A modular organic neuromorphic spiking circuit for retina-inspired sensory coding and neurotransmitter-mediated neural pathways

Signal communication mechanisms within the human body rely on the transmission and modulation of action potentials. Replicating the interdependent functions of receptors, neurons and synapses with organic artificial neurons and biohybrid synapses is an essential first step towards merging neuromorphic circuits and biological systems, crucial for computing at the biological interface. However, most organic neuromorphic systems are based on simple circuits which exhibit limited adaptability to both external and internal biological cues, and are restricted to emulate only specific the functions of an individual neuron/synapse. Here, we present a modular neuromorphic system which combines organic spiking neurons and biohybrid synapses to replicate a neural pathway. The spiking neuron mimics the sensory coding function of afferent neurons from light stimuli, while the neuromodulatory activity of interneurons is emulated by neurotransmitters-mediated biohybrid synapses. Combining these functions, we create a modular connection between multiple neurons to establish a pre-processing retinal pathway primitive.

Supplementary Discussion 1: The organic spiking circuit and the benefits of the OECTbased design.
This section introduces the devices and materials that compose the spiking circuits (Supplementary Figure 1a,b).Before presenting the single devices characteristics, an overview is provided describing the state-of-the-art materials that are commonly used to design neuromorphic spiking circuits, highlighting the benefits of employing organic electrolyte materials/devices (OECT) comparing organic non-electrolyte (OFET) and inorganic technologies (Supplementary Table 1).
The circuital design of current spiking circuits is predominantly based on inorganic materials, particularly on silicon-based devices 1 (Supplementary Discussion 1, Supplementary Table 1).However, neuromorphic applications operating at the biointerface clearly favour organic materials 2 , and specifically OECT-based neuromorphic circuits.Next to their relatively soft and flexible properties, high tunability and low operational voltage, organic mixed ion-electron conductors conduct both electrons and ions 3 .This mixed conduction allows organic materials to closely match the operating timescales of their biological counterparts 4 , reduce interface impedance 5 , and mimic ionbased biological functions such as neuronal ion-flux communication and neurotransmitter-receptor binding 2 (synaptic function), which are required to interact with biological tissues and to design adaptive biointerfaces 6 .Although showing desirable biocompatibility and flexibility properties, standard OFETs materials fail to translate (with the same sensibility and reliability of OECTs) the interaction of a complex bio-interfaced environment into a change of the neuromorphic circuit electrical characteristics.
Supplementary Table 1.List of the critical characteristics of artificial neurons and neuromorphic circuits based on different technologies.Critical neuromorphic characteristics comparing inorganic materials (grey), organic non-electrolyte materials (OFET-based, green), organic electrolyte materials (OECT-based, brown) neuromorphic systems.The devices and materials that compose the spiking circuits of this work (Supplementary Figure 1 a,b) are presented and the single device voltage and transfer characteristics are displayed (Supplementary Figure 3, Supplementary Figure 4, Supplementary Figure 5 and Supplementary Figure 6).Note that for all the experiments presented in the manuscript main text, the capacitor was disconnected from the circuit (see Fig. 1) by omitting the electrolyte that connects the two sides of the capacitor.mm.Supplementary Figure 1c shows a 3D rendering of the circuit, highlighting the single circuital elements with the same colour code.
In Supplementary Figure 1d a photograph of the circuit substrate shows again the different components and how these are interfaced with the electrolyte (PBS).The spiking circuit capacitor (PEDOT: PSS OECT, blue) receives the external input in forms of voltage potential charging from 0 to 0.26 V (see Supplementary Figure 6).A 330 kOhm resistor is used to connect the capacitor to the Vin.The capacitor is connected to the ground through a switch device (series connection through the channel terminal of this device, orange).In the experiments presented in the manuscript main text, the capacitor was disconnected from the circuit and consequently, the internal capacitance of the OECTs of the inverters and switch was used to integrate the input signal.Indeed, as reported in Supplementary Discussion 7, the overall system capacitance is dominated by the capacitance of the inverters and switch with a negligible contribution deriving from the capacitor element.This is the result of the scaling process which allowed us to accommodate all the elements on a quarter of a glass slide.The switch (P-3O OECT) opens and close the ground connection of the capacitor depending on its resistance (Supplementary Figure 5).The polymer P-3O was synthesized following a previously reported protocol 25 .The switch device is in ON/OFF state depending on the voltage applied to its gate terminal and allows the capacitor to charge and discharge (Supplementary Figure 6).The inverter pair of the neuromorphic circuits is constituted by four identical p(C4-T2-C0-EG) OECTs (Supplementary Figure 1a, green).The polymer p(C4-T2-C0-EG) was synthesized following a previously reported protocol 26 .p(C4-T2-C0-EG) is an ambipolar material so the inverter of these OECTs can behave both as n-type and p-type transistors (Supplementary Figure 3).To the first inverter a VDD1= 0.6V is applied, while on the second inverter VDD2 = 0.8V is applied.The VDD unbalance was introduced to obtain a sharper inversion point at the second inverter (increasing de-facto the gain and the inversion range of the second capacitor) which is key to replicate neural spikes.When the capacitor (or the internal capacitance) charge reaches the inverters threshold (0.3 V), the first inverter characteristic voltage, due to the low VDD1, moves from 0.6 to 0 V (Supplementary Figure 4, left axis).The second inverter, receiving as input voltage the output of the first inverter, displays a characteristic curve with an inversion point (0.26 V) with the voltage moving from 0 to 0.8 V (Supplementary Figure 4, red axis).The output terminal of the second inverter is connected to the gate terminal of the switch device so that the voltage inversion (0 to 0.8 V) of this system modulate the change of the state of the switch from OFF to ON (Supplementary Figure 5).
When the switch is in the ON state the connection to the ground is closed and the capacitor discharges to 0 V, allowing a new cycle to start.
In Supplementary Figure 7 it is displayed a frequency modulation experiment with relevant parameters reported in Supplementary Table 2.In this case the capacitor (internal capacitance) was directly connected to an external voltage supplier and the voltage input was increased from 0.1 to 1.3 V. Through this experiment the threshold voltage for spikes initiation is identified.For values of the input voltage below 0.2V, the neuromorphic circuit does not spike: the capacitor charging time is slow and the circuit VOUT reach an equilibrium state close to 0 V or some oscillations are shown.In the input voltage window 0.2 to 0.3 V the neuromorphic circuit shows irregular spikes.In this regime, due to the low input voltage, the total internal capacitance does not fully charge and then discharge in between each spike, so uncomplete spikes are recorded.Hence, it is impossible to extract a value for the spike frequency, since the spikes are irregularly spaced.Increasing the input voltage to 0.3 V, the neuromorphic circuit fires regularly spaced spikes.The extracted spike frequencies are reported in Supplementary Table 1, these are normalized respect to the starting frequency 0.1 Hz at 0.3 V, and the frequency increase with VIN is reported in percentage (Supplementary Figure 7 and Supplementary In Supplementary Figure 8

Light sensor
In this section we describe the mechanisms of coupling of a commercially available light sensor to the neuromorphic spiking circuit to mimic the functions of an afferent neuron and perform sensory coding of light stimuli.
The light sensor we used is the DFRobot's Analog Ambient Light Sensor (DFR0026) with a voltage supply of 5 V (Supplementary Fig. 20).The light sensor was shielded from the ambient light except for a small opening on top of the sensor.With the use of an LED and a green 555 nm filter, we controlled the light stimuli delivering 0.01 mW/cm 2 (low light intensity) and 0.07 mW/cm 2 (high light intensity), which correspond to 11000 lux (ambient daylight) and 70000 lux (direct sunlight) 27 .As such, we replicated two common conditions experienced by the human retina: an ambient daylight close to the levels of an outdoor space, and the direct illumination encountered when facing the sun.
Supplementary Figure 9. Ambient light sensor response.Analog ambient light sensor (supply voltage 5 V).In the table, the parameter extracted from the calibration of the sensor are reported with details on the optical power and the light conditions.
Hence, the light sensor was directly coupled to the input of the neuromorphic circuit as depicted in Supplementary Figure 10.As reported in Fig. 1      In Supplementary Figure 13a,b the neurotransmitter-mediated neuromorphic device included in the synaptic modulator (presented in the manuscript), the so-called bio-hybrid synapse, is depicted, showing the Gate, Source, Drain ITO electrodes.The design of this device is based on a previous work 28 .
In the same panel, the fluidic module is in pale blue and shows the direction of the flow.Supplementary The device work as a conventional organic electrochemical transistor (OECT) based on the material poly(3,4-ethylene-dioxythiophene):polystyrene sulfonate (PEDOT:PSS), with a gate constituted by the same material.Hence, the device operates in depletion mode and the application of a voltage bias at the gate electrode causes a de-doping of the film due to cations penetrating its bulk (Supplementary Fig. 1   d,e). 29This mechanism is reversed upon removal of the voltage bias.Employing voltage pulses as inputs, the device can operate as an ENODE achieving a volatile control of the conductivity of the organic mixed ionic/electronic conductor when no neurotransmitters are present.In this condition, a reversible conductance modulation is achieved (Fig. 1b, black dotted traces).
However, in presence of a neurotransmitter, the PEDOT:PSS channel conductance decreases in a nonreversible way due to the de-doping process (Fig. 1b, red and blue solid traces) 30 .This mechanism is strongly influenced by protons and electrons produced during the oxidation reaction thus depending on the type of neurotransmitter and its concentration in the solution as previously shown. 30upplementary Figure 15 the bio-hybrid synapse conductance modulation is reported in case dopamine (left) and serotonin (right) are employed.In this experiment the input voltage applied to the bio-hybrid gate consist of three periodic voltage pulses of 300 mV (dopamine) and 400 mV (serotonin) with time width 3 s and delay of 10 s 28 .The conductance GD depends on the concentration of neurotransmitter and on the number of voltage pulses applied (in these graphs only three).
Increasing the number of input gate pulses, the dynamic range of the device saturates (as seen in Fig. 1 c for DA) and the conductance modulation per pulse decreases.
Form the previous work, we present an overview on the mechanism of conductance modulation in such devices.The conductance modulation of the bio-hybrid synapse can be explained by the two possible reactions that can occur at the PEDOT:PSS electrodes.The first reaction is the two-electron oxidation of dopamine or serotonin as described: Where DA is dopamine, DQ is dopamine o-quinone 31 and 5HTQ is the serotonin quinone 32,33 .The Eq.
1 and 2 reactions result in the de-doping of PEDOT:PSS ( the decrease of its conductance) as shown in Supplementary Fig. 13 and given by the following equation.
Where the electron reduces PEDOT + , eliminating a hole and leading to a decrease in conductivity, and the proton or other cationic species (Cat + ) compensates the negative charge of the sulfonate group on PSS.The previous reactions lead to a decrease in the channel conductance bio-hybrid synapse.Our neuromorphic system demonstrates for the first time an artificial network of neurons connected through chemical synapses (biohybrid synapses) which work to adjust the synaptic weights depending on the neurotransmitter environment and regulate in cascade the spike frequency along a neural pathway.
This system has been developed with the scope of future bio-integration.In this context, a previous work by Keene et al. demonstrated the use of the bio-hybrid synapse (with a device working on the same principles of our synapse) as a platform interfaced with PC-12 cells.It allowed the transduction of chemical signals of these cells (released dopamine) into permanent conductance modulations of the organic neuromorphic device's channel terminal (chemical to electrical signal transduction) 30 .In the more recent work of Matrone et al., to investigate the interaction of electroactive neurotransmitters ( dopamine and serotonin) without interfering agents ( additional molecules expressed by the cell models), the sending chemical signal ( neurotransmitter release) was substituted by a controllable microfluidic module 28 .In the same context, in our work the chemical messengers' activity is emulated by the activation of the microfluidic module.However, even though this design choice introduced a gap between biological and artificial synapses for a practical control of neurotransmitters release, the proposed system has been clearly developed on the same design rules used by previous bio-interfaced devices 30 .Hence, the biohybrid synapse can be interfaced with cells models that express electroactive neurotransmitters such as dopamine and serotonin 34 .In this case, the biohybrid synapse conductance state depends on the electrolyte environment (which is controlled by the interfaced cells).Hence, the spiking circuit connected to this synapse will show frequency adaptation depending on the cell activity (neurotransmitter expression).
For example, a sending neuron signal can be used to trigger the release of neurotransmitters (electrical stimulation) from the interfaced cells, and promote the oxidation of these molecules to control the output of the synaptic modulator 35 , eliciting in cascade an increase of the receiving neuron spikes frequency.
On the other hand, on a still artificial design level without the use of bio-interfaced tissues, alternative approaches to locally deliver neurotransmitters and replicate the synaptic machinery have been also proved 36,37 .

Supplementary Discussion 4: Synaptic modulator with dopamine and serotonin for frequency tuning and signal transmission.
In this section the strategy to employ the bio-hybrid synapse to modulate the spike frequency of the neuromorphic circuit is presented.The bio-hybrid synapse from previous section (Supplementary Figure 13) is included in a voltage divider configuration (Supplementary Figure 16a) connected in series with a 1 kOhm resistor, hereinafter referred as synaptic modulator (Synmod).The VOUT of this voltage divider depends on the ratio of the resistance of the two elements comprising the circuit according to the formula: We also note that the VOUT is controlled by the value of VDD applied, as explained at the end of this section.In the above formula we assume the bio-hybrid synapse to be a tuneable resistor.Indeed, the resistance of the PEDOT:PSS-based can be tuned by applying a gate potential > 0 V , i.e. switching off the OECT.In Supplementary Figure 13, the conductance of the bio-hybrid synapse was modulated applying periodic voltage pulses (generated by an external voltage supply) on the gate of the device, exploiting the oxidation of dopamine and serotonin.However, in this work the Synmod mimic the function of a biological synapse, the VOUT is connected to neuromorphic spiking circuit, replicating the functions of an interneuron (Fig. 1 d and Supplementary Figure 17), while the bio-hybrid synapse receives the spikes from an afferent neuron.Supplementary Figure 22 depicts the laboratory setup that allowed the measurements.Each neuron is connected to the Arkeo system with the clips depicted in the photograph.The artificial synapses are connected to Arkeo and so to the neurons via probes.The two syringe pumps were used to control the presence of DA and 5-HT in the bio-hybrid synapses.In this configuration, in the neutral electrolyte PBS (no neurotransmitter in solution) the Afferent neuron spikes applied to the gate of the biohybrid synapse enable a reversible modulation of its resistance.
Indeed, the VOUT of the voltage divider does not change (Supplementary Figure 18).When DA or 5-HT are introduced in solution, using a fluidic module as depicted in Supplementary Figure 22, the oxidation of these molecules leads to an increase of the resistance of the bio-hybrid synapse that depends on the numbers of pulses applied and on the concentration of the neurotransmitters.Hence, the voltage divider output increases.This is used as input voltage to charge the capacitor (internal capacitance) of the spiking circuit.In short, the spike frequency is modulated (increases) depending on the type of neurotransmitter used and on the pulsing time.7. Spike modulation data for different serotonin concentrations at high light intensity.Spike modulation data as extracted from the trace in Supplementary Figure 20 for 5-HT 0.05 mM and DA 0.01mM (high light intensity as afferent neuron input), comprising the time, the voltage divider (synaptic modulator) output, the time between each pair of spikes, the corresponding frequency and the percentage of frequency change.In order to perform the experiment of Fig. 3, two spiking circuits (Interneurons) were connected in series using two bio-hybrid synapses in Synmod configuration to influence the spike frequency of the corresponding neuron.The first Interneuron (DA-Interneuron) receives the input (spikes) on its biohybrid synapse mediated by DA from an Afferent neuron (Supplementary Figure 23).Supplementary

Supplementary Table
Figure 22 depicts the laboratory setup that allowed the measurements.Each neuron is connected to the Arkeo system with the clips depicted in the photograph.The artificial synapses are connected to Arkeo and so to the neurons via probes.The two syringe pumps were used to control the presence of DA and 5-HT in the artificial synapse.At the beginning of the experiment, the artificial synapses operate with PBS, so their conductance is not affected by the incoming input signal (spikes), thus their synaptic modulator output is stable.The global input to this neural pathway is represented by the spikes generated by an afferent neuron subject to high light intensity and low light intensity (Fig. 3b).The pattern of the afferent neuron spikes has been previously recorded with the Arkeo System.To ensure reproducible results, the two recorded spike patterns (high light and low light intensity, Supplementary Figure 11) have been used as input using Arkeo System as a voltage supplier.
Considering Fig. 3, in the case of low light intensity, when the first pump is activated, dopamine is expressed at the first synapse, so the synaptic modulator output increases (+ 0.04 mV) in turns increasing the frequency of the corresponding neuron (+ 12%).At this stage, if 5-HT is not expressed in the associated synapse, the synaptic modulator output of the last neuron does not change and no modulation of its frequency is recorded (black).When 5-HT is introduced in the artificial synapse through the microfluidic pump, a synaptic modulator output (blue, + 0.035 mV) and neuron frequency modulation (blue, +11%) is recorded.
We here contextualize the platform's performance and functions with emphasis on the presented cascaded neural pathway, establishing a direct parallel to specific biological circuits displaying similar functions and thus commenting on its applicability as a bio-interfaced system.
Building on Supplementary Discussion 3, we still consider the work by Keene et al. demonstrating the use of biohybrid synapses as devices allowing the transduction of chemical signals from PC12 cells (released dopamine) into permanent conductance modulations of the organic neuromorphic device's channel terminal.This work established a chemical to electrical signal transduction approach which has been thoroughly investigated from an electrochemical perspective by Matrone et al. 28 .In our synaptic modulator, dopamine and serotonin allow a semi-permanent conductance modulation (synaptic weight update) which is used to control the activity of a connected spiking neuron, eliciting not only frequency modulations in persistently spiking neural networks (Figs. 1 and 3) but also triggering spikes (neural activation), and so signal transmission, in a silent pathway (Fig. 2).By design, the biohybrid synapse used in the synaptic modulator can be employed as an active interface with tissue and cells models.
Indeed, leveraging the biocompatibility of PEDOT:PSS, cells can be directly seeded on the gate terminal of this device, with extensive reports supporting these applications in literature [38][39][40] .In this configuration, the presence of physiologically released electroactive neurotransmitters, such as dopamine and serotonin, allows to operate the presented neuromorphic neural pathway as an adaptive platform which does not merely record biochemical clues into electrical signals but more importantly locally transduce these relevant signals in a neuromorphic fashion.
By examining biological neural pathways involving a complex interconnection of neuromodulation processes, we establish a direct parallel between the neuromorphic platform and midbrain neurons thus validating the significance of this work.
DA neurons in the ventral tegmental area bidirectionally regulate the activity of 5-HT neurons in the dorsal raphe nucleus, a biological scenario which has been replicated through the neuromorphic platform and is illustrated in Figs. 2 and 3. 41 We investigated the biological phenomena at the base of the interaction between dopamine and serotonin neurons, and hereby highlight technological gaps which must be addressed to establish an intimate coupling between neuromorphic devices and biological pathways.First, as analyzed in detail in Supplementary Discussion 7, we reaffirm that only the amplitude of change of the spikes frequency represent the brain encoding language which allow to translate both external and internal stimuli into neural data.As such, midbrain interneurons perform computing functions through frequency modulations of +/-30% (respect to the baseline frequency, Supplementary Fig. 30), which correspond to the same frequency changes demonstrated by our neuromorphic platform in Figs. 2 and 3. On the other hand, viable strategies to increase the operative frequency of the neuromorphic systems are investigated and suggested in Supplementary Discussion 7.
Finally, focusing on the electrical to chemical signal transduction, we here investigate the concentrations of neurotransmitters involved in biological neurons functions, to prove the neuromorphic platform applicability in bio-hybrid scenarios.
For DA and 5-HT, two kind of signalling patterns (in the midbrain and retina) have been identified as phasic and tonic, supposedly associated to the biological functions of teaching signal (direct signal processing) and motivational drive (indirect signal processing), respectively.Direct signal processing is associated to the release of neurotransmitters in the pre-to post-synaptic space (4 mm) upon presynaptic firing (neurons phasic activity).This mechanism represents a core computational primitive which we intended to replicate in this work.According to the suggestion that the synaptic compartment has to be distinguished from the extra-synaptic compartment, synaptic transmission can be modeled on the hypotheses that neurotransmitters release is highly localized to the synapses and that neurotransmitters uptake strongly contributes to this tight localization 42 .Under these assumptions, the concentration of neurotransmitters peaks during firing and returns to the baseline level (4-50 nM) in the ms range.While it is still difficult to estimate or directly measure the peak concentration reached during a firing event, models predict a change of the instantaneous (local) level of neurotransmitters to be in the tens of mM.In this work we used concentrations of neurotransmitters ranging from 0.025 mM to 0.1 mM (25 to 100 µM) to simulate the phasic release of dopamine and serotonin at the synaptic terminal.While lower concentrations of neurotransmitters (below 10 µM) can be still used to elicit the conductance modulation phenomena described in Supplementary Discussion 3, the range of concentrations employed in this work still correspond to a realistic biological scenario which can be replicated by bio-interfacing the platform with neuronal model cells such as PC12 (dopamine release).
As such, midbrain and retina neuronal pathways depending on the interaction of afferent and interneurons mediated by the neurotransmitter dopamine and serotonin have been replicated by the design of the modular neuromorphic platform.

Supplementary Discussion 6: Power Consumption of the spiking circuit and the synaptic modulator
The spiking circuit replicating the functions of a biological neuron employs three voltage sources.
Hence, the total power consumption Ptot can be estimated considering the single power sources contributions, i.e. the input voltage VIN and the two voltages VDD1 and VDD2 that are required to operate the first and second inverter, respectively.As such, first the power for each voltage source has been calculated and then the single contributions have been summed: ./. =  01 +  233) +  2334 (6)     We evaluated the power consumption of the spiking circuit used to emulate an afferent neuron operating in both the high frequency regime (emulating the biological scenario of high light intensity) and the low frequency regime (low intensity).As shown in Supplementary Figure 24, the main power contribution is from the second inverter of the circuit (red curve).The energy consumed per spike-event by the circuit was estimated by integrating the power over the time of a single spike.The energy per spike Es consumed by the neuromorphic circuit in the high spiking-frequency regime (VIN = 1.3 V, Supplementary Figure 24) is 6.2*10 -5 J. On the other hand, in the low frequency regime the Es is 1.2*10 -4 J (Supplementary Figure 25).Hence, these values confirm that the neuromorphic circuit requires higher power than the one required by its biological counterpart.

Supplementary Figure 1 .
photolithography dxf file.d) Photograph of the integrated circuit realized on a glass substrate, highlighting the total lateral footprint, scalebar = 15 mm.

Supplementary Figure 2 .
Fabrication process for integrated spiking circuit.Thermal evaporation of 10/100 nm Ti/Au layers, photoresist (S1805) spin-coating and patterning, chemical etching of Au (Au-etchant) and Ti (HCl), double PaC deposition ( insulating and sacrificial layers) and RIE patterning, polymer thin films spin-coating and peel off of the sacrificial PaC.
the spike frequency extracted from the previous data are plotted.With a linear regression (red line, y = a + bx) it is proved that the voltage/ frequency relation is sublinear (b = 0.13) in the voltage window examined (0.3 -1.3 V).

4 VSupplementary Discussion 2 :
INPUT (V) y= a + bx Coefficient values ± one standard deviation a = 0.08 ± 0.01 b = 0.13 ± 0.01 Sensory coding with light c and Supplementary Figure 11, two different light intensities have been employed to test the light sensory coding capability of the neuromorphic circuit and mimic the aforementioned biologically plausible conditions.Relevant parameters, extracted from the spiking pattern of Fig. 1c, are reported in Supplementary Table

Supplementary Figure 12 .Supplementary Table 4 .Supplementary Figure 13 .
Three-spikes output of the neuromorphic circuit.The output generated by the neuromorphic circuit responding to the light conditions (low and high light intensity) whose duration triggers a train of three spikes.Spike modulation data for short time light stimuli.Spikes data for short time light stimuli ( triggering three spikes) as extracted from the trace in Supplementary Figure 12, comprising the light sensor output, the time between each pair of spikes, the corresponding frequency and the percentage of frequency change.The biohybrid synapse and the voltage divider.a) The connections of the synaptic modulator and its components and b) close-up of the bio-hybrid synapse component.c) The top view of the bio-hybrid synapse.

Figure 13c panel depictsSupplementary Figure 14 .
Figure 13c panel depicts top view of the transparent glass where the grey areas represent the ITO electrode, the dark blue rectangles are the PEDOT:PSS films used as gate and channel of the device, the light blue area corresponds to overlap of the microfluidic module with the substrate.As such, the channel and gate PEDOT:PSS areas, the active areas in contact with the electrolyte, both measure 3 × 5 mm.The circles indicates the position of the inlet hole of the microfluidic module.

TimeSupplementary Figure 20 .Supplementary Figure 21 .
Neuromodulation with serotonin at low high light intensity.Neuromodulation with serotonin using the Synaptic modulator showing the modulation of the Interneuron spike frequency depending on the serotonin concentration.The input signal to the Synaptic modulator is a train of spikes from the afferent neuron under high light intensity.The frequency modulation is in percentage (dark blue for high 5-HT concentration and light blue for low 5-HT concentration) with respect to the spike frequency of PBS, black (0.10 Hz).Neuromodulation with serotonin at low low light intensity.Neuromodulation with serotonin using the Synaptic modulator showing the modulation of the Interneuron spike frequency depending on the serotonin concentration.The input signal to the Synaptic modulator is a train of spikes from the afferent neuron under low light intensity.The frequency modulation is in percentage (dark blue for high 5-HT concentration and light blue for low 5-HT concentration) with respect to the spike frequency of PBS, black (0.10 Hz).

Supplementary Table 8 .Supplementary Figure 22 .
Spike modulation data for different serotonin concentrations at low light intensity Spike modulation data as extracted from the trace in Supplementary Figure 21 for 5-HT 0.05 mM and DA 0.01 mM (low light Spiking neurons to voltage dividers setup.Photograph of the setup used in the laboratory to connect each neuron (DA and 5-HT) to its synaptic modulator (DA and 5-HT) synapse and activate the fluidic module to control the expression of the neurotransmitters.

Figure 24 .
Power consumption of the neuromorphic spiking circuit at high frequency.Power for each voltage source (VIN black, VDD1 pink and VDD2 red) in the neuromorphic circuit high spiking regime (frequency 0.25 Hz).

P
IN = power input voltage P VDD1 = power drain voltage 1 P VDD2 = power drain voltage 2

Table 2 )
. All these results are obtained from three different spiking circuit to check device reproducibility.Hence, the frequency values reported in the table are the average over three devices/experiments (standard deviation StDev is highlighted in Supplementary Table1).Besides, still in Supplementary Table2, the variation of the spike width is reported respect to the spike frequency.As the spike frequency change with the frequency, at each input voltage the full-width-at-half-maximum is extracted from the previous series of data and reported in the table.Likewise, the amplitude of the spike in reported in the same table.
the percentage of frequency change.The values reported are averaged over three devices/ experiment to check the neuromorphic circuit reproducibility.
and Supplementary Figure8The spike frequency modulation.Spike frequency modulation (depending on the V input) as extracted from Supplementary Figure7traces.The plot shows that the relation is sublinear in the voltage region 0.3 -1.3 V.
Calibration experiments are here reported (Supplementary Figure11) showing a modulation of the spike frequency from 0.1 Hz to 0.25 Hz, as extracted in Supplementary Table8, when moving from low light (0.3 V light sensor output) to high light intensity (1.2 V sensor output).Additionally, we investigate the neuron output response to voltage inputs that trigger the same number of spikes, in order to check the reproducibility and robustness of our encoding mechanism when stimuli of different duration are applied.In Supplementary Figure12the duration of light stimuli of different intensities is adjusted to ensure trains with only three spikes.As evident, the low light condition requires a duration time of 35 s to elicit a train of three spikes, while the high light condition three spikes train only requires a stimulus of 16 s.In both cases, train of spikes with amplitude, frequency and percentage modulation similar to the condition (long term stimuli) presented in Supplementary Figure11are recorded and listed in Supplementary Table4.In summary, the output of the spiking circuit does not depend on the duration of the stimulus, if the stimulation time, depending on the input voltage, allows to trigger the spikes activity.