Spiking Neuromorphic Hardware Accelerator with ReRAM

The proposed innovation targets the scope of Radiation-Tolerant AIML Learning Hardware as outlined in subtopic H6.22 of the 2024 SBIR solicitation. A RRAM-based radiation-tolerant spiking neuromorphic computing architecture within an unsupervised spiking neural network model will be developed in Phase I. This architecture leverages RRAM's unique properties to enhance the efficiency and robustness of AI systems in space environments. By leveraging RRAM's radiation tolerance, our proposed architecture provides a solution to ensure uninterrupted operation of AI systems in space missions, addressing a critical challenge where traditional approaches fall short. Evaluation will involve fault injection testing and TID radiation margining, utilizing SPICE simulation to demonstrate the proof of concept of a radiation-tolerant design approach. Hardware development will include schematic representation and physical layout in Cadence database, utilizing GlobalFoundries 22FDX technology process with RRAM IP provided by the University at Albany, SUNY. The design will incorporate critical circuitry such as input drivers, leaky integrate and fire circuits, and detection circuitry for firing signals. Additionally, circuits for LTD and LTP programming, as well as a pad ring with design for test circuitry, will be implemented for testing purposes. Radiation tolerance of the circuitry will be addressed using techniques like triple redundant latches, extended corner simulations, and back gate biasing. The outcome of Phase I will be a proof of concept utility demonstrating the benefits and efficiency gains of the proposed design, directly applicable for Phase II research and beyond, facilitating the identification of efficiencies in larger designs.