Design and analysis of memristor-based reliable crossbar architectures

The conventional transistor-based computing landscape is already undergoing dramatic changes. While transistor-based devices’ scaling is approaching its physical limits in nanometer technologies, memristive technologies hold the potential to scale to much smaller geometries. Memristive devices are used majorly in memory design but they also have unignorable applications in logic design, neuromorphic computing, sensors among many others. The most critical research and development problems that must be resolved before memristive architectures become mainstream are related to their reliability. One of such reliability issue is the sneak-paths current which limits the maximum crossbar array size. This thesis presents various designs of the memristor based crossbar architecture and corresponding experimental analysis towards addressing its reliability issues. Novel contribution of this thesis starts with the formulation of robust analytic models for read and write schemes used in memristive crossbar arrays. These novel models are less restrictive and are suitable for accurate mathematical analysis of any mn crossbar array and the evaluation of their performance during these critical operations. In order to minimise the sneak-paths problem, we propose techniques and conditions for reliable read operations using simultaneous access of multiple bits in the crossbar array. Two new write techniques are also presented, one to minimise failure during single cell write and the other designed for multiple cells write operation. Experimental results prove that the single write technique minimises write voltage drop degradation compared to existing techniques. Test results from the multiple cells write technique show it consumes less power than other techniques depending on the chosen configuration. Lastly, a novel Verilog-A memristor model for simulation and analysis of memristor’s application in gas sensing is presented. This proposed model captures the gas sensing properties of titanium-dioxide using gas concentration to control the overall memristance of the device. This model is used to design and simulate a first-of-its-kind sneak-paths free memristor-based gas detection arrays. Experimental results from a 88 memristor sensor array show that there is a ten fold improvement in the accuracy of the sensor’s response when compared with a single memristor sensor