Automated Fabrication and Surface Feature Classification for Atomic Silicon Devices

Abstract

The development of the modern transistor has sparked a technological revolution which has flourished for the past 70 years. Advancements in transistor design and fabrication have allowed for their continued shrinking in size and increase in operation speed. With the continued reduction in size of transistors to ~ 10,000s of atoms, their expected increase in operation speed has reached a roadblock. In order to see any continued increase in operation speed, novel device designs made from few to single atoms or molecules must be explored. One such candidate is Binary Atomic Silicon Logic (BASiL) which relies on selectively patterning dangling bonds (DBs) on an otherwise hydrogen terminated silicon (H-Si) surface. These DBs act as quantum dots allowing for the controlled localization of charge within DB structures. While proof of concept studies have shown the utility in the application of BASiL devices to supplement and replace the transistor, further studies are needed to realize its full potential. The automated fabrication of such devices is limited by the natural distribution of defects commonly found on the H-Si surface. By utilizing a combination of imaging modes of a scanning tunneling microscope (STM) and a non-contact atomic force microscope (nc-AFM), a catalogue of surface defects is created. This catalogue confirms the classification of previously reported and unreported defects providing experiential and theoretical evidence into their proposed assignments. The knowledge gained from this comprehensive study is used to create a neural network capable of segmenting and identifying defects of the H-Si(100)-2x1 surface. This network can distinguish between the defect free surface, charged defects, and uncharged defects through seven different classes. The locations of these defects are used for the automated fabrication of DB devices by patterning a DB structure in the most viable area of the surface as determined by the neural network. The accuracy and number of classes of the neural network is improved by upgrading the quality of the training data used for the neural network enabling additional applications in surface quality assessment.

Utilizing one of the nc-AFM imaging modes refined in the creation of the defect catalogue, the AFM tip is turned into a charge sensor capable of distinguishing between the three charge states of the DB. This charge sensitivity is used to study DB wires fabricated on the H-Si(100)-2x1 surface revealing a previously unobserved ionic charge phase. Performing spectroscopic analysis, we identify higher energy configurations corresponding to an alternative lattice distortion as well as tip-induced charging effects. By varying the length and orientation of these DB structures, we further highlight key features in the formation of these ionic surface phases which provide the experimental insights needed for the incorporation of DB wires for BASiL device control. The combined efforts of developing a surface defect catalog, automated device fabrication scheme, and an understanding of the behaviour of DB wires work to further the development and application of BASiL devices enabling the replacement of current transistor technology by atom scale devices.