About the Event
Mainstream non-volatile memory technology, dominated by the floating gate transistor, has historically improved in density, performance and cost primarily by means of process scaling. This simple geometrical scaling now faces significant challenges due to constraints of electrostatics and reliability. Thus, novel non-transistor based memory paradigms are being widely explored. Among the various contenders for next generation storage technology, RRAM devices have got immense attention due to their high-speed, multilevel capability, scalability, simple structure, low voltage operation and high endurance.
In this thesis, electrical and material characterization is carried out on a MIM device system and formation / annihilation of nanoscale filaments is shown to be the reason behind the resistance switching. The MIM system is optimized to include an in-cell resistor which is shown to improve device endurance and reduce stuck-at-one faults. For highest density, the devices were arranged in a crossbar geometry and vertically integrated on CMOS decoders to demonstrate the feasibility of practical data storage applications.
Next, we show that these binary RRAM devices exhibit native stochastic nature of resistive switching. Even for a fixed voltage on the same device, the wait time associated with programming is not fixed and is random and broadly distributed. However, the probability of switching can be predicted and controlled by the programming pulse. These binary devices have been used to generate random bit-streams with predicable bias ratios in time and space domains. The ability to produce random bit-streams using binary resistive switching devices based on the native stochastic switching principle may potentially lead to novel non-von-Neumann computing paradigms.
Further, sub-1nA operating current devices have been developed. This ultra-low current provides energy savings by minimizing programming, erase and read currents. Despite having such low currents, excellent retention, on/off ratio and endurance have been demonstrated.
Finally a scalable approach to simple 3D stacking is discussed. By implementation of a vertical sidewall-based architecture, the number of critical lithography steps can be reduced. A vertical device structure based on a W / WOx / Pd material system is developed. This scalable architecture is well suited for development of analog memory and neuromorphic systems.