Biohybrid Electrically Conductive Fibers—Cryogel OECTs for Neuromorphic Computing

Both functionality and the complexity of biological systems challenge current human technologies. Contrary to biological systems, present artificial means of information processing rely on CMOS-based hardware with rigid, fixed architectures and limited interaction modalities. Despite the recent advances in neuromorphic software implementations, hardware alternatives remain elusive. This work proposes a biocompatible and biomimetic system capable of information processing under physiological conditions in the context of reservoir computing as a step towards bridging both worlds.<br/><br/>Reservoir computing relies on non-linear, physical dynamical systems (reservoirs) coupled with minimal conventional artificial neural networks to perform non-conventional computing. The core and reservoirs in this study are three-dimensional (3D) biomimetic macroporous hydrogel scaffolds (cryogels) comprised of sulfated/sulfonated polymers (SSPH) such as heparin, a highly negatively charged natural glycosaminoglycan (GAG) covalently crosslinked with synthetic 4-arm poly(ethylene glycol) (starPEG) via carbodiimide chemistry at subzero temperature [1, 2]. Mechanical and morphological properties of the sponge-like scaffolds can be controlled via the composition of the precursor solution and the freezing conditions during their synthesis.<br/>Poly(3,4-ethylene dioxythiophene) (PEDOT) is electropolymerized from EDOT solutions into these highly hydrated and tough scaffolds which are contacted with Pt-coated electrodes. The resulting sets of highly branched fiber-like structures composed of PEDOT:SSPH organic mixed ionic-electronic conductors (OMIEC) are stably embedded in the pore walls of the cryogel scaffolds.<br/><br/>To employ the cryogels as reservoirs, we developed 3D printed structures that integrate positional fixation for cryogel cylinders, up to 24 platinum electrodes, and liquid handling ports while allowing visual monitoring of the reservoir in a sealed environment. Interfacing with the artificial neural network layer is achieved by the electrical connection of the platinum electrodes to analog I/O multichannel boards and a PC running Python scripts. Ultimately, reservoirs are fabricated by electropolymerized PEDOT fiber-like structures in the scaffold and exploiting the resulting dynamic behavior derived from ionic-electronic transport coupling.<br/><br/>This work showcases the OMIEC character of the PEDOT:SSPH fiber-like structures by demonstrating their organic electrochemical transistor (OECT) p-type depletion mode behavior under a three-terminal configuration. In addition, parametrical exploration of the PEDOT integration process highlights the intertwining of PEDOT and GAG morphologies, contrary to 2D and free-standing fibers previously reported [3] in which PEDOT:dopant morphology is highly dependent on the electropolymerization signal parameters. Memory-capacity measurements combined with simple predictive tasks confirm the neuromorphic computing capabilities of the proposed system.<br/><br/>Furthermore, network density and performance limits are still to be assessed. The 3D system increases the degrees of freedom and connectivity of the recurrent network and creates integration challenges. Subsequent steps towards <i>in vitro</i> applications are planned. The potential of the cryogel scaffolds as delivery systems for signaling proteins due to their GAG component was previously reported [2]. The biomolecular toolbox compatible with the biomimetic, highly hydrated reservoirs tempt to elucidate the full potential of their modulation by biochemical signals. Coupling biochemical affinity, ionic-electronic transport, and <i>in situ</i> real-time computing with the possibility of hosting cells would be a step towards merging the biological and semiconductor worlds.<br/>REFERENCES:<br/>[1] Welzel, P. et al (2012). https://doi.org/10.1021/bm300605s<br/>[2] Sievers, J. et al (2021) https://doi.org/10.1016/j.biomaterials.2021.121170<br/>[3] Cucchi, M. et al (2021) https://doi.org/10.1002/aelm.202100586