value, and (Section 3.4 “By modeling the membrane potential of the integrate-and-fire neuron with noisy inputs (a valid assumption with circuit noise and/or noisy spike inputs) as a type of Brownian motion, a closed-form expression to relate the input and output firing rates of the neuron was determined, and thus its derivative.”)
wherein the determining of the second conductance value comprises counting a number of pairs, of the PSPs of the second biological neuron and the APs of the first biological neuron, that have respective firing times within a first threshold, and selectively adjusting a conductance value of the second cross point memory based on the counted number of pairs, to be the second conductance value. (3.2 Paragraph 2 Figure 1-2 "In pair-wise STDP learning, spikes sent from pre- and post-synaptic have their voltage amplitudes below the program and erase switching thresholds (V+th and V−th ) of a bipolar RRAM device. RRAM switching events may occur only if this spike pair overlaps and creates a net potential (Vnet) greater than the switching threshold, as illustrated in Figure 2b,c. Here, for Vnet > V+th , RRAM is incrementally programmed (conductance is increased) causing long-term potentiation (LTP) in the synapse. On the other hand, for the case Vnet < V−th , the RRAM is incrementally erased (conductance is decreased) and long-term depression (LTD) occurs in the synapse. In case of no temporal overlap, the pre-synaptic pulse is integrated in the neuron and thus should have a net positive area and smaller amplitude than the program or erase thresholds. This in turn sets a constraint for the voltage spikes that V−th < Vspk(t) < V+th must always be ensured to avoid disturbing the RRAM state. This scheme effectively converts the time overlap (∆t) of pre and post spikes into program or erase voltage pulses”).
Serb and Saxena are considered analogous art to the claimed invention because they are in the same field of endeavor being spiking neural networks. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the biological neuron and silicon spiking neuron connect of Serb with the spiking neural network method of Saxena. One would want to do this to better replicate the natural neural networks of biology.
Regarding claim 45, Saxena in view of Serb teaches claim 44 as outlined above. Saxena further teaches:
the first cross point memory, the second cross point memory, and final cross point memory are a same cross point memory. (3.1 "CMOS neurons and RRAM synapses are organized in a crossbar network to realize a single-level of neural interconnections as shown in Figure 1. In this architecture, each input neuron is connected to another output neuron through a two-terminal RRAM to form a crossbar, or cross-point, array.")
Regarding claim 46, Serb teaches:
An electronic device comprising: (Discussion 2nd paragraph “Synaptors were implemented by relying on two separate physical electronic components: one for signal trans mission and one for plasticity. Biological-to-electronic (in BAsyn) and electronic-to-biological (in ABsyn) signal transmission was realized through microelectrodes.)
a processor configured to: (Methods “The central part of the artificial side of the bio-hybrid system is formed by a reconfigurable on-line learning spiking neuromorphic processor (ROLLS)28, which contains neuromorphic CMOS circuits emulating short-term plasticity (STP) properties of synapses29 and long-term plasticity (LTP) ones30. In addition, this processor comprises mixed signal analogue-digital circuits which implement a model of the adaptive exponential integrate-and-fire neuron31.”)
obtain action potentials (APs) of a first biological neuron of a natural neural network and post-synaptic potentials (PSPs) of a second biological neuron of the natural neural network; (Figure 1. “MR1 resistive states are translated into weighted voltage stimuli. These are delivered to BN through the CME capacitance (CCME) causing EPSP-like depolarisations, in turn leading to action potential firing (Supplementary Fig. 1).”)
adjust, based on a firing time difference between an AP of the first biological neuron and a PSP of the second biological neuron, a conductance value of the first cross point memory; (Figure 2. “Firing frequency is modulated in four phases, targeting the induction of plasticity as per the sequence: none/LTP/none/LTD, using the chosen plasticity rule. (b) BN firing response to ABsyn inputs. After LTP induction, the original 10 Hz pacemaker stimulation becomes capable of eliciting BN action potentials, thus reflecting the increase of postsynaptic potential amplitudes to above threshold. Firing persists until the commencement of the depotentiation/depression phase. (c) MR2 weight evolution. Data points denote resistance values for the intended LTP (red), LTD (blue) or no polarity change (black) phases. The right vertical axis indicates the corresponding weight. X-axis common to all panels.”)
Serb does not teach:
a crossbar memory having first electrodes, including a first electrode, that overlap second electrodes, including a second electrode, of the crossbar memory at respective cross points of the crossbar memory, with the first electrode and the second electrode overlapping to respectively connect to a first memory corresponding to a first cross point of the crossbar memory; and
determine a first conductance value of a first cross point memory corresponding to the first and second biological neurons, as a first mode; determine a second conductance value of a second cross point memory corresponding to the first and second biological neurons, as a second mode; and determine a final conductance value of a final cross point memory based on at least one of the first conductance value and the second conductance value.
However, Saxena does:
a crossbar memory having first electrodes, including a first electrode, that overlap second electrodes, including a second electrode, of the crossbar memory at respective cross points of the crossbar memory, with the first electrode and the second electrode overlapping to respectively connect to a first memory corresponding to a first cross point of the crossbar memory; and (3.1 "CMOS neurons and RRAM synapses are organized in a crossbar network to realize a single-level of neural interconnections as shown in Figure 1. In this architecture, each input neuron is connected to another output neuron through a two-terminal RRAM to form a crossbar, or cross-point, array.")
determine a first conductance value of a first cross point memory corresponding to the first and second biological neurons, as a first mode; determine a second conductance value of a second cross point memory corresponding to the first and second biological neurons, as a second mode; and determine a final conductance value of a final cross point memory based on at least one of the first conductance value and the second conductance value. (Section 3.2 first paragraph "STDP states that the synaptic weight w is updated according to the relative timing of the pre- and post-synaptic neuron firing… As illustrated in Figure 2a, a spike pair with the pre-synaptic spike arrives before the post-synaptic spike results in increasing the synaptic strength, or long-term potentiation (LTP); a pre-synaptic spike after a post-synaptic spike results in decreasing the synaptic strength, or long-term depression (LTD). Changes in the synaptic weight plotted as a function of the relative arrival timing of the post-synaptic spike with respect to the pre-synaptic spike is called the STDP learning function or learning window.” And Section 3.3 second paragraph: “During the training phase, i.e., when the signal T = 1, the voltage spikes with positive pulse and negative tail are propagated in the forward (pre spikes) as well as the backward direction (post spikes). This enables learning by adjusting the synaptic weights using STDP based program or erase mechanism seen in Figure 2.” Training implies an iterative method meaning the adjusting happens multiple times).
Serb and Saxena are considered analogous art to the claimed invention because they are in the same field of endeavor being spiking neural networks. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the biological neuron and silicon spiking neuron connect of Serb with the spiking neural network method of Saxena. One would want to do this to better replicate the natural neural networks of biology.
Regarding claim 47, Saxena in view of Serb teaches claim 46 as outlined above. Serb further teaches:
adjust the firing time difference based on the PSP of the second biological neuron; and (Figure 2. “Firing frequency is modulated in four phases, targeting the induction of plasticity as per the sequence: none/LTP/none/LTD, using the chosen plasticity rule. (b) BN firing response to ABsyn inputs. After LTP induction, the original 10 Hz pacemaker stimulation becomes capable of eliciting BN action potentials, thus reflecting the increase of postsynaptic potential amplitudes to above threshold. Firing persists until the commencement of the depotentiation/depression phase. (c) MR2 weight evolution. Data points denote resistance values for the intended LTP (red), LTD (blue) or no polarity change (black) phases. The right vertical axis indicates the corresponding weight. X-axis common to all panels.”)
Saxena further teaches:
control another firing time difference between the AP of the first biological neuron and the PSP of the second biological neuron to be the adjusted firing time difference, and based on the controlled other firing time difference, adjust another conductance value of the first cross point memory, to be the first conductance value. (Section 3.4 “By modeling the membrane potential of the integrate-and-fire neuron with noisy inputs (a valid assumption with circuit noise and/or noisy spike inputs) as a type of Brownian motion, a closed-form expression to relate the input and output firing rates of the neuron was determined, and thus its derivative.” and Serb teaches the biological neurons as outlined above)
Regarding claim 48, Saxena in view of Serb teaches claim 46 as outlined above. Saxena further teaches:
the processor is further configured to count a number of pairs, of the PSPs of the second biological neuron and the APs of the first biological neuron, that have respective firing times within a first threshold, and selectively adjusting a conductance value of the second cross point memory based on the counted number of pairs, to be the second conductance value. (3.2 Paragraph 2 Figure 1-2 "In pair-wise STDP learning, spikes sent from pre- and post-synaptic have their voltage amplitudes below the program and erase switching thresholds (V+th and V−th ) of a bipolar RRAM device. RRAM switching events may occur only if this spike pair overlaps and creates a net potential (Vnet) greater than the switching threshold, as illustrated in Figure 2b,c. Here, for Vnet > V+th , RRAM is incrementally programmed (conductance is increased) causing long-term potentiation (LTP) in the synapse. On the other hand, for the case Vnet < V−th , the RRAM is incrementally erased (conductance is decreased) and long-term depression (LTD) occurs in the synapse. In case of no temporal overlap, the pre-synaptic pulse is integrated in the neuron and thus should have a net positive area and smaller amplitude than the program or erase thresholds. This in turn sets a constraint for the voltage spikes that V−th < Vspk(t) < V+th must always be ensured to avoid disturbing the RRAM state. This scheme effectively converts the time overlap (∆t) of pre and post spikes into program or erase voltage pulses” and Serb teaches the biological neurons as outlined above).
Claims 18 & 32 are rejected under 35 U.S.C. 103 as being unpatentable over Serb in view of Saxena and Poon (US 20110137843 A1).
Regarding claim 18, Serb in view of Saxena teaches claim 1 as outlined above. Neither Saxena nor Serb teach the computer-readable medium. However, Poon does:
A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the method of claim 1. ([0033] “In accordance with another aspect of the present invention, the above method can be associated with a computer-readable medium including instructions for implementing the method.”)
Saxena, Serb and Poon are considered analogous art to the claimed invention because they are in the same field of endeavor being spiking neural networks. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the spiking neural network method of Saxena with the computer-readable medium of Poon.
Regarding claim 32, Serb in view of Saxena teaches claim 24 as outlined above. Neither Saxena nor Serb teach the computer-readable medium. However, Poon does:
A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the method of claim 24. ([0033] “In accordance with another aspect of the present invention, the above method can be associated with a computer-readable medium including instructions for implementing the method.”)
Saxena, Serb and Poon are considered analogous art to the claimed invention because they are in the same field of endeavor being spiking neural networks. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the spiking neural network method of Saxena with the computer-readable medium of Poon.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DANIEL PATRICK GRUSZKA whose telephone number is (571)272-5259. The examiner can normally be reached M-F 9:00 AM - 6:00 PM ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Li Zhen can be reached at (571) 272-3768. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/DANIEL GRUSZKA/Examiner, Art Unit 2121
/Li B. Zhen/Supervisory Patent Examiner, Art Unit 2121
Prosecution Insights