Spiking Neuromorphic Accelerator with ReRAM Proof of Concept Completed

• February 11, 2025
• Christian Oliver
• 0.5 Minute Read

GMS has completed the schematic and layout requirements for a fully testable proof-of-concept Spiking Neural Network (SNN).  

The leaky integrate-and-fire (LIF) and spike detection circuits were then designed, together representing one neuron. The full SNN circuit was then designed, integrating the synaptic memristor array and LIF neurons to create input and output layers. 

The Radiation tolerance was addressed in key analog and digital support circuits through radiation hardening by design (RHBD) techniques. 

Extensive simulations allowed to validate performance and functionality under normal conditions and extreme temperatures. 

Projected power consumption and compute performance, scaled for a large processor and various technology nodes, have been benchmarked against state-of-the-art solutions.

It shows that a high performant AI processor can be designed with realistic ReRAM size constraints, if high volume manufacturing at advanced foundries can be achieved.  This study will serve as the foundation for a larger, more advanced post Phase-1 design.