Memristors with diffusive dynamics as synaptic emulators for neuromorphic computing

May combine two claims.

Postsynaptic calcium helps regulate long-term potentiation and depression; calcium accumulation controls CaMKII, calcineurin, and protein phosphatase 1, which persistently alter synaptic strength by changing AMPAR number or conductance.

“The calcium inside the postsynaptic membrane also plays important regulating roles in long-term potentiation and depression where Ca 2+ accumulation is necessary to regulate enzymes, i.e.Ca 2+ /calmodulindependent protein kinase II (CaMKII), calcineurin, and protein phosphatase 1 42,43,46,47 , which in turn triggers rapid and persistent modification of synaptic strengths by changing the number and”

• cites [42] Burgoyne, R. D. Neuronal calcium sensor proteins: generating diversity in neuronal Ca2+ signalling. Nat. Rev. Neurosci. 8, 182–193 (2007). sourcePartly supported

  The 'Different signals, diverse outcomes' section says postsynaptic Ca2+ signals can trigger LTP or LTD, and 'Ca2+ sensor proteins: calmodulin' links Ca2+/calmodulin to CaMKII and identifies calcineurin as a calmodulin target (page numbers unavailable). The paper does not establish the claimed pathway through protein phosphatase 1 or say that these enzymes persistently change synaptic strength by altering AMPAR number or conductance. Its References discuss possible AMPA-receptor endocytosis in LTD through hippocalcin, a different mechanism. (LLM, conf=0.87)

  Experimental — accuracy not yet measured
• cites [43] Clapham, D. E. Calcium signaling. Cell 131, 1047–1058 (2007). sourceNot checked — no full text of the source
• cites [46] Malenka, R. C. & Bear, M. F. LTP and LTD: an embarrassment of riches. Neuron 44, 5–21 (2004). sourcePartly supported

  In the passages headed “LTD Expression and Maintenance” (page unavailable), the review links LTP and LTD to AMPAR phosphorylation and identifies CaMKII as acting on an AMPAR phosphorylation site. Other passages describe postsynaptic Ca²⁺ as a trigger for LTD and report roles for calcineurin and protein phosphatase 1. The retrieved passages support changes in AMPAR surface expression, but do not establish the claim that these enzymes persistently change synaptic strength through altered AMPAR conductance. (LLM, conf=0.78)

  Experimental — accuracy not yet measured
• cites [47] Caporale, N. & Dan, Y. Spike timing-dependent plasticity: a Hebbian learning rule. Annu. Rev. Neurosci. 31, 25–46 (2008). sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

Nanoparticle pinning can arise from interactions with impurities, the substrate, or the ionic lattice.

“Pinning can occur through interactions of the nanoparticles with impurities, the substrate and/or the ionic lattice 41 .”

• cites [41] Yi, W. et al. Quantized conductance coincides with state instability and excess noise in tantalum oxide memristors. Nat. Commun. 7, 11142 (2016). sourceCould not verify

  The Noise spectra section discusses fluctuating atomic-scale contacts in memristors, while Methods mentions a substrate only as part of the device construction. Searches for nanoparticle pinning and interactions with impurities, substrates, or ionic lattices found no passage addressing the claim; our copy therefore does not verify it. The retrieved passages have no page numbers. (LLM, conf=0.76)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

A qualitatively similar simulation result was obtained when the interfacial energy was modelled as particle-particle interactions.

“Moreover, a qualitatively similar result (although requiring considerably more computational resources) has been obtained when we modelled the interfacial energy as an interaction between particles 41 .”

• cites [41] Yi, W. et al. Quantized conductance coincides with state instability and excess noise in tantalum oxide memristors. Nat. Commun. 7, 11142 (2016). sourceCould not verify

  The Discussion (page unavailable) describes a Molecular Dynamics–Langevin simulation that reproduces a crossover in memristor noise spectra. Searches for interfacial energy, particle–particle interactions, and the claimed modelling comparison found no passage in our copy addressing them. (LLM, conf=0.75)

  Experimental — accuracy not yet measured

May combine two claims.

Substantial relaxation occurred in systems with large wetting contact angles, including MgOx:Ag, SiOxNy:Ag, and HfOx:Ag, in line with reports of silver nanoparticle pseudoelasticity and the idea that interfacial energy helps rupture filaments in volatile switching.

“The material systems exhibiting a substantial relaxation were those with large wetting contact angles 32 , such as MgOx:Ag, SiOxNy:Ag, and HfOx:Ag in Fig. 1b, consistent with the reported psudoelasticity of silver nanoparticles 33 and the hypothesis that interfacial energy facilitates filament rupture in volatile switching [34][35][36] .”

• cites [32] Stoneham, A. M. Systematics of metal–insulator interfacial energies: a new rule for wetting and strong catalyst-support interactions. Appl. Surf. Sci. 14, 249–259 (1983). sourcePartly supported

  Split into 2 atoms — 0 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 1 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured
• cites [33] Sun, J. et al. Liquid-like pseudoelasticity of sub-10-nm crystalline silver particles. Nat. Mater. 13, 1007–1012 (2014). sourceNot checked — no full text of the source
• cites [34][35][36]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Opposite-polarity bias produced symmetric hysteresis loops, indicating unipolar threshold switching that differs notably from non-volatile drift memristors, particularly when turning off.

“Symmetric hysteresis loops were observed with the opposite polarity bias (Supplementary Fig. S3), showing that the threshold switching is unipolar in nature and significantly different from non-volatile drift-type memristors, especially in the OFF-switching process.”

May combine two claims.

Like ECM cells, these devices use mobile noble-metal species, but they differ markedly in symmetry, voltage polarity, metal concentration and profile, and transient switching.

“These devices are similar to electrochemical metallization memory (ECM) [21][22][23][24][25][26] cells in terms of utilizing mobile species of noble metals, but they differ substantially in terms of the structural symmetry and operating voltage polarities, metal concentration and profile, and transient switching behaviour.”

• cites [21][22][23][24][25][26]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

XPS showed the Ag in the device was metallic, and HRTEM micrographs confirmed Ag nanocrystals in SiOxNy:Ag.

“X-ray photoelectron spectroscopy (XPS) revealed that the Ag was metallic, which was further confirmed by HRTEM micrographs showing Ag nanocrystals in SiOxNy:Ag”

May combine two claims.

HRTEM and simulations indicate that silver atoms spread when voltage is applied and re-form clusters on their own at zero bias as interfacial energy is minimized.

“In situ high-resolution transmission electron microscopy and nanoparticle dynamics simulations both demonstrate that Ag atoms disperse under electrical bias and regroup spontaneously under zero bias because of interfacial energy minimization,”

May combine two claims.

The on/off resistance ratio reached five orders in SiOxNy:Ag and more than ten orders in HfOx:Ag, which the authors report as the highest for threshold-switching devices to date.

“The resistance ratio between the conducting and insulating states was 5-orders of magnitude in SiOxNy:Ag and over 10-orders in HfOx:Ag devices, the highest reported in threshold switching devices so far [27][28][29][30] .”

• cites [27][28][29][30]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Turn-on slopes for volatile switching were roughly 10 mV per decade in MgOx:Ag and SiOxNy:Ag and about 1 mV per decade in HfOx:Ag, reported as the sharpest so far.

“The volatile switching had sharp turn-on slopes of ~10mV/decade in MgOx:Ag and SiOxNy:Ag, and an extraordinary ~1mV/decade in HfOx:Ag, the sharpest demonstrated to date [27][28][29][30] .”

• cites [27][28][29][30]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Reliable threshold switching was observed under a 100 nA compliance in ambient conditions with a typical relaxation time constant of about 11 ms.

“Reliable threshold switching was observed under a 100nA compliance in ambient conditions with a typical relaxation time constant of ~11ms (Supplementary Fig. S5b-c).”

May combine two claims.

If another pulse comes before the particles are reabsorbed, additional particles enter the gap, so device conductance rises gradually, resembling paired-pulse facilitation.

“If a subsequent pulse arrives before particles are re-absorbed, i.e., if the time between pulses is shorter than the diffusion relaxation time, more particles are pushed into the gap between terminals resulting in a gradual increase in device conductance, similar to the paired-pulse facilitation (PPF) phenomenon in bio-synapses.”

• AssumptionThe authors state this as an assumption: The model treats electron flow as a series of tunnelling hops: input terminal to closest metal particle, particle to particle, and last particle to output terminal.
• AssumptionThe authors state this as an assumption: The tunnelling resistance amplitude is taken to be identical for every island.
• AssumptionThe authors state this as an assumption: Each mobile metal nanoparticle is described by an overdamped Langevin equation with a trapping potential and a random force whose strength depends on the local device temperature.
• AssumptionThe authors state this as an assumption: The simulation model omits redox reactions.

May combine two claims.

Too much stimulation makes conductance inflect, reproducing paired-pulse facilitation followed by paired-pulse depression.

“This results in an inflection of the device conductance due to excessive stimulation, capturing another synaptic behaviour, i.e., PPF followed by PPD (paired pulse depression).”

• AssumptionThe authors state this as an assumption: The model treats electron flow as a series of tunnelling hops: input terminal to closest metal particle, particle to particle, and last particle to output terminal.
• AssumptionThe authors state this as an assumption: The tunnelling resistance amplitude is taken to be identical for every island.
• AssumptionThe authors state this as an assumption: Each mobile metal nanoparticle is described by an overdamped Langevin equation with a trapping potential and a random force whose strength depends on the local device temperature.
• AssumptionThe authors state this as an assumption: The simulation model omits redox reactions.

May combine two claims.

In time, most nanoparticles merge into larger clusters to reduce interfacial energy, restoring the high-resistance state and nearly the original distribution, which explains the observed volatility.

“Eventually, most of the nanoparticles have merged into larger clusters to minimize interfacial energy, and the high resistance state is re-established along with the original particle distribution almost restored (compare Fig. 4a1 and Fig. 4a4), leading to the observed volatility in Fig. 2.”

• AssumptionThe authors state this as an assumption: The model treats electron flow as a series of tunnelling hops: input terminal to closest metal particle, particle to particle, and last particle to output terminal.
• AssumptionThe authors state this as an assumption: The tunnelling resistance amplitude is taken to be identical for every island.
• AssumptionThe authors state this as an assumption: Each mobile metal nanoparticle is described by an overdamped Langevin equation with a trapping potential and a random force whose strength depends on the local device temperature.
• AssumptionThe authors state this as an assumption: The simulation model omits redox reactions.

May combine two claims.

PPF and PPD were shown with diffusive memristors alone and resemble short-term plasticity, since conductance changes from the OFF state fade with time.

“The above PPF and PPD demonstrations were realized with diffusive memristors only, resembling short-term plasticity 45 in synapses because any conductance change from the OFF state of diffusive memristors will vanish over time.”

• cites [45] Zucker, R. S. & Regehr, W. G. Short-term synaptic plasticity. Annu. Rev. Physiol. 64, 355–405 (2002). sourceNot checked — no full text of the source

May combine two claims.

With long enough intervals between high-voltage pulses, a conducting bridge may form first, but it breaks before the next pulse and particles return to the terminals.

“Third, sequential high voltage pulses with a long enough interval (low frequency) may form a conducting bridge first, but before the next pulse arrives the bridge breaks and the particles are re-absorbed at the terminals.”

• AssumptionThe authors state this as an assumption: The model treats electron flow as a series of tunnelling hops: input terminal to closest metal particle, particle to particle, and last particle to output terminal.
• AssumptionThe authors state this as an assumption: The tunnelling resistance amplitude is taken to be identical for every island.
• AssumptionThe authors state this as an assumption: Each mobile metal nanoparticle is described by an overdamped Langevin equation with a trapping potential and a random force whose strength depends on the local device temperature.
• AssumptionThe authors state this as an assumption: The simulation model omits redox reactions.

May combine two claims.

τr decreased as ambient temperature increased, consistent with a diffusion activation energy of 0.27 eV.

“As shown in Fig. 3b, τr decreased as the ambient temperature increased, consistent with a diffusion activation energy of 0.27eV (inset of Fig. 3b)”

May combine two claims.

After power is removed, temperature falls and nanoparticles begin to coalesce, diffusing slowly toward minimum-energy positions near the terminals.

“As soon as the power is turned off, the temperature drops, and the nanoparticles start to coalesce (Fig. 4a3), i.e. particles slowly diffuse to their minimum energy positions near the device terminals.”

• AssumptionThe authors state this as an assumption: The model treats electron flow as a series of tunnelling hops: input terminal to closest metal particle, particle to particle, and last particle to output terminal.
• AssumptionThe authors state this as an assumption: The tunnelling resistance amplitude is taken to be identical for every island.
• AssumptionThe authors state this as an assumption: Each mobile metal nanoparticle is described by an overdamped Langevin equation with a trapping potential and a random force whose strength depends on the local device temperature.
• AssumptionThe authors state this as an assumption: The simulation model omits redox reactions.

May combine two claims.

Applying a voltage pulse raises local temperature through Joule heating and tilts the potential via electric forces on induced-charge particles, which breaks up larger clusters.

“When a voltage pulse is applied (Fig. 4a), the local temperature increases due to Joule heating and the potential is tilted by electric forces acting on particles with induced charge, both of which cause larger clusters to break up.”

• AssumptionThe authors state this as an assumption: The model treats electron flow as a series of tunnelling hops: input terminal to closest metal particle, particle to particle, and last particle to output terminal.
• AssumptionThe authors state this as an assumption: The tunnelling resistance amplitude is taken to be identical for every island.
• AssumptionThe authors state this as an assumption: Each mobile metal nanoparticle is described by an overdamped Langevin equation with a trapping potential and a random force whose strength depends on the local device temperature.
• AssumptionThe authors state this as an assumption: The simulation model omits redox reactions.

May combine two claims.

More uniform nanoparticle distribution in the gap lowers resistance; current and temperature then rise, and positive feedback creates a conductive channel.

“As the nanoparticles become more uniformly distributed in the gap, the resistance drops, the current and temperature increase, and a positive feedback results in the formation of a conductive channel (Fig. 4a2).”

• AssumptionThe authors state this as an assumption: The model treats electron flow as a series of tunnelling hops: input terminal to closest metal particle, particle to particle, and last particle to output terminal.
• AssumptionThe authors state this as an assumption: The tunnelling resistance amplitude is taken to be identical for every island.
• AssumptionThe authors state this as an assumption: Each mobile metal nanoparticle is described by an overdamped Langevin equation with a trapping potential and a random force whose strength depends on the local device temperature.
• AssumptionThe authors state this as an assumption: The simulation model omits redox reactions.

May combine two claims.

Smaller ∆t gives lower diffusive memristor resistance and a larger drift memristor resistance change, and the reverse for larger ∆t.

“A smaller ∆t corresponds to a smaller diffusive memristor resistance and results in a greater resistance change in the drift memristor and vice versa (Fig. 6d).”

May combine two claims.

This facilitation-to-depression inflection had not been clearly shown before in two-terminal devices; it was predicted in Fig. 4b and seen experimentally in Fig. 5c.

“This important feature of bio-synapses, which has not been clearly demonstrated previously on two terminal devices, was predicted in Fig. 4b and observed experimentally in Fig. 5c.”

May combine two claims.

The elongated nanoparticle cluster that probably formed the conductive bridge shrank quickly from 14.8 nm long to a circular shape 7.6 nm across by 5.7 s, which indicates Ostwald ripening.

“The elongated cluster of nanoparticles that had likely formed the conductive bridge rapidly contracted from a length of 14.8nm to a circular profile with a diameter of 7.6nm by 5.7s, indicating Ostwald ripening 31 .”

• cites [31] Voorhees, P. W. The theory of Ostwald ripening. J. Stat. Phys. 38, 231–252 (1985). sourceNot checked — no full text of the source

May combine two claims.

From steady state, the device shows PPD under 196 Hz stimulation and current rises (facilitation) when frequency is increased to 5000 Hz.

“The device in its steady state shows PPD upon low frequency (196Hz) stimulation and experiences an increase in current (facilitation) once the stimulation frequency is raised (5000Hz).”

May combine two claims.

Shorter interpulse intervals produced larger diffusive-memristor conductance increases, so more voltage dropped across the drift memristor, switching it through the voltage-divider effect.

“Similar to Fig. 5b, a shorter tzero resulted in a greater increase in the conductance of the diffusive memristor and thus a larger voltage drop across the drift memristor, which thereby switched due to the voltage divider effect.”

Probably not a substantial claim: may not pass S3 (load-bearing).

At 4.6 s, the Ag cluster reached a diameter of about 4.2 nm.

“At 4.6s, the cluster reached a diameter of ~4.2nm.”

May combine two claims.

Longer interpulse intervals caused smaller diffusive-memristor conductance increases, so less voltage dropped across the drift memristor and its resistance change was smaller or undetectable.

“A longer tzero resulted in a smaller increase in the diffusive memristor conductance and thus a smaller voltage drop across the drift memristor, leading to a smaller or non-detectable resistance change in the drift memristor.”

May combine two claims.

Once a channel formed, current rose abruptly by several orders of magnitude and then grew slowly with bias as the channel thickened.

“Upon channel formation, the current jumped abruptly by several orders of magnitude, and then slowly increased further under bias as the channel thickened.”

May combine two claims.

The first spike does not switch the drift memristor because most voltage falls across the diffusive memristor and turns it on first.

“The drift memristor is not switched by the first spike, because the majority of the voltage drops across the diffusive memristor and turns it ON first.”

May combine two claims.

On the first voltage pulse, field-assisted diffusion moves some silver particles out of the left cluster and they begin to bridge the terminals.

“First, when the initial voltage pulse is applied, electric field-assisted diffusion pumps some of the Ag-particles out of the 'left' cluster and they start to bridge terminals.”

• AssumptionThe authors state this as an assumption: The model treats electron flow as a series of tunnelling hops: input terminal to closest metal particle, particle to particle, and last particle to output terminal.
• AssumptionThe authors state this as an assumption: The tunnelling resistance amplitude is taken to be identical for every island.
• AssumptionThe authors state this as an assumption: Each mobile metal nanoparticle is described by an overdamped Langevin equation with a trapping potential and a random force whose strength depends on the local device temperature.
• AssumptionThe authors state this as an assumption: The simulation model omits redox reactions.

May combine two claims.

When the pre-spike precedes the post-spike, drift memristor conductance rises (potentiation); when it follows, depression occurs.

“If the pre-spike appears before the postspike, the drift memristor conductance increases (potentiation).If the pre-spike follows the post-spike, depression occurs.”

May combine two claims.

In the combined element, the diffusive memristor has much higher resistance than the drift memristor when OFF and much lower resistance when ON.

“In the combined element, the resistance of the diffusive memristor in its OFF state is much larger than that of the drift memristor, while the resistance of its ON state is much smaller than that of the drift memristor.”

May combine two claims.

In the HRTEM experiment, a delay of roughly 2 s occurred, during which current stayed below 5 nA and silver nanoparticles appeared in the gap.

“We observed a delay time of ~2s during which the measured current was <5nA and Ag nanoparticles formed in the gap region”

May combine two claims.

In situ HRTEM first revealed the microscopic basis of threshold switching and relaxation in the diffusive memristor, and nanoparticle simulations explained it.

“The microscopic nature of both the threshold switching and relaxation of the diffusive memristor is revealed for the first time by in situ HRTEM and explained by nanoparticle dynamics simulation.”

May combine two claims.

The simulated response to a voltage pulse shows a delayed response and gradual relaxation to the low-conducting state, consistent with the experimental threshold-switching property in Fig. 3a.

“curve) induced by a voltage pulse (red curve) shows a delay in the response and a gradual relaxation towards the low conducting state, consistent with the experimental threshold switching property in Fig. 3a.”

May combine two claims.

When the post-spike precedes the pre-spike, the device is depressed; when the pre-spike precedes the post-spike, it is potentiated.

“When the post-spike precedes the pre-spike, the device is reset (depressed), and when the prespike precedes the post-spike, the device is set (potentiated).”

May combine two claims.

Figure 4 shows relaxation on time scales of about 10-20 1/κ, indicating that the simulated system is governed by diffusion rather than temperature relaxation.

“As one can see from Fig. 4, the relaxation occurs on time scales of the order 10-20 1/κ, thus, the simulated system is clearly governed by diffusion rather than the temperature relaxation 19 .”

• cites [19] Kim, S. et al. Experimental demonstration of a second-order memristor and its ability to biorealistically implement synaptic plasticity. Nano Lett. 15, 2203–2211 (2015). sourceNot checked — no full text of the source

May combine two claims.

In Figure 4b, the simulated conductance reached 25% of its maximum after 4 pulses and about 75% after 7 pulses.

“from Fig 4b we conclude that the conductance reached 25% of the maximum after 4 pulses and about 75% after 7 pulses.”

May combine two claims.

A combined device made of a diffusive and a drift memristor shows spike-rate-dependent plasticity and true spike-timing-dependent plasticity.

“Device current response (blue) to multiple dependent plasticity (SRDP and true STDP) behaviour of a combined device consisting of a diffusive and a drift memristor.”

May combine two claims.

In each spike, the long low-voltage pulse turns the diffusive memristor on, and the short high-voltage pulse switches the drift memristor.

“The long low voltage pulse in each spike turns the diffusive memristor ON, and the short high voltage pulse switches the drift memristor.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The device did not fail during measurement, indicating high robustness.

“The device did not fail during the measurement, demonstrating high robustness.”

May combine two claims.

Diffusive memristors exhibit highly non-linear, fast, and repeatable threshold switching.

“Highly non-linear, fast, and repeatable threshold switching behaviours of diffusive memristors”

May combine two claims.

The device has a finite turn-on delay and a finite relaxation time before returning to the high-resistance state after the switching pulse is removed.

“The device requires a finite delay time to turn ON and has a finite relaxation time before it goes to the high resistance state after the switching pulse is removed.”

May combine two claims.

For long tzero, the diffusive memristor's conductance change is smaller, leading to a smaller weight change in the drift memristor.

“For long tzero, the change in the conductance of the diffusive memristor is lower (see Fig. 5b), resulting in a lower weight change of the drift memristor.”

May combine two claims.

After external bias is removed at 5.0 s, the filament deforms and shrinks into a round spherical nanocluster, implying an interfacial-energy-driven diffusion mechanism.

“When the external biasing is removed at 5.0s, the filament starts to deform, shrinking to a round spherical nanocluster implying an interfacial energy driven diffusion mechanism.”

May combine two claims.

Over the studied range, the characteristic single hopping time scale changed from about 2/κ to 6/κ, and the diffusion relaxation time was about 20/κ to 60/κ.

“In this range the characteristic single hopping time scale 𝑒𝑒 𝑤𝑤 𝑝𝑝 /𝑘𝑘 𝐵𝐵 𝑇𝑇 changed from about 2/κ to 6/κ and the diffusion relaxation time (estimated as a characteristic jump time multiplied by the number of jumps to reach the interfacial minimum near the device terminal) was about 20/κ to 60/κ.”

May combine two claims.

CMOS circuits have mimicked synaptic calcium dynamics, yet three-terminal devices match biological counterparts poorly in mechanism and require many complex circuits for synaptic behavior.

“CMOS circuits have been employed to mimic synaptic Ca 2+ dynamics, but three-terminal devices bear limited resemblance to bio-counterparts at the mechanism level and require significant numbers and complex circuits to simulate synaptic behavior [1][2][3] .”

• cites [1][2][3]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

Various drift-type memristors have been used in neuromorphic architectures.

“Various types of memristors based on ionic drift (drift-type memristor) [4][5][6][7][8] have recently been utilized for this purpose in neuromorphic architectures [9][10][11][12][13][14][15] .”

• cites [4][5][6][7][8]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [9][10][11][12][13][14][15]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Lu and co-workers approximated chemical-synapse calcium dynamics with second-order drift memristors using thermal dissipation or mobility decay, and showed STDP with non-overlapping spikes plus other synaptic functions.

“Recently, Lu's group adopted second-order drift memristors to approximate the Ca 2+ dynamics of chemical synapses by utilizing thermal dissipation 19 or mobility decay 20 , which successfully demonstrated STDP with non-overlapping spikes and other synaptic functions”

• cites [19] Kim, S. et al. Experimental demonstration of a second-order memristor and its ability to biorealistically implement synaptic plasticity. Nano Lett. 15, 2203–2211 (2015). sourceNot checked — no full text of the source
• cites [20] Du, C., Ma, W., Chang, T., Sheridan, P. & Lu, W. D. Biorealistic implementation of synaptic functions with oxide memristors through internal ionic dynamics. Adv. Funct. Mater. 25, 4290–4299 (2015). sourceNot checked — no full text of the source

May combine two claims.

That approach is simple and repeatable, yet its dynamics depart markedly from real synapses, which restricts the fidelity and range of synaptic functions.

“This approach features repeatability and simplicity, but the significant differences of the dynamical response from actual synapses limit the fidelity and variety of desired synaptic functions.”

• cites [19] Kim, S. et al. Experimental demonstration of a second-order memristor and its ability to biorealistically implement synaptic plasticity. Nano Lett. 15, 2203–2211 (2015). sourceNot checked — no full text of the source
• cites [20] Du, C., Ma, W., Chang, T., Sheridan, P. & Lu, W. D. Biorealistic implementation of synaptic functions with oxide memristors through internal ionic dynamics. Adv. Funct. Mater. 25, 4290–4299 (2015). sourceNot checked — no full text of the source

May combine two claims.

MOS-based memristor emulators have similar issues, though they can simulate various synaptic functions including STDP.

“Similar issues also exist in MOS-based memristor emulators [16][17][18] , although they are capable of simulating a variety of synaptic functions including spike-timing-dependent plasticity (STDP).”

• cites [16][17][18]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Drift memristors have demonstrated qualitative synaptic behavior, but their memory-optimized fast switching and non-volatility fail to reproduce plasticity faithfully.

“Although qualitative synaptic functionality has been demonstrated, the fast switching and non-volatility of drift memristors optimized for memory applications do not faithfully replicate the nature of plasticity.”

Classified as probably not a substantial claim. May combine two claims.

Calcium dynamics initiates both short- and long-term synaptic plasticity and underlies memory and learning.

“Ca 2+ dynamics is responsible for initiating both short-and long-term plasticity of synapses, forming the basis of memory and learning [42][43][44] .”

• cites [42][43][44]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The exact shapes of the interfacial and pinning potentials are not essential; the model only requires two distinct energy scales and a weaker pinning potential with many wells.

“The particular shape of neither the interfacial potential nor the pinning energy profile are essential; the only property that matters is that the potential 𝑈𝑈 has two different energy scales and the weaker pinning potential has many wells (for example it can be random with fluctuating 𝑤𝑤 𝑝𝑝 and 𝑖𝑖 𝑝𝑝 ).”

• AssumptionThe authors state this as an assumption: The model's potential energy landscape has two distinct energy scales: a stronger interfacial term that forms large metallic clusters near the device terminals and a weaker pinning term with many wells between electrodes.

May combine two claims.

The diffusive memristor's dynamics supply an intrinsic timing mechanism, so SRDP and STDP need neither complex pulse shaping nor overlapping spikes.

“Because the dynamics of the diffusive memristor provides an intrinsic timing mechanism for the combined element, the spike-ratedependent plasticity and STDP do not require complex pulse engineering or spike overlapping.”

May combine two claims.

Silver dynamics in diffusive memristors functionally resemble synaptic calcium behavior and enable direct, natural emulation of several short- and long-term plasticity functions, including PPF, PPD, PPD after PPF, SRDP, and STDP.

“The Ag dynamics of the diffusive memristors functionally resemble the synaptic Ca 2+ behavior in chemical synapses and lead to a direct and natural emulation of multiple synaptic functions for both short-term and long-term plasticity, such as PPF, PPD, PPD following PPF, SRDP and STDP.”

May combine two claims.

The diffusive memristor can also act as a selector with strong transient nonlinearity, which is important for large crossbar neural-network arrays.

“In addition to providing a synapse emulator, the diffusive memristor can also serve as a selector with a large transient nonlinearity that is critical for the operation of a large crossbar array as a neural network.”

May combine two claims.

The combination of electrical bias and diffusion causes the device to switch to a high-conducting state and depletes nanoparticles at one terminal.

“The electrical bias together with diffusion is responsible for switching to a high conducting state and depletion of nanoparticles at one of the device terminals.”

• AssumptionThe authors state this as an assumption: The model's potential energy landscape has two distinct energy scales: a stronger interfacial term that forms large metallic clusters near the device terminals and a weaker pinning term with many wells between electrodes.

May combine two claims.

This lowers circuit and algorithm design complexity and permits low-energy operation.

“This substantially reduces the complexity of both circuit and algorithm design and enables lowenergy operations.”

May combine two claims.

Stronger thermal fluctuations increase Joule heating and vice versa, breaking up large Ag clusters into a string of smaller nanoparticles that forms a conductive path between terminals, as observed in HRTEM experiments.

“The stronger fluctuations support larger Joule heating and vice versa, breaking up the large Ag clusters to form a string of smaller nanoparticles and thus a conductive path between the terminals, as observed experimentally in the HRTEM studies.”

May combine two claims.

Ag diffuses into the gap between Ag nanoclusters with field assistance, and the Ag filament is cleared from the gap by interfacial energy or mechanical stress after the voltage is removed, closely emulating Ca2+ dynamics.

“diffusion into the gap region between Ag nanoclusters with field assistance and clearance of Ag filament from the gap region by the interfacial energy or mechanical stress once the voltage signal is removed, which is a close emulation of Ca 2+ dynamics.”

Probably not a substantial claim: may not pass S4 (based).

Equations (4) and (5) are sufficient to describe the memristor dynamics under the current imposed by its embedding circuit.

“Equations ( 4) and ( 5) are sufficient to describe the memristor dynamics subject to the current through the device, which is controlled by the electric circuit in which it is embedded.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Programming is at the instrument noise level (<100 nA) for the current measurement range used.

“programming is at the instrument noise level (<100nA) for the current measurement range used.”

Classified as probably not a substantial claim. May combine two claims.

The normalization quantities can be obtained from simulations averaged over 30 realizations and from experimental I-V hysteresis loops, directly linking simulation and experiment units.

“These quantities can be easily obtained from both simulations (averaged over 30 realizations) and experimental I-V hysteresis loops, thus directly linking the units in the simulations and the experiments.”

May combine two claims.

The diffusion time scale depends on temperature, the potential profile, and electric drift, so it varies with voltage-pulse intensity and temperature.

“The first one depends on temperature, potential profile, and electric drift, thus, this time scale changes depending on voltage pulse intensity and temperature.”