Realization of neuromorphic systems for the detection, transduction and classification of bio-electronic signals

This thesis work is the result of a partnership between ChiLab-Materials and Microsystems Laboratory (Politecnico di Torino) in Chivasso and the Institute of Materials for Electronics and Magnetism (IMEM) at the National Research Council (CNR) in Parma. In the first part, the fabrication process was done, by realizing the devices through standard cleanroom procedures, such as metal deposition, etching and photolithography. The devices layout was improved, so that the total number of devices per single die passed from 8 to 16. Three different channel sizes were considered: this choice was important to evaluate the behavior of memristors with different channel aspect ratio. After that, the attention was focused on the inkjet printing process, chosen for the deposition of the active material (PANI) on the devices channel. The inkjet printer used here relies on a piezoelectric actuation principle: the droplet ejection is obtained by stimulating the nozzle with a voltage signal. The process improvement included different aspects. Firstly, the ink synthesis was optimized. Different concentrations were investigated to improve both the printability and the conductive properties. At this point, the inkjet printer parameters were optimized, starting from the voltage signal waveform definition up to the spot parameters. The inkjet printing conditions varied according to the channels sizes and the pattern dimensions to be printed, since the polymeric ink quantity needed was different. Furthermore, important role was played by the final drying step of the printed pattern. Two solutions were considered: a short, fast drying procedure and a long, slow one. In the first case, the variability among the devices was greater, affecting negatively the final printed pattern, which showed a non-homogeneous and uniform volume. In the second case, instead, the variability among the devices was evidently reduced, with a much better volume uniformity and printed pattern repeatability. In the second part of the activity, the electrical characterization of organic memristors was carried out at CNR-IMEM in Parma. The measurements were done in liquid electrolyte (HCl 1M). A comparison between short-dried and long-dried devices was proposed. In fact, short-drying procedure impacted negatively on the performances, since the variability among devices with same channel aspect ratio was evident. On the opposite, long-drying procedure lead to much better results, with much less variability and more repeatability for twin devices. The electrical characterization tests included the analysis of the hysteresis curves at different working frequency (scan rate), followed by the extraction of the minimum and maximum values of conductance. In the best cases, these devices showed a stable behaviour, with a constant gap separating minimum and maximum conduction values. The geometry factor of devices channels is relevant here, since channel conduction is directly proportional with the channel aspect ratio. To show the dependence of the resistance on applied voltage and time, the kinetics in voltage and time were investigated. The synaptic weight variation was evaluated as function of the applied voltage pulse amplitude and time width. To prove the reliability and stability of organic memristors, endurance tests were executed on the best performing devices (in both short and long dry procedure cases). Finally, neuromorphic properties were analysed, in particular Long-Term Potentiation (LTP) and Long-Term Depression (LTD).