Solution processed organic semiconductors contain extrinsic environmental species that cause device instabilities as they are difficult to remove during low temperature processing and are able to penetrate into organic electronics after fabrication. This dissertation is centered on the search for and development of solid state molecular additives to improve device stability associated with atmospheric defects in organic field effect transistors. To achieve this, an extended study was undertaken of the influence of over 95 different molecular additives expected to improve the stability characteristics of organic field effect transistors based on the literature.

This dissertation demonstrates that positive bias and light stress stability can be improved in both p-type diF-TES ADT and IDT-BT organic field effect transistors by incorporating solid state small molecular additives. Simulations predict that the additives improve stability by introducing a competitive recombination pathway in order to prevent the trapping of electrons in the LUMO of the semiconductor by atmospheric species. Improvements with the solid state molecular additives are achieved by controlling the LUMO and morphology of the additive polymer blend.

Secondly, improvements in the environmental and negative bias stress stability of organic field effect transistors are also observed with molecular additives. The solid state small molecular additives that significantly improve device characteristics are limited to a subset of molecules with similar structure to tetracyanoquinodimethane. It is demonstrated that this subset of solid state molecular additives correlates with a chemical reaction between the molecules and water. The chemical reaction appears to change the molecular additives into a new chemical species, plausibly consuming water, modifying the pH and doping the semiconductor, resulting in improved organic field effect transistor characteristics.

Thirdly, machine learning techniques are used to accurately predict which solid state additives are capable of improving device performance. The machine learning algorithm uses neural passing networks for feature generation, due to its ability to capture physical plausible features such as functional groups. The algorithm screened over 1.5 billion molecular structures and found plausible molecular structures based on expert knowledge.

Fourthly, novel analogue neuromorphic computer architectures based on anti-ferromagnetic and analogue transistors are modeled. The proposed architecture presents both trainable anti-ferromagnetic based synapses for learning and non-trainable voltage controlled synapses for computationally demanding inferences. This dissertation suggests that combining both the fully controllable and trainable networks is a promising route forward for analog neuromorphic computers.