Guidelines for visual synapse characterisation for PV experts

Version 1.0 Experimental Setup Guidelines for Artificial Visual Synapse Characterisation SOLIS Project - Work Package 1 SOLIS Consortium Abstract This document is addressed to Photovoltaic (PV) researchers interested to investigate their devices for visual synapses and neuromorphic applications. It is written as part of the Marie Curie Staff Exchange project SOLIS https://cordis.europa.eu/project/id/101183049. The document provides guidelines for establishing an experimental platform to characterise artificial visual synapses using optical stimulation. Designed for thin film photovoltaic researchers, it covers hardware configurations, synchronisation strategies, and standardized measurement protocols for synaptic plasticity characterisation. Contents 1 Introduction 3 2 Hardware Configurations 3 2.1 Configuration A: Dual-Channel Source meter (Keithley 2636A) . . . . . . . . . . 3 2.2 Configuration B: Waveform Generator + Measurement Instrument . . . . . . . . 3 2.3 Budget-Conscious Example Setup (∼e2,350) .................... 4 3 Optical Calibration 4 3.1 Power Calibration Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 4 Timing and Synchronisation 4 4.1 TemporalParameters.................................. 4 4.2 Configuration A: Keithley 2636A Synchronisation . . . . . . . . . . . . . . . . . . 4 4.3 Configuration B: External Triggering . . . . . . . . . . . . . . . . . . . . . . . . . 5 4.4 Instrument Speed Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 5 Integration Time and Noise (NPLC) 5 6 Standard Characterisation Protocols 6 6.1 General Measurement Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 6.2 Long-Term Potentiation (LTP) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 6.3 Long-Term Depression (LTD) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 6.4 Short-Term Plasticity: Paired-Pulse Facilitation (PPF) . . . . . . . . . . . . . . . 7 6.5 Spike-Rate-Dependent Plasticity (SRDP) . . . . . . . . . . . . . . . . . . . . . . 7 6.6 Spike-Timing-Dependent Plasticity (STDP) . . . . . . . . . . . . . . . . . . . . . 7 6.7 Current-Voltage (JV) Hysteresis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 7 Data Acquisition and Analysis 8 7.1 DataStructure ..................................... 8 7.2 KeyMetrics....................................... 9 7.3 TypicalResults..................................... 9 SOLIS-Guidelines for visual synapses characterisation 1 Version 1.0 8 Experimental Best Practices 9 8.1 DeviceConsiderations ................................. 9 8.2 EnvironmentalControl................................. 9 9 Software Integration 10 10 Summary and Recommendations 10 10.1 Minimum Equipment Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . 10 10.2 Critical Success Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 10.3RecommendedWorkflow................................ 11 11 Conclusion 11 12 Appendix A: Equipment Specifications 11 12.1 Waveform Generator Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . 11 12.2 Digital Multimeter Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 12.3 LED and Driver Specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 12.4OpticalComponents .................................. 12 13 Appendix B: Timing Diagram Examples 13 13.1 Configuration A: Keithley 2636A . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 13.2 Configuration B: WFG + DMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 14 Appendix C: Sample Data Analysis 14 14.1 Conductance Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 14.2 Plasticity Metrics Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 14.3SRDPAnalysisExample................................ 14 14.4STDPAnalysisExample................................ 15 15 Appendix D: Quick Reference Tables 15 15.1 Typical Protocol Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 15.2ReadVoltageSelection................................. 15 15.3 Instrument Settings Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 16 Appendix E: Material-Specific Considerations 16 16.1 Common Thin Film PV Materials . . . . . . . . . . . . . . . . . . . . . . . . . . 16 16.2 Wavelength-Dependent Response . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 SOLIS-Guidelines for visual synapses characterisation 2 Version 1.0 1 Introduction Visual synapses operate through optically-controlled memristive behavior, leveraging persistent photoconductivity in thin film semiconductors. Characterisation requires: •Pulsed optical stimulation (write operation) •Electrical conductance readout (read operation) •Precise temporal control (ms to s timescales) •Synchronized triggering between stimulation and measurement This guide addresses two primary hardware configurations: dual-channel Source meter units (e.g., Keithley 2636A) and combined waveform generator + measurement instrument setups. 2 Hardware Configurations 2.1 Configuration A: Dual-Channel Source meter (Keithley 2636A) Required equipment: •Keithley 2636A (or equivalent dual-channel SMU) •Fiber-coupled LED with driver •LED driver with external trigger capability •Optical fiber and collimator •Calibrated optical power meter Connection scheme: •Channel A: Device terminals (apply read voltage, measure current) •Channel B: LED driver trigger input (pulse generation) •Optical path: LED →fiber →collimator →device Advantages: Integrated control, high timing precision, simplified synchronisation via software. 2.2 Configuration B: Waveform Generator + Measurement Instrument Required equipment: •Dual-channel waveform generator (e.g., Multicomp Pro MP751062) •Digital multimeter with trigger input (e.g., Keysight 34460A) or single-channel SMU •Fiber-coupled LED with triggered driver (e.g., Thorlabs LEDD1B + M430F1) •Multimode optical fiber and collimator •Calibrated optical power meter Connection scheme: •WFG Channel 1: LED driver external trigger input (optical stimulation pulses) •WFG Channel 2: Device bias (apply read voltage Vread) •DMM input: Device current measurement •Trigger synchronisation: WFG output →DMM external trigger input Critical consideration: The measurement instrument must support external triggering to synchronize current acquisition with the pulse sequence. SOLIS-Guidelines for visual synapses characterisation 3 Version 1.0 2.3 Budget-Conscious Example Setup (∼e2,350) Component Model/Specs Cost (e) Waveform generator Multicomp Pro MP751062 (2-ch, 60 MHz) 250 Digital multimeter Keysight 34460A (6.5-digit, triggerable) 1,300 Fiber-coupled LED Thorlabs M430F1 (430 nm, 5.3 mW) 250 LED driver Thorlabs LEDD1B (T-Cube, ext. trigger) 350 Optical components Fiber + collimator 200 Total 2,350 Table 1: Low-cost experimental setup for visual synapse characterisation Note: Multiple LEDs at different wavelengths are recommended to study spectral response. 3 Optical Calibration 3.1 Power Calibration Protocol Accurate characterisation requires calibrated incident optical power density at the device surface. Procedure: 1. Position optical power meter detector at device plane. 2. Measure optical power Popt [mW] for each LED at operating current. 3. Calculate power density: Pd=Popt/Abeam [mW cm−2] 4. Record beam diameter at device position (measure spot size with beam profiler or imaging). 5. Verify spatial uniformity across device active area. Requirements: •Calibrated Si or InGaAs photodiode detector (wavelength-dependent) •Power meter with nW to mW range •Document: λ,Popt,Abeam, LED current, measurement distance 4 Timing and Synchronisation 4.1 Temporal Parameters Synaptic characterisation in thin film device should happen on timescales from milliseconds to seconds. Ideally we would go faster, but for the equipment described here, this is the fastest we can hope for: Tperiod ≥tpulse +tdelay +tsettle +tmeas (1) tpulse :optical stimulation duration (10–1000 ms) (2) tdelay :post-pulse relaxation (5–100 ms) (3) tsettle :measurement stabilization (2–20 ms) (4) tmeas :current acquisition window (1–50 ms) (5) 4.2 Configuration A: Keithley 2636A Synchronisation Synchronisation is managed internally via the controlling software: 1. Channel B outputs voltage pulse to trigger LED driver. 2. Software waits tpulse +tdelay. SOLIS-Guidelines for visual synapses characterisation 4 Version 1.0 3. Channel A applies Vread, waits tsettle. 4. Channel A measures current. 5. Software calculates conductance: G=I/Vread. NPLC setting: Use NPLC = 0.1–1 for balance between speed and noise rejection (see Section 5). 4.3 Configuration B: External Triggering Requires explicit hardware synchronisation: Setup: 1. WFG Channel 1 generates optical stimulus trigger (TTL pulse, duration tpulse). 2. WFG Channel 2 applies constant Vread to device during read phase. 3. DMM external trigger configured to start acquisition at t=tpulse +tdelay. 4. Trigger signal: use WFG marker output or delayed copy of Channel 1 pulse. Critical: Verify DMM trigger latency (typically 1–5 ms). Adjust tdelay accordingly. Measurement sequence: for each pulse: t=0: WFG Ch1 triggers LED (ON for t_pulse) t=t_pulse: LED turns OFF t=t_pulse + t_delay: WFG triggers DMM acquisition t=t_pulse + t_delay + t_settle: DMM samples current Record I, calculate G = I / V_read 4.4 Instrument Speed Considerations •WFG: Typical trigger jitter <1µs, adequate for ms-scale pulses. •DMM (e.g., Keysight 34460A): Integration time depends on NPLC; trigger delay ∼2 ms. Use fast trigger mode if available. •Keithley 2636A: Buffer-based acquisition with ∼17 ms per point at NPLC=1 (60 Hz line frequency). Use lower NPLC for faster measurements. However, from my experience, it is better to limit yourself to about 30ms unless you operate at very high read voltages so the current is high, often above 0.5V in our devices. •LED driver: Rise/fall time typically <1µs. Verify driver specifications but it is normally not a limitation. 5 Integration Time and Noise (NPLC) The Number of Power Line Cycles (NPLC) controls integration time and noise rejection: NPLC Integration Time (60 Hz) Application 0.01 0.17 ms Fast transients, high noise 0.1 1.7 ms Synapse readout, moderate noise 1.0 17 ms Standard DC measurements 10 170 ms Ultra-low noise, slow dynamics Table 2: NPLC selection guidelines Recommendation: For synaptic characterisation with tpulse ≥10 ms, use NPLC = 0.1–1. SOLIS-Guidelines for visual synapses characterisation 5 Version 1.0 6 Standard Characterisation Protocols 6.1 General Measurement Sequence All protocols follow this base structure: 1. Initialize: Measure baseline conductance G0(zero illumination). Ensure that the baseline is almost flat at the chosen read voltage, to make sure that the read voltage does not significantly contribute to the plasticity. 2. Stimulate: Apply optical pulse sequence. 3. Read: Measure conductance after each pulse: apply Vread, measure I, compute G= I/Vread. 4. analyse: Calculate plasticity metrics (see equations below). Read voltage: We recommend to use Vread = 0.05–0.2 V (non-destructive, linear regime). Verify device IV characteristics first, and as previoulsy said, be sure that it does not significantly contribute to the plasticity. 6.2 Long-Term Potentiation (LTP) Objective: Demonstrate sustained conductance increase under repeated optical stimulation. Protocol: 1. Measure G0(initial conductance, dark conditions). 2. Apply N= 50–200 optical pulses: •Pulse intensity: Pd= 1–10 mW cm−2 •Pulse width: tpulse = 50–500 ms •Period: T= 200–1000 ms 3. After each pulse: wait tdelay, measure Gn. 4. Plot Gnvs. pulse number n. Metrics: ∆GLTP =GN−G0(6) ∆G%=GN−G0 G0 ×100% (7) Expected behavior: Gnincreases and saturates at Gmax. 6.3 Long-Term Depression (LTD) Objective: Demonstrate conductance decrease. It often requires reverse bias or dark relaxation, but in some cases, we have observed depression assisted by light (which is ideal, as you then have a fully optically controlled synapse). Protocol (dark relaxation): 1. Potentiate device to Gmax (apply LTP sequence). 2. Measure conductance decay in dark: sample G(t)at intervals (e.g., every 10 s for 10 min). 3. Fit decay: G(t) = G∞+ (Gmax −G∞)e−t/τ Protocol (electrical reset): 1. After LTP, apply negative voltage pulses (if applicable) or opposite polarity. 2. Monitor Gndecrease over Mreset pulses. Metrics: ∆GLTD =G0−GM(8) τ:decay time constant (9) SOLIS-Guidelines for visual synapses characterisation 6 Version 1.0 6.4 Short-Term Plasticity: Paired-Pulse Facilitation (PPF) Objective: Measure conductance enhancement from closely-spaced pulse pairs. Protocol: 1. Apply two identical optical pulses separated by interval ∆t= 10–100 ms. 2. Measure G1after first pulse, G2after second pulse. 3. Repeat for multiple ∆tvalues (e.g., 10, 20, 50, 100 ms). 4. Allow relaxation between trials (∼60 s in dark). Metrics: PPF(∆t) = G2 G1 ×100% (10) Expected behavior: PPF >100% for ∆t<τrelax, where τrelax is the conductance relaxation time constant. 6.5 Spike-Rate-Dependent Plasticity (SRDP) Objective: Characterise synaptic weight change as a function of input frequency. Protocol: 1. Define frequency range: fmin = 1–10 Hz, fmax = 50–500 Hz. 2. Use Nf= 8–15 logarithmically-spaced frequencies: fi=fmin ·fmax fmin i/(Nf−1) , i = 0, . . . , Nf−1(11) 3. At each fi: •Measure G0. •Apply Npulse = 30–100 pulses at frequency fi(period T= 1/fi). •Measure final conductance Gf. •Calculate ∆G(fi) = Gf−G0. •Allow full relaxation (5–10 min dark) before next frequency. 4. Plot ∆Gvs. f. Typical pulse parameters: •tpulse = 10–50 ms (constrain: tpulse < T/2) •Fixed intensity across all frequencies Expected behavior: ∆Gincreases with f(potentiation regime) or decreases (depression regime), possibly showing critical frequency transition. 6.6 Spike-Timing-Dependent Plasticity (STDP) Objective: Map synaptic weight change as a function of pre-post spike timing. This can be hard to do in a fully optical setup. Below is a proposition for a protocol. Protocol: 1. Define timing offset range: ∆t∈[−50,+50] ms (10–20 points). 2. At each ∆tj: •Measure G0. •Apply Npairs = 30–100 spike pairs: – Pre-synaptic spike: Optical pulse (tpulse = 10–20 ms) – Post-synaptic spike: Electrical pulse on device (e.g., +0.5 V, 10 ms) applied at time t+ ∆tj –Configuration A: Channel A applies post-spike, Channel B triggers LED for prespike. –Configuration B: WFG Ch1 triggers LED (pre), WFG Ch2 applies voltage pulse (post) with programmed delay. SOLIS-Guidelines for visual synapses characterisation 7 Version 1.0 •Inter-pair period: Tpair = 200–500 ms. •Measure Gf, calculate ∆G(∆tj)=Gf−G0. •Allow relaxation before next ∆t. 3. Plot ∆Gvs. ∆t. Timing convention: ∆t=tpre −tpost (12) •∆t>0: Pre before post →expected LTP (potentiation) •∆t<0: Post before pre →expected LTD (depression) Expected STDP window: ∆G(∆t) = (A+exp(−∆t/τ+) ∆t>0 −A−exp(∆t/τ−) ∆t<0(13) where τ±≈10–20 ms (biological inspiration). 6.7 Current-Voltage (JV) Hysteresis Objective: Characterise memristive behavior in the device electrical response. Protocol: 1. Dark sweep: •Sweep voltage: V=−0.2to +1.0V (forward), then reverse (backward). •Step size: ∆V= 10–20 mV. •Measure current at each point. 2. Illuminated sweep: •Apply constant illumination (Pd= 1–10 mW cm−2). •Repeat forward-backward voltage sweep. 3. Light-soaking effect: •Pre-illuminate device (e.g., 60 s at 10 mW cm−2). •Immediately measure dark JV curve. •Compare with initial dark JV (measure hysteresis width). Analysis: •Hysteresis area: Ahyst =HI dV •Shift in turn-on voltage or resistance state between forward/backward sweeps •Persistence of conductance change after light removal 7 Data Acquisition and Analysis 7.1 Data Structure Record for each measurement: •Time stamp (relative to sequence start) •Pulse index n •Measured current I[A] •Applied read voltage Vread [V] •Calculated conductance G=I/Vread [S] •Optical parameters: λ,Pd,tpulse •Environmental conditions: temperature, ambient light (if not in dark box) Metadata: Device ID, area, materials, measurement date, protocol parameters. SOLIS-Guidelines for visual synapses characterisation 8 Version 1.0 Metric Definition Interpretation ∆G Gf−GiAbsolute conductance change ∆G%(Gf−Gi)/Gi×100% Relative change Dynamic range Gmax/Gmin On/off ratio PPF G2/G1×100% Short-term facilitation Nonlinearity Deviation from linear G(n)Synaptic update symmetry Table 3: Synaptic plasticity metrics 7.2 Key Metrics 7.3 Typical Results LTP: Expect ∆G>20–100% over 50–200 pulses. Saturation indicates Gmax reached. SRDP: Logarithmic or sigmoidal ∆G(f)relationship. Critical frequency fcmarks LTD-toLTP transition. STDP: Asymmetric window with ∆G > 0for ∆t > 0and ∆G < 0for ∆t < 0. Time constants τ±= 10–50 ms typical. JV Hysteresis: Pinched loop at origin indicates memristive behavior. Hysteresis width correlates with memory retention. 8 Experimental Best Practices 8.1 Device Considerations •Compliance: Set current compliance 20% above expected maximum to prevent device damage. •Bias polarity: Verify device polarity (some devices require specific bias direction for potentiation). •Active area: Define and document device illuminated area for power density calculations. •4-wire vs. 2-wire: Use 4-wire (Kelvin) sensing for low-resistance devices (< 100 Ω). 8.2 Environmental Control •Dark box: Essential. Eliminate ambient light during measurements. •Temperature: Ideally it should be monitored and stable, as we are measuring very small currents. Ideally... •Humidity: Can be important for air-sensitive materials, but has not been an issue for us so far. Measurement quality •Repeatability: We should try to perform measurements per protocol and report mean ± standard deviation. But some devices may degrade. •Relaxation time: Allow sufficient dark recovery between measurements (typically 5– 10 min) to avoid cumulative effects. If you have a way to reset the device, optically or electrically, even better. •Baseline stability: Verify G0returns to initial value after relaxation. If not, device may show fatigue or irreversible damages. •Cable shielding: If doing very low current measurements, we recommend shielded coaxial cables to minimise electromagnetic interference. SOLIS-Guidelines for visual synapses characterisation 9 Version 1.0 •Ch1: Pulse function, amplitude 3.3–5 V (TTL), width tpulse, period T •Ch2: DC function, amplitude Vread •Trigger mode: Continuous or single (protocol-dependent) 16 Appendix E: Material-Specific Considerations 16.1 Common Thin Film PV Materials CdTe: •Bandgap: 1.45 eV (λ < 850 nm) •Recommended LED: 430–625 nm •Known metastability: Light soaking improves fill factor (Cu migration) •Expected synaptic behavior: Moderate LTP, strong persistence CIGS/Kesterite (CZTS): •Bandgap: 1.0–1.5 eV (tunable with Ga or composition) •Recommended LED: 530–850 nm •Known metastability: VOC increase under light soaking (defect reordering) •Expected synaptic behavior: Strong LTP/LTD, good dynamic range Sb2S3/Sb2Se3: •Bandgap: 1.1–1.7 eV (S vs. Se) •Recommended LED: 430–730 nm •Known metastability: Interface defects, persistent photoconductivity •Expected synaptic behavior: Fast response, wavelength-selective Perovskite (CH3NH3PbI3): •Bandgap: 1.55 eV •Recommended LED: 530–730 nm •Known metastability: Ion migration, strong hysteresis •Expected synaptic behavior: Very strong memristive behavior, but stability concerns •Caution: Measure in inert atmosphere or encapsulated devices 16.2 Wavelength-Dependent Response For materials with strong absorption edge, synaptic response varies with wavelength: Above bandgap (hν > Eg): •Strong absorption, high carrier generation •Fast conductance increase (strong LTP) •May saturate quickly Near bandgap (hν ≈Eg): •Moderate absorption •Balanced LTP/LTD dynamics •Optimal for SRDP/STDP characterisation Below bandgap (hν < Eg): •Defect-mediated absorption •Slow, persistent conductance change •Long time constants (suitable for memory applications) Recommendation: Characterise with 2–3 wavelengths spanning above/near/below bandgap. Document Version: 1.0 (September 2025) SOLIS-Guidelines for visual synapses characterisation 16 Version 1.0 Contributors: SOLIS Consortium (UPC, UniLiv, TalTech, UniVerona, MASCIR, UWC, NIP, Ritsumeikan) Acknowledgments: This work is part of the SOLIS project (HORIZON-MSCA-SE-2023), focusing on solar cell-inspired inorganic semiconductor synaptic systems for low-energy edge computing and visual learning. https://cordis.europa.eu/project/id/101183049 License: This document is released under CC BY 4.0. The described methods and protocols are freely available for academic and research purposes. Contact: For questions, contributions, or reporting errors in this guide, contact the SOLIS coordination team: Prof Jonathan Major, University of Liverpool (task leader) Email: [email protected] Prof. Zacharie Jehl Li-Kao, UPC Barcelona (coordinator) Email: [email protected] Companion Software: The SOLIS characterisation software (Python-based, open-source) is available at: https://github.com/SOLIS-project/Keithley-2636A-Parameter-analyser Citing this guide: SOLIS Consortium (2025). Experimental Setup Guidelines for Artificial Visual Synapse Characterisation. SOLIS Project Technical Report, WP1. DOI: [to be assigned] SOLIS-Guidelines for visual synapses characterisation 17