The Akida1000 Reference SoC is a complete neural processing device designed for integration into AI systems, featuring 1.2 million neurons and 10 billion synapses. This SoC is configurable as a standalone embedded AI accelerator or can be paired with an MCU or CPU for enhanced performance. The Akida1000 uses event-based processing optimized for real-time neural network tasks, gaining efficiency from its large SRAM and scalable neural processing units, which are fully configurable to meet application-specific needs. It supports fully integrated networks, eliminating the need for frequent weight swaps between memory systems, thus enhancing throughput and reducing power consumption.
