Study of power dissipation in MolFCN neural structures

Since 1960s, Moore’s law describes the technological evolution of electronic devices through an exponential growth in computing performance. To perform increasingly complex operations, a higher energy consumption is needed, demanding energetic resources and compromising the device's performance. Moore’s law is becoming difficult to be respected, thus approaching its end. New technologies must be investigated based on different principles with respect to silicon-based ones; such technologies are named beyond-CMOS. Molecular Field-Coupled Nanocompunting (MolFCN) stands out as a promising Beyond CMOS technology thanks to its high device density and promising high-frequency operation at room temperature. MolFCN encodes binary information through charge localization in aggregation regions of molecules. The application of external electric fields, applied through the so-called molecule hosting structures, allows forcing specific charge configurations encoding (input field) or deleting (clock field) information. This current-less working principle promises low-power dissipation that must be explored to assess technology potentialities. Indeed, the current literature does not present definitive studies on MolFCN power characterization. To address such research gap, this thesis explores the potentiality of the technology, taking as a case-study a neural application. Initially, the work focuses on realizing molecule-hosting structures. By using Synopsys Sentaurus framework, a wire has been initially explored by creating a structure that is able to host a sequence of molecules: it is a dielectric trench structure characterized by two upper electrodes and a bottom one. The project then proceeds by investigating more complex architectures, exploring the realization of angles, T connections and, finally, the majority voter, core of the technology for logic operations and for neural implementations. Such technological structures are then verified through Sentaurus Device by means of a finite element method simulation, permitting the analysis of the electric fields with different voltage combinations applied to the electrodes. The simulations confirm theoretical predictions by showing clock fields reducing towards the center of the structure. The obtained results demonstrate that the top electrodes create overlap between the electric fields, which should be taken into account in the design of MolFCN. Therefore, this work results in the optimization of structure geometries, bringing to an improvement of the layouts efficiency in providing correct application of the input and clock electric fields. The thesis proceeds by verifying the correct functioning of the optimized structures by analyzing the interactions between the molecules placed in the trench. This is achieved through the Self Consistent ElectRostatic Potential Algorithm (SCERPA), a tool capable of studying the behavior of molecular charges in an iterative manner. In this way, it is possible to fine-tune the voltages applied to the electrodes in order to achieve the correct propagation of logical information. Finally, the provided electric fields and the electrode potentials are used to evaluate the power dissipation associated with the functioning of molecule-hosting structures. In particular, the thesis considers the neural cell as a case-study device and evaluates its power consumption. It demonstrates, for the first time, the possibility of tying SCERPA to a well-established simulation toolchain like the Sentaurus one.