In this thesis, a novel memory based on III-V compound semiconductors is studied, both theoretically and experimentally, with the aim of developing a technology with superior performance capabilities to established and emerging rival memories. This technology is known as ULTRARAM™.
The memory concept is based on quantum resonant tunnelling through InAs/AlSb heterostructures, which are engineered to only allow electron tunnelling at precise energy alignment(s) when a bias is applied. The memory device features a floating gate (FG) as the storage medium, where electrons that tunnel through the InAs/AlSb heterostructure are confined in the FG to define the memory logic (0 or 1). The large conduction band offset of the InAs/AlSb heterojunction (2.1 eV) keeps electrons in the FG indefinitely, constituting a non-volatile logic state. Electrons can be
removed from the FG via a similar resonant tunnelling process by reversing the voltage polarity.
This concept shares similarities with flash memory, however the resonant tunnelling mechanism
provides ultra-low-power, low-voltage, high-endurance and high-speed switching capability.
The quantum tunnelling junction is studied in detail using the non-equilibrium Green’s function (NEGF) method. Then, Poisson-Schrödinger simulations are used to design a high-contrast readout procedure for the memory using the unusual type-III band-offset of the InAs/GaSb heterojunction. With the theoretical groundwork for the technology laid out, the memory performance is modelled and a high-density ULTRARAM™ memory architecture is proposed for random-access memory applications. Later, NEGF calculations are used for a detailed study of the process tolerances in the tunnelling region required for ULTRARAM™ large-scale wafer manufacture.
Using interfacial misfit array growth techniques, III-V layers (InAs, AlSb and GaSb) for ULTRARAM™ were successfully implemented on both GaAs and Si substrates. Single devices and 2×2 arrays were then fabricated using a top-down processing approach.
The memories demonstrated outstanding memory performance on both substrate materials at 10, 20 and 50 µm gate lengths at room temperature. Non-volatile switching was obtained with ≤ 2.5 V pulses, corresponding to a switching energy per unit area that is lower than DRAM and flash by factors of 100 and 1000 respectively. Memory logic was retained for over 24 hours whilst undergoing over 10^6 readout operations. Analysis of the retention data suggests a storage time exceeding 1000 years. Devices showed promising durability results, enduring over 10^7 cycles
without degradation, at least two orders of magnitude improvement over flash memory. Switching of the cell’s logic was possible at 500 µs pulse durations for a 20 µm gate length, suggesting a subns switching time if scaled to modern-day feature sizes. The proposed half-voltage architecture is shown to operate in principle, where the memory state is preserved during a disturbance test of > 10^5 half-cycles. With regard to the device physics, these findings point towards ULTRARAM™ as a universal memory candidate. The path towards future commercial viability relies on process development for aggressive device and array-size scaling and implementation on larger Si wafer