Ion-tunable antiambipolarity in mixed ion–electron conducting polymers enables biorealistic organic electrochemical neurons

Biointegrated neuromorphic hardware holds promise for new protocols to record/regulate signalling in biological systems. Making such artificial neural circuits successful requires minimal device/circuit complexity and ion-based operating mechanisms akin to those found in biology. Artificial spiking neurons, based on silicon-based complementary metal-oxide semiconductors or negative differential resistance device circuits, can emulate several neural features but are complicated to fabricate, not biocompatible and lack ion-/chemical-based modulation features. Here we report a biorealistic conductance-based organic electrochemical neuron (c-OECN) using a mixed ion–electron conducting ladder-type polymer with stable ion-tunable antiambipolarity. The latter is used to emulate the activation/inactivation of sodium channels and delayed activation of potassium channels of biological neurons. These c-OECNs can spike at bioplausible frequencies nearing 100 Hz, emulate most critical biological neural features, demonstrate stochastic spiking and enable neurotransmitter-/amino acid-/ion-based spiking modulation, which is then used to stimulate biological nerves in vivo. These combined features are impossible to achieve using previous technologies.


Supplementary Tables
Supplementary Table 1  This interaction is validated using ex-situ FTIR spectra before and after voltage cycling in the NH4Cl and dopamine-based electrolytes (Supplementary Figures 10-12). FTIR spectra of pristine BBL are taken first. This is followed by applying voltages corresponding to the first and second reduction potentials of BBL obtained from cyclic voltammetry (Supplementary interactions. This is in line with the lack of VP shift in the case of guanidinium and confirms the hypothesis that the VP shift originates from the interaction of NH4 + /neurotransmitter with C=O and nitrogen of BBL via hydrogen bond. Similar hydrogen bonding interaction between amines and C=O group in polymer has been utilized to enable sensing in previous reports 3 . It has also been known that NH4 + can enable high-charge densities in covalent organic frameworks due to H-bonding 2 . In our case, we observed that the electrons injected per monomer unit are much higher in the case of NH4 + compared to Na + for a given gate voltage. This causes the multiply charged species to be formed at a lower voltage in the case of NH4 + , thus shifting the VP to lower values (Supplementary Figure 14). components at t3. The Na-OECT crosses the maximum current and hence the Na-current drops.

Supplementary
The K-OECT turns on after the delay. We observed water electrolysis above 1.3 V, as evident from the high gate leakage current (Supplementary Figure 28). However, applying such a high voltage did not affect the stability of the OECT considerably (Supplementary Figures 29 and 30). The spiking in a neuron is caused by the opening and closing of voltage-gated Na and K ion channels. There is a concentration difference of ions inside and outside neurons maintained by ion pumps. This is modeled as Na and K batteries and forms the lower and upper limits of the action potential. When a current is injected, the membrane capacitance Cmem causes the voltage to increase, and the voltage-gated Na channels activate and inactivate, followed by a delayed activation of the K channel. The whole system can be modeled using four differential equations given by Hodgkin and Huxley.
In the circuit of the Hodgkin-Huxley model shown Figure 2d, the relation between neuron membrane voltage Vmem and the influx of current Iin(t) can be written as 4 : where Cmem is the membrane capacitance and ∑ ( ) is the sum of the ionic currents passing through the membrane in biological terms. Hodgkin and Huxley further formulated the three ionic currents of K channel, Na channel, and the leakage as where EK, ENa, and EL are the reversal potentials and n, m, and h are ion channel gating variables that control the activation of the K channel, the activation of the Na channel, and the inactivation of the Na channel, respectively. 1 can be rewritten as follows:

Supplementary
where iK(t) and iNa(t) are the ionic currents passing through the K channel and the Na channel respectively.
Given the K-OECT operates in a linear region when activated, the current of the K channel is therefore expressed as: where CV is the volumetric capacitance of BBL polymer semiconductor, W, L, and d are the channel width, length, and thickness respectively, Rdk and Cdk provide the activation delay of the K-OECT. In the absence of Cdk, the gate capacitance of the K-OECT serves the same function.
The voltage to the gate of the Na-OECT is amplified with either inverting or non-inverting amplifier as discussed above and is a function of membrane voltage as f(Vmem, t). The drain of the Na-OECT is applied with ENa while the source of the Na-OECT is connected to the node of Vmem. Two thresholds VNa-m and VNa-s are employed to simulate the activation and the inactivation of the Na channel, respectively. The current of the Na channel is then written as a combination of two equations of OECTs operated in saturation region as: where Pin represents the input current source, PNa and PK are the Na and K channels, and PINV is the inverter amplifier.
Supplementary Figure 39a shows the sample action potential of a slow neuron with a period of 350 ms and the corresponding Na and K currents. For the same period, the power consumption of each source is shown (Supplementary Figure 39b) It is observed that there is a delay between the applied input pulse and the generated spike in the c-OECN (Figure 3d). This latency encodes the input pulse strength because this delay will be proportional to the input amplitude. The OECN can demonstrate the integration of the input signals, ie when two inputs producing subthreshold spikes occur in a short interval of time, it is integrated to generate a spike and enable coincidence detection (Figure 3e). However, if a second input occurs during the hyperpolarisation phase of a previous action potential, there is no new spike generation, thus exhibiting refractoriness (Figure 3f). Due to the presence of subthreshold oscillations associated with action potential, the c-OECN exhibits resonance or frequency preference, ie when two subthreshold inputs occur at a frequency in phase with the oscillations, a spike is generated (Figure 3g). This enables frequency modulated interactions with the neuron. As opposed to the leaky integrate and fire-based OECN which exhibits a fixed threshold for spike generation, the c-OECN has a variable threshold that depends on the neuron's prior activity. As shown in Figure 3h, a signal which generates subthreshold output can create a spike if the threshold is lowered by a preceding inhibitory input. Rebound spike, where the c-OECN spikes in response to an inhibitory input (Figure 3i), and accommodation where a slowly varying input does not invoke a spike while a smaller but sharper input (Figure   3j) can incite a spike can also be observed.