Ultrathin Phase-Change Material Films and Devices for Analog In-Memory Computing

As artificial intelligence is gaining popularity, its energy use keeps increasing. This thesis offers progress towards a possible mitigation strategy to this, in the form of analog in-memory computing, whereby multiplication can be performed directly in the computer memory. This requires nonbinary memory elements, e.g. cells that can store values ranging from 0 to 1000 instead of standard binary computer memory that can store only 0 or 1. In addition, these cells should not lose their information over time or require energy input to keep their information. Phase-change memory is a promising technology that satisfies those requirements. It relies upon a phase-change material, the atomic configuration of which can be reversibly switched between ordered (crystalline, high conductance) and disordered (amorphous, low conductance). Based on the amount of disordered material, the conductance can be tuned to many values. In the first part of this thesis, a phase-change material consisting of antimony with a very small amount of germanium (approximately 5 atomic percent) is investigated as fabricated and after switching. While its low switching energy and high operating temperature are promising for use in in-memory computing, its low switching endurance is a significant drawback. In the second part of the thesis, a novel phase-change memory device architecture is presented, showing excellent performance with respect to switching energy, amount of storable values, and read noise. The so-called disc-type device uses an extremely thin circular phase-change material layer (5 nanometers thick), a small electrode in the center, and a perimeter electrode to achieve this.