Nanofiber Channel Organic Electrochemical Transistors for Low‐Power Neuromorphic Computing and Wide‐Bandwidth Sensing Platforms

Abstract Organic neuromorphic computing/sensing platforms are a promising concept for local monitoring and processing of biological signals in real time. Neuromorphic devices and sensors with low conductance for low power consumption and high conductance for low‐impedance sensing are desired. However, it has been a struggle to find materials and fabrication methods that satisfy both of these properties simultaneously in a single substrate. Here, nanofiber channels with a self‐formed ion‐blocking layer are fabricated to create organic electrochemical transistors (OECTs) that can be tailored to achieve low‐power neuromorphic computing and fast‐response sensing by transferring different amounts of electrospun nanofibers to each device. With their nanofiber architecture, the OECTs exhibit a low switching energy of 113 fJ and operate within a wide bandwidth (cut‐off frequency of 13.5 kHz), opening a new paradigm for energy‐efficient neuromorphic computing/sensing platforms in a biological environment without the leakage of personal information.

Sulfur atoms are contained in both PEDOT and PSS; the sulfur atoms are included within the thiophene ring in PEDOT and in the sulfonate moiety in PSS 1,2 . To further understand the component changes after DMSO treatment, the ratios of PEDOT and PSS must be analyzed separately. The S 2p peaks occur at two binding energies ( Figure   1g,h). The lower binding energy peak between 163.5 and 164.5 eV corresponds to the sulfur atoms in PEDOT 1,2 . The higher binding energy peak near 167.5 eV corresponds to the sulfur atoms in PSS 2,3 . The estimated area ratios of PEDOT to PSS are 1:2.32 and 1:3.07 for as-spun and DMSO-treated nanofibers, respectively. The PSS content is increased by 33% on the surface of the nanofibers after DMSO treatment, while PEDOT moves further inside and is more densely packed in the nanofibers. Therefore, the PEDOT:PSS/PAAm nanofibers have PEDOT-rich cores and excess PSS and PAAm in the outer region. This composition change is related to PAAm and PSS being hydrophilic and PEDOT being hydrophobic, causing some amount of PAAm and PSS to be attracted to the surface during DMSO treatment due to the high polarities of these two compounds and DMSO.

Supplementary Note 2-Model of nanofiber channel OECT operation
We introduce here the drain current equation for nanofiber channel OECTs with the volumetric capacitance (C * ) based on the model by Bernards and Malliaras 4 . The Bernards and Malliaras model starts with Ohm's law applied to the channel and the following assumption in the channel: where is the electric current density via the channel, and is channel conductivity, given by the following: where is the number of nanofibers, is the average radius of the nanofibers, ℎ is the hole mobility, and ( ) is the hole density, given by the following: − is the density of sulfonate groups that are compensated for by holes in PEDOT:PSS, and + ( ) is the density of cations that enter the channel when a positive V G is applied to the electrolyte. The channel is treated as a volumetric capacitor, giving the following: Because the steady-state current density throughout the OECT channel is constant, integrating equation (1.7) over the length of the channel yields a straightforward current-voltage relationship.
In the saturation regime (i.e., > ̶ ), the current and transconductance are given by the following: Fabrication process of the nanofiber channel OECT Nanofiber channel OECTs require fabrication methods that have not been used in film channel OECTs. This is because the nanofibers are suspended on a second PaC layer rather than adhered to the substrate within a channel after the transfer process, as shown in Figure S1d. In this state, channels cannot be defined since the nanofibers are peeled off together with the sacrificial second PaC layer during the peel-off process.
Therefore, a novel method for attaching nanofibers to the substrate is needed to obtain channels.
To fabricate a nanofiber channel OECT, we used DMSO for not only improving the conductivity of the nanofiber but also channel patterning. DMSO is usually used to increase the conductivity of PEDOT:PSS by mixing with the PEDOT:PSS dispersion.
DMSO was dropped on the nanofiber dissolved PAAm at the nanofiber surface, and then the channel was filled with nanofibers, as demonstrated in Figure S1e. If either excessive DMSO is dropped on a nanofiber or DMSO treatment is implemented on an as-spun nanofiber without crystallization annealing, the architecture of the nanofiber will collapse. Under a limited DMSO content, the morphology of the nanofiber will remain intact, but the nanofiber will not dissolve. The nanofiber will then remain suspended on the second PaC layer rather than being attached to the substrate, hence disabling channel patterning. Thus, controlling the DMSO content and moderating the dissolution of nanofibers are important for the fabrication of nanofiber channel OECTs.

Transfer method of electrospun nanofibers
Typically, channel formation in OECTs is accomplished via the spin-coating process. Spin coating allows not only channel formation on a whole substrate at a time but also determination of the channel thickness by controlling the rotation speed of the chuck and the number of spins. During spin coating, a polymer solution is coated onto the entire exposed surface of a substrate, which enables the fabrication of OECTs with the same channel material and thickness in a single batch. To make different channel materials or thicknesses on the substrate, a further photolithography process is needed.
However, the solvents used in conventional photolithography, such as developer and photoresist stripper, have deleterious effects, including delamination and swelling, on organic films. Therefore, the formation of various kinds of channels is challenging in film channel OECTs processed via spin coating. This challenge makes it difficult to fabricate neuromorphic devices with low-conductivity channels and sensors with highconductivity channels on a single substrate.
For nanofiber channel OECTs, various amounts of nanofibers are separately electrospun on each collector (e.g., Si wafer) ( Figure S2a   Photographs of (b) a bare cover glass and (c) a cover glass coated with 50 µl of nanofibers. The cover glass position is represented by the yellow arrows.

AFM images
The morphology of DMSO-treated PEDOT:PSS/PAAm nanofibers and the dissolution of a joint at the cross junction between two nanofibers were confirmed by AFM. The nanofibers appeared to be melted and connected at the cross junction and adhered to substrates. The heights of the two nanofibers were 53 (region 1) and 39 nm (region 2), respectively, and the height of the cross junction of the two nanofibers was 83 nm (region 3), demonstrating that the nanofibers were moderately melted and connected rather than simply touching. This morphology allows channels to adhere well to the substrate.

Tunable channel conductivities
Channel conductance can be easily controlled on a single substrate, even by spin coating. However, this approach degrades spatial resolution for application such as EEGs (electroencephalograms) or ECGs (electrocardiograms) because OECTs with large channels are needed to obtain large conductance and signal-to-noise ratio.
Therefore, control of channel conductivity is preferred to channel conductance.
To demonstrate that nanofiber channel OECTs with various channel conductivities in same channel sizes can be fabricated on a single substrate, the devices were

Fastest response of the nanofiber channel OECT
The electrospun volume of 0.5 µl yields the fastest f cut-off in this paper; the value of 13.5 kHz. Even if g m is small, devices with high f cut-off (i.e., response speed) could be optimal building blocks for integrated bioelectronics, such as digital logic gates (e.g., NAND, NOR) 20 .

Calculation of signal-to-noise ratio
The signal-to-noise ratio (SNR) was measured to show the reliability of the results of calculation of switching energy. The SNR was estimated by the results of V G = 10 mV from Figure S11, and the equation is as follows: where is the power of a signal, is the power of background noise, r s is the distance from the signal peak to the average leakage current (I avg ) baseline, and 2 is the variance of the noise defined as 2 = 1 ∑( − ) 2 . The signal amplitude (r s ) and 76 points of baseline I G were selected for the calculation of their variation. The measured SNR was 1.6×10 1 , which can be converted to 12 dB from the following relationship: ( ) = 10 10 Figure S12. Calculation of SNR of the data when measuring switching energy. a) A spike was triggered by V G = 10 mV from Figure S11. I avg was used as baseline for the determination of r s . b) Leakage current points were selected from (a) for the calculation of standard deviation.

Emulation of biological synapses
In biological synapses, the paired-pulse depression (PPD) phenomenon is observed when two spikes are transmitted to a postsynaptic neuron with a narrow time interval between spikes. As the interspike interval increases, the influence of the first spike on the second spike decreases, and eventually, any causality between the two spikes disappears. To establish this functionality in the nanofiber channel OECT, a long channel (L = 5 mm) is adopted to exhibit spike-and-recover behavior wherein hole transport in the channel, rather than ion migration, is the limiting step 16,17 .

Short-term and long-term plasticity
After successive positive V G pulses, relaxation of conductance occurs because cations diffuse back spontaneously to the electrolyte. However, the change in conductance is stabilized at the conductance gap from the initial current in the case of pulse widths of 20, 50, and 100 ms, implying long-term plasticity. The conductance gap increases with a longer pulse width because more ions are injected into the channel and rest within the channel. A short pulse width of 10 ms shows a nonpermanent change in synaptic weight, resulting in short-term depression.
The precise pulse width that causes the transition from short-term plasticity to longterm plasticity is difficult to define because the transition gradually occurred, not sharply at some point, by various conditions such as the magnitude of V G , pulse width, pulse interval, and the number of pulses. Larger V G and/or pulse width and/or shorter pulse interval and/or more the number of pulses inject more ions into deeper nanofibers inside, followed by more ions resting inside nanofibers, causing the changed state to slowly recover to an initial state. This phenomenon continuously occurs.