GoAI provides support to test and deploy AI inference software by connecting to Caffe and ARM CMSIS-NN frameworks. This allows users to train and test a model, quantize and retest the trained model on a microcontroller within the GOWIN FPGA and then accelerate their model in FPGA fabric to enable real-time performance.
GOWIN’s GoAI accelerator provides an AHB interface, which allows customers to control the accelerator with a state machine, a soft processor or a hardened Arm Cortex-M3 processor. It also provides easier connectivity with FPGA fabric allowing developers to connect the accelerator to various interfaces such as a MIPI cameras or microphones.
"Due to strict area and computational power requirements, edge AI applications require highly flexible domain-specific frameworks." said Jianhua Liu, director of software engineering for Gowin, in a statement.
Gowin has a demonstration with a neural network trained with the CIFAR10 dataset and configured within the GoAI accelerator.
Related links and articles: