Hardware for Memristive Neuromorphic Systems with Reliable Programming and Online Learning

Alternative computing technologies are highly sought after due to limitations on transistor fabrication improvements. Fabricated memristive technology allows for a non-volatile analog memory for neuromorphic computing. In an integrated CMOS process, the synapse circuits designed for a spiking neuromorphic system can use memristors to regulate accumulation in the neuron circuits. Testing the fabricated memristive devices composed of hafnium oxide and developing a model to represent the key device characteristics lead to specific design choices in implementing the analog memory core of the synapse circuit. The circuits I designed for neuromorphic computing in this process take advantage of the unique capabilities of the memristive device to store a programmable analog memory reliably and efficiently. I designed the peripheral circuitry required including the circuits for programming the memristor and for online learning capabilities.