Neural and AI image signal processing VLSI IP to support computer vision
For computer vision applications in smartphones, industrial automation and automotive electronics, there is a need to enhance the image quality to recognise image more accurately. To implement that on silicon chip directly, leading VLSI design service solution provider VeriSilicon is offering Artificial intelligence based image signal processing technology to integrate into silicon.
The key features and benefits of its technology:
It combines VeriSilicon's own programmable Neural Network Processing (NPU) technology with Image Signal Processing (ISP) technology. To ensure functional safety, it meets safety critical standards such as ISO 26262 and IEC 61508.
Low-power consuming NPUs deliver pixel-level processing for 4K video featuring self-adaptive resolution calculation and multi-frame fusion function, and take the noise out from the pictures captured in low light, so that the picture is good enough for AI image enhancement for computer vision and evem better.
VeriSilicon uses its own high-speed Synchronous communication interface called FLEXA for the ISPs to access data directly from NPUs bypassing CPU.
To reduce power consumption AI-ISP dynamically optimizes power consumption and memory access/utilization based on workload, and also ensures low-latency and low-DDR bandwidth image quality upgrades.
VeriSilicon says its NPU technology, customizable for different chip sizes and power budgets, is a cost-effective, premium neural network acceleration engine with a complete software stack and SDK that supports mainstream deep learning frameworks including Tensorflow, PyTorch, ONNX, TVM, IREE and more.
VeriSilicon claims its ISPs are highly code-efficient and capable of high-performance real-time image processing, with low area, low power consumption, and high throughput while having a low memory footprint.
“Artificial intelligence is ubiquitous. VeriSilicon’s industry-leading NPUs have been delivered to more than 60 customers in the past few years and been embedded into more than 110 system-on-chips (SoCs). VeriSilicon is now once again pioneering the deep integration of AI technology into domain-specific processors, pushing the envelope of traditional technologies. The AI-ISP is designed to set new standards for image quality beyond current limits for computer vision technology, and is especially suited for 4K video processing that demands ultra-low power consumption in extremely low-light scenarios with near-zero illuminance such as mobile phones and automobiles,” said Wei-Jin Dai, Executive VP and GM of IP Division at VeriSilicon. “With VeriSilicon’s Glass to Glass (from camera-in to display-out) intelligent pixel processing IP portfolio and the innovative FLEXA™ IP interconnection technology, we are working closely with key customers to provide them with efficient, low-power and seamless AI-integrated solutions that perform beyond the limitations of traditional technologies.”