Stacked-die DMS chip packs 1.1 trillion ops/sec processing power

Image sensor technology expert OmniVision has launched a stacked-die automotive driver monitoring chip. Highly integrated device packs AI neural processor, image signal processor and DDR3 memory. This AI powered processor provides processing speeds up to 1.1 trillion operations per second which is required for eye gaze and eye tracking algorithms in driver monitoring systems. With 1K MAC of convolutional neural network acceleration, the QAZ8000 is designed to consume just one watt of power in typical conditions. The neural processor is supported by popular TensorFlow, Caffe, MXNet, and ONNX toolchains. The other processor cores available on chip includes Arm Cortex A5 CPU cores with Neon technology for accelerated video encoding/decoding and on-chip video analytics algorithms, along with hardware for image processing, video encoding and RGB/IR processing. The video input can be both from RGB and infrared image sensors so that it supports videos taken during both day and night. To give an example, the on chip video encoder can read 5 megapixel image captures from OmniVision’s automotive image sensors, and also delivers quality output up to 2K resolution video at 30 frames per second (fps). The delay between ignition and activation of driver monitoring system camera is significantly reduced, since boot-up time for the OAX8000 is faster. OmniVision claims boot-up t...