{"id":"1708bec5-5ac3-4169-9a57-2f330de0f357","arxiv_id":"1908.08602","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"PbS nanoplatelets on InP exhibit single-electron tunneling with short- and long-term memory, and the authors model them as low-energy, MHz-speed neuromorphic building blocks.","lead":"This paper shows that tiny lead-sulfide crystals on indium-phosphide can trap single electrons, producing staircase-like current steps and memory effects. The authors argue these structures could act as extremely low-energy artificial synapses, estimating about 0.2 femtojoules per operation, but that figure comes from a model extrapolation.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 5 MHz and 0.2 fJ claims rest on an unmeasured 200 ns write-pulse width that is extrapolated from a fitted impulse-response model, so a direct pulse-width measurement is needed to validate them.","rationale":"The central quantitative claims of the paper, the ~0.2 fJ energy per operation and 5 MHz maximum frequency, are extrapolations from a fitted SET model rather than direct measurements. The paper's own Section 2 describes the plateau as a trap-limited two-step process, and Section 3 treats the same plateau as Coulomb blockade; the discrimination is not experimentally tested. More importantly for the numeric headlines, the 200 ns minimum write pulse is a component of an assumed impulse-response shape that is fit to integrated current data, not resolved in time. No pulse-width-sweep experiment is shown, so the assertion that shorter pulses have no effect is unjustified. The equivalent-circuit capacitance C2 is also unexpectedly small, suggesting the fit may be absorbing unknown conduction physics. These issues do not invalidate the observed memory and memory-fading phenomena, which are interesting and appear reproducible, nor do they rule out the SET interpretation; they mean the specific energy and speed figures should be treated as conditional until direct pulse measurements are made. The proposed test would settle whether the 200 ns write-pulse assumption is physical. Because the reader's verdict already conditions acceptance on addressing this gap, the verdict should remain unchanged.","tokens_in":8777,"tokens_out":10211,"duration_ms":102695,"concrete_test":"Perform a direct pulse-width-dependence measurement on the same PbS/InP junction at 77 K with a high-bandwidth current preamplifier. Apply a fixed-amplitude write pulse (e.g., +0.3 V) with widths logarithmically spaced from 50 ns to 10 µs, followed by a fixed probe pulse to read the device conductance; repeat the sequence after resetting the device with a negative bias pulse. If a measurable, reproducible conductance change appears only for pulse widths substantially longer than 200 ns (i.e., >1 µs), or if the recorded current transient shows the preamplifier's rise time rather than an intrinsic 200 ns constant current plateau, then the estimated 5 MHz frequency and 0.2 fJ energy are not supported by the evidence. A complementary check is to record the current transient directly at 100 MHz bandwidth to verify the 200 ns plateau.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing assumption is in Section 4: the minimum write-pulse width is set to 'approximately 200 ns' and is used to compute a 5 MHz maximum frequency and a 0.2 fJ Joule energy per synaptic operation. This 200 ns value is not a directly measured quantity. In Section 3, it is one of four parameters (700 pA, 1 pA, 200 ns, 4 ms) obtained by modelling the integrated current response as a fast constant high-current component followed by an exponential decay (Figure 4b). The source data in Figures 3(a) and 3(c) are voltage ramps of 0.07-4.6 s and current samples taken after delays, so the 200 ns timescale is below the explicit time resolution of the measurements and is only recovered through a fitting/deconvolution step. If this fast component is an artifact of the current preamplifier bandwidth or of the assumed two-component model, the claim that pulses shorter than 200 ns have no effect is unsupported. The energy and speed headlines would then be invalid. An additional risk is that the fitted capacitance C2 = 2e-19 F is about three orders of magnitude smaller than a parallel-plate estimate for a 100-nm nanoplatelet with a ~1-nm oxide, so the fitted SET parameters may absorb other physics; however, the 200 ns pulse-width assumption is the single most consequential one for the quantitative claims.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports scanning tunneling spectroscopy measurements on PbS nanoplatelets grown on p-type InP with a self-formed oxide layer, observing a voltage plateau in I-V curves that shifts with repeated scans, a current that depends on acquisition time and pulse delay, and memory-like relaxation behavior. The authors model the junction as a double-junction single-electron tunneling (SET) circuit, extract capacitances and resistances from a least-squares fit of the I-V curves, and fit the time-dependent current response to a fast constant component followed by an exponential decay. From this they extrapolate a 200 ns minimum pulse width for conductance change, a 5 MHz maximum operating frequency, and a 0.2 fJ minimum Joule energy per synaptic operation. They further simulate short- and long-term plasticity and a spiking-neuron response using the fitted model. The central quantitative claims therefore rest on the fitted parameters and the 200 ns timescale.","tokens_in":9144,"tokens_out":5015,"duration_ms":43687,"significance":"The qualitative observations, including a reproducible plateau, threshold shifting, and memory-fading behavior in a scalable liquid-phase-grown nanostructure, are of potential interest for neuromorphic device research. If the quantitative energy and speed claims were validated by direct measurements, they would be a strong selling point for ultra-low-energy synaptic hardware. Credit is due for providing real experimental data and for testing the data against a SET equivalent-circuit framework. However, the load-bearing quantitative predictions (0.2 fJ, 5 MHz, 200 ns) are not directly measured, lack error bars, and are derived from a fitting model described only verbally. The plasticity demonstrations are simulations, not experimentally validated predictions. These issues undermine the paper's headline claims as currently presented.","major_comments":[{"comment":"The 200 ns minimum pulse width underlying the 5 MHz and 0.2 fJ claims is not directly measured; it is recovered from fitting the integrated current response to a two-component model (700 pA constant high-current response for 200 ns followed by a 4 ms exponential decay), whereas the raw data in Figure 3(a) and 3(c) are acquired on timescales of 0.07–4.6 s and ~30 ms respectively. The paper provides no error analysis, no model-selection justification, and no control experiment (e.g., short-pulse measurements or characterization of the current preamplifier bandwidth) to demonstrate that the 200 ns feature is physical rather than an artifact of the deconvolution. Consequently, the assertion that pulses shorter than 200 ns have no effect on the conductance is unsupported.","section":"Section 3 (Figure 4b) and Section 4"},{"comment":"The minimum energy estimate of 0.2 fJ is computed as the Joule energy for a 1 V, 1 nA pulse of 200 ns duration, using the steady-state plateau current from Figure 3(a). However, the fitted impulse response in Section 3 gives a constant high-current response of 700 pA, not 1 nA, and the manuscript does not justify why the plateau current should apply to a 200 ns write pulse. The paper should either directly measure the energy of a switching event or explicitly state the assumptions and discuss how the estimate changes if the operating point or the pulse current is different.","section":"Section 4"},{"comment":"The SET model fitting is described only verbally: the text states that a combination of diode curves and sigmoid functions is used and that C1, R1, C2, R2 are extracted from the plateau onset, width, slope, and edge curvature. No equations, confidence intervals, or goodness-of-fit metrics are provided. The extracted C2 = 2×10^-19 F is about three orders of magnitude smaller than a parallel-plate estimate for a ~100-nm nanoplatelet with a ~1-nm oxide, and the explanations (dead layers, depletion) are qualitative. Without a reproducible fitting procedure and error estimates, the physical interpretation of the fitted parameters as evidence of single-electron tunneling is not established, and alternative trap-limited conduction mechanisms are not excluded.","section":"Section 3"},{"comment":"The plasticity demonstrations in Figure 5 are simulations performed using the fitted model parameters, yet the text presents them as if they were predictions of the physical device, blurring the line between measurement and extrapolation. The distinction between measured data and model output should be made explicit in the main text and in the figure captions; otherwise, the experimental evidence for short- and long-term plasticity is overstated.","section":"Section 3 and Figure 5"}],"minor_comments":[{"comment":"The abstract and introduction state that the minimum energy is \"of the order of 1 fJ,\" while Section 4 and the conclusion state 0.2 fJ; this numerical inconsistency should be reconciled.","section":"Abstract / Introduction / Conclusion"},{"comment":"The term \"Quantum Blockade\" is used with reference [6], which is titled \"Coulomb blockade of single-electron tunneling\"; please standardize the terminology to \"Coulomb blockade.\"","section":"Section 3"},{"comment":"There is a typo: \"The sample is negatively bias at V0\" should be \"The sample is negatively biased at V0.\"","section":"Section 2"},{"comment":"There is a stray Chinese full-width comma in the sentence \"Moreover，there is a memory fading equivalence\"; please replace it with a standard comma.","section":"Section 2"},{"comment":"The statement \"Pulses shorter than this will not have any effect\" is too strong, as no such short pulses were applied; suggest softening to \"are predicted to have no effect according to the fitted model.\"","section":"Section 4"}],"recommendation":"major_revision","confidential_remarks":"The paper's scientific value hinges on its quantitative energy and speed claims, but these are extrapolations from a fitted model without direct experimental support. In revision, the authors should either add direct measurements of short-pulse behavior or substantially downgrade the 0.2 fJ and 5 MHz claims to speculative estimates clearly labeled as model-based. The qualitative memory and thresholding observations are interesting, but the current presentation overstates the strength of the evidence for the SET mechanism and for plasticity."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper deserves a serious referee, but not for the numbers in the abstract. The genuinely new thing here is the specific combination: PbS nanoplatelets grown on InP, a self-formed oxide at the interface, and transport showing a Coulomb staircase plus short- and long-term memory. That combination is not in the earlier PbS SET work or in the neuromorphic SET proposals, and the fabrication and characterization are careful. The STEM/EDS evidence for the oxide layer is solid, and the trap-mediated explanation for the plateau shift is plausible.\n\nWhat works well is the experimental phenomenology. The voltage-shifting plateau, the dependence on set-point current and acquisition time, and the time-resolved decay are all real observations. The double-junction SET model fit in Figure 4a captures the I-V shape, and the resistance values are consistent with the 4 kΩ condition. The plasticity demonstrations in Figure 5 are at least internally consistent with the fitted model.\n\nThe soft spots are all in the extrapolation from fit to headline metrics. The 200 ns minimum pulse width is not measured. It comes from a two-component model of the integrated current response, where the fast component is below the explicit time resolution of the voltage-ramp and delay-time measurements. If that fast component is a preamplifier artifact or an artifact of the assumed model, the 5 MHz frequency ceiling and the 0.2 fJ energy claim have no basis. The abstract says 1 fJ while the text says 0.2 fJ; that inconsistency needs fixing. Also, the fitted C2 = 2e-19 F is roughly three orders of magnitude below a naive parallel-plate estimate, and the hand-waving about dead layers and perimeter depletion is not enough to make the value credible. Finally, Figures 5c-f are simulations run with the same fitted parameters, so they do not constitute independent prediction or validation; they are demonstrations of what the model can do.\n\nNone of this undermines the core experimental result. It does mean the paper should be framed as a building-block demonstration with tentative extrapolations, not as a device that already achieves sub-fJ per operation. A referee should ask for direct pulse-width measurements at the nanosecond scale, error bars on the fitted parameters, and an honest reconciliation of the two energy numbers.\n\nBottom line: send it to peer review. With a revision that either validates the 200 ns pulse or drops the quantitative claims, it would be a useful contribution for people working on low-power synaptic devices.","headline":"Solid experimental building-block paper; the 0.2 fJ and 5 MHz claims rest on a fitted 200 ns parameter and need direct pulse-width data before they're cited.","tokens_in":9705,"tokens_out":1536,"would_cite":false,"duration_ms":17333,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A single PbS/InP junction exhibits Coulomb blockade and trap-based memory, making it a candidate for ultra-low-energy neuromorphic synapses.","keywords":["single-electron tunneling","Coulomb blockade","PbS nanoplatelets","InP","neuromorphic computing","synaptic plasticity","short-term memory","trap-mediated transport"],"falsifier":"Measure a single PbS/InP junction with pulses shorter than 200 ns and check whether its subsequent probe current still changes; also take the same I-V data at room temperature and at 77 K. If a sub-200 ns pulse produces a persistent conductivity change, or if the Coulomb staircase disappears when the stated resistance and temperature conditions are violated, the central energy and speed claim fails.","tokens_in":8613,"feed_emoji":"🧠","tokens_out":5325,"duration_ms":51136,"temperature":0.7,"pith_summary":"This paper argues that a single PbS nanoplatelet sitting on p-type InP, with its self-formed oxide interface, behaves as a single-electron tunneling device and as a synaptic element. The same junction that shows a Coulomb-blockade plateau also shows memory, whose strength fades over seconds to hundreds of seconds, giving short- and long-term plasticity. Fitting the current-voltage curves to a double-junction SET circuit yields junction capacitances of about 0.2 to 0.5 aF and resistances of 4 to 20 GΩ. Extrapolating the fitted model, the authors estimate a minimum write pulse of about 200 ns, corresponding to a 5 MHz maximum operating frequency and an energy cost of roughly 0.2 fJ per synaptic operation. If those estimates hold, the structure is a candidate building block for large-scale, liquid-phase-grown neuromorphic networks.","feed_headline":"PbS/InP synapse learns on single electrons at 0.2 fJ","feed_subtitle":"Coulomb blockade plus oxide-trap memory yields short- and long-term plasticity, with a projected 5 MHz operating speed.","key_machinery":"The load-bearing object is a double tunneling junction: the STM tip/vacuum/PbS nanoplatelet junction in series with the PbS/self-formed oxide/p-InP junction, each modeled as a resistor in series with a parallel resistor-capacitor pair. Fitting steady-state I-V curves gives $C_1 = 5 \\times 10^{-19}$ F, $R_1 = 20$ GΩ for the tip junction and $C_2 = 2 \\times 10^{-19}$ F, $R_2 = 4$ GΩ for the oxide junction, with a single-electron charging energy $E_e = e^2/(2C)$. The trap states inside the oxide are the memory: electrons trapped during positive bias raise the plateau threshold until they detrap, producing the observed memory fading. This SET-plus-trap mechanism is what converts a simple rectifying junction into a device with non-linear threshold behaviour, short- and long-term plasticity, and a predicted ultra-low switching energy.","core_discovery":"The central claim is that the amorphous oxide layer between PbS and InP acts as a Coulomb island with trap states, so electron transport is a two-step process: tunneling from the STM tip through the nanoplatelet into an oxide trap, then out of the trap. This creates a Coulomb-blockade staircase with a single observable step, whose onset shifts with repeated scans because charge accumulates in oxide traps and detraps slowly. The authors show the same physical effect produces both long-term depression under 100 ms pulse trains and short-term depression after a single 1 ms pulse, and they emulate a spiking neuron by feeding random weighted pulses through the measured transfer function. The paper's quantitative claim is that the fitted SET parameters imply a 200 ns minimum write pulse, a 5 MHz operating ceiling, and about 0.2 fJ dissipated per operation, compared with roughly picojoule-scale biological synapses.","pith_inferences":["If the 200 ns plateau is a genuine minimum write pulse rather than a measurement artifact, then read/write separation in an array would require write pulses near 200 ns and shorter read pulses, which may create a speed-accuracy tradeoff not discussed in the paper.","Because the oxide layer is inhomogeneous and its thickness varies under each nanoplatelet, device-to-device variation in trap density could produce a population of synapses with different thresholds and time constants, which a network could exploit as a form of stochastic or heterogeneous computation.","The condition $E_e > k_B T$ suggests room-temperature operation is plausible in principle, but all transport measurements were taken at 77 K; a direct room-temperature I-V test would be a natural follow-up.","The trap-mediated SET mechanism implies that the memory is volatile by construction; using it for long-term storage would require periodic refresh or a second, non-volatile mechanism, which the paper does not explore."],"forward_implications":["The same junction can serve as both a weighted synapse and a spiking neuron in an emulated neural network, since voltage pulses change its conductivity and threshold while probe pulses read the state.","Arrays of such devices could be grown in liquid phase and stacked vertically, providing a scalable route to neuromorphic hardware without lithographic single-electron transistors.","The device's memory-fading gives natural short-term plasticity, with time constants near 1 s and 100 s, which can be used for temporal signal processing.","The projected energy and speed of a synaptic operation, about 0.2 fJ and 5 MHz, sit below the roughly picojoule and kilohertz figures of biological synapses."],"supporting_citations":[{"why":"Prior structural and electrical characterization of the PbS/InP heterostructure that establishes the inhomogeneous self-formed oxide layer the memory mechanism depends on.","marker":"[24]"},{"why":"Shows that repeated scanning tunneling spectra shift the onset voltage when charge traps are occupied, the effect the paper reads as memory.","marker":"[25]"},{"why":"Provides the single-quantum-level transport result used to justify the two-step trap-mediated tunneling current plateaus.","marker":"[29]"},{"why":"Supplies the double-junction single-electron tunneling model used to fit the I-V curves.","marker":"[30]"},{"why":"Supplies the simulation framework connecting wave functions to current-voltage characteristics for Coulomb blockade devices.","marker":"[31]"},{"why":"Shows how to extract junction capacitance and resistance values from scanning-tunneling-microscope Coulomb blockade data, the fitting method used here.","marker":"[32]"},{"why":"Gives the biological synapse energy and speed figures (picojoules, kilohertz) that the paper compares its projected 0.2 fJ and 5 MHz against.","marker":"[23]"},{"why":"Explains why nanoscale junction capacitance need not scale with area, used to defend the small fitted $C_2$ value.","marker":"[33]"}],"fun_headline_variants":["PbS/InP single-electron tunneling yields neuromorphic plasticity","Coulomb staircase in PbS/InP enables 0.2 fJ synaptic operation","Single-electron memory and plasticity in PbS/InP junctions","Neuromorphic PbS/InP building blocks with MHz speed and fJ energy","SET-based PbS/InP synapse: short-term and long-term plasticity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The energy and frequency estimates assume that the fitted 200 ns constant-current response is the shortest pulse that changes the device conductivity and that the fitted capacitances and resistances accurately represent the physical junctions; if the plateau comes from trap-limited conduction rather than Coulomb charging, those numbers would not transfer to real devices.","fun_headline_variants_meta":{"raw":{"variants":["PbS/InP single-electron tunneling yields neuromorphic plasticity","Coulomb staircase in PbS/InP enables 0.2 fJ synaptic operation","Single-electron memory and plasticity in PbS/InP junctions","Neuromorphic PbS/InP building blocks with MHz speed and fJ energy","SET-based PbS/InP synapse: short-term and long-term plasticity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000216,"raw_usage":{"total_tokens":1373,"prompt_tokens":830,"completion_tokens":543,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":446,"completion_tokens_details":{"reasoning_tokens":443}},"tokens_in":446,"tokens_out":543,"duration_ms":5042,"temperature":1.0,"reasoning_tokens":443,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:34:21.010589+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure a single PbS/InP junction with pulses shorter than 200 ns and check whether its subsequent probe current still changes; also take the same I-V data at room temperature and at 77 K. If a sub-200 ns pulse produces a persistent conductivity change, or if the Coulomb staircase disappears when the stated resistance and temperature conditions are violated, the central energy and speed claim fails.","supporting_citations":[{"cited_title":"Trap-Free Heterostructure of PbS Nanoplatelets on InP(001) by Chemical Epitaxy,","cited_arxiv_id":null,"evidence_quote":"Prior structural and electrical characterization of the PbS/InP heterostructure that establishes the inhomogeneous self-formed oxide layer the memory mechanism depends on."},{"cited_title":"Measuring charge trap occupation and energy level in CdSe/ZnS quantum dots using a scanning tunneling microscope,","cited_arxiv_id":null,"evidence_quote":"Shows that repeated scanning tunneling spectra shift the onset voltage when charge traps are occupied, the effect the paper reads as memory."},{"cited_title":"Probing the Carrier Capture Rate of a Single Quantum Level,","cited_arxiv_id":null,"evidence_quote":"Provides the single-quantum-level transport result used to justify the two-step trap-mediated tunneling current plateaus."},{"cited_title":"I-V characteristics of coupled ultrasmall-capacitance normal tunnel junctions,","cited_arxiv_id":null,"evidence_quote":"Supplies the double-junction single-electron tunneling model used to fit the I-V curves."},{"cited_title":"From wave-functions to current-voltage characteristics: overview of a Coulomb blockade device simulator using fundamental physical parameters,","cited_arxiv_id":null,"evidence_quote":"Supplies the simulation framework connecting wave functions to current-voltage characteristics for Coulomb blockade devices."},{"cited_title":"Scanning-tunneling-microscope observations of Coulomb blockade and oxide polarization in small metal droplets,","cited_arxiv_id":null,"evidence_quote":"Shows how to extract junction capacitance and resistance values from scanning-tunneling-microscope Coulomb blockade data, the fitting method used here."},{"cited_title":"Synaptic Energy Use and Supply,","cited_arxiv_id":null,"evidence_quote":"Gives the biological synapse energy and speed figures (picojoules, kilohertz) that the paper compares its projected 0.2 fJ and 5 MHz against."},{"cited_title":"Anomalous Capacitance of Thin Dielectric Structures,","cited_arxiv_id":null,"evidence_quote":"Explains why nanoscale junction capacitance need not scale with area, used to defend the small fitted $C_2$ value."}],"review_version":1}