BrainChip’s Akida AKD1000 SoC serves as a reference hardware accelerator design that engineers can use to test BrainChip’s NPU architecture, IP, and software. It can be paired with your choice of MCU or CPU in order to accelerate event-based AI compute tasks, with power-efficient operation thanks to neuromorphic – i.e. brain-like – performance that ignores irrelevant data to save on computing power and memory requirements. Though BrainChip’s software development ecosystem is used to program and run the AKD1000, it is compatible with any CPU and is OS agnostic.
BrainChip’s Akida runs all computations within the hardware and offers off-line training for few-shot datasets. There are 80 configurable NPUs with a neural network size that can be customised to suit your application requirements, whether that be facial recognition, visual wake word detection, or another use case.
The on-chip processor is based on an M-class architecture with FPU & DSP, and on-chip communication interfaces are provided via mesh network. Data input is provided by PCI Express 2.1, USB 3.0, and standard peripheral interfaces. Learn more by filling out the form below, and ipXchange will facilitate evaluation of this disruptive technology for use in your commercial project.
