Breaking the limits of computer vision, VeriSilicon AI-ISP technology brings innovative image enhancement experience

Image quality upgrade with low power consumption, low latency and low DDR bandwidth achieved through VeriSilicon FLEXATM technology

September 6, 2022, Shanghai, China - VeriSilicon, a leading Silicon Platform as a Service (SiPaaS®) company, today announced the launch of its innovative artificial intelligence image enhancement AI-ISP technology, which can provide advanced image enhancement effects that surpass traditional computer vision technology for applications such as smartphones, automotive electronics, and industrial Internet of Things.

VeriSilicon's AI-ISP technology innovatively integrates the company's scalable and programmable neural network processing (NPU) technology, which has been widely used in the market , and image signal processing (ISP) technology, which has passed the dual certification of ISO 26262 and IEC 61508 functional safety standards. Among them, the efficient NPU technology provides pixel-level processing for 4K video with extremely low power consumption, supports adaptive precision calculation and multi-frame fusion functions, and has excellent noise reduction performance in low light, realizing artificial intelligence image enhancement technology that exceeds computer vision; VeriSilicon's unique innovative FLEXATM low-power and low-latency synchronous interface communication technology allows the ISP to read, write and access data directly from the NPU, and the internal communication between the two does not require CPU intervention; AI-ISP also adopts an intelligent workload balancing architecture, which can dynamically optimize power consumption, memory access rate and utilization based on workload, truly realizing low-power, low-latency and low-DDR bandwidth image quality upgrades.

VeriSilicon's NPU technology can be customized for different chip sizes and power budgets. It is a cost-effective and high-quality neural network acceleration engine. Its complete software stack and SDK support mainstream deep learning frameworks, including Tensorflow, PyTorch, ONNX, TVM, IREE, etc. VeriSilicon has established an ecological partner community to develop artificial intelligence algorithms covering multiple market segments and provide customers with complete solutions. VeriSilicon's ISP has high code efficiency and can complete high-performance real-time image processing. It has the characteristics of low area, low power consumption and high throughput, while taking up very little memory.

"AI technology is everywhere. VeriSilicon's industry-leading NPU has been delivered to more than 60 customers and used in more than 110 system-on-chips (SoCs) in the past few years . Now, VeriSilicon is once again taking the lead in deeply integrating AI technology into specific domain processors, breaking the limits of traditional technology. The AI-ISP is designed to push image quality beyond the limits of computer vision technology, especially for 4K video processing that requires extremely low power consumption, and extremely low-light applications with near-zero illumination, such as mobile phones and automobiles," said Weijin Dai, Executive Vice President and General Manager of IP Business Unit at VeriSilicon. "With VeriSilicon's intelligent pixel processing IP portfolio from camera input to display output (Glass to Glass), and VeriSilicon's innovative FLEXATM IP interconnect technology, we are working closely with key customers to provide them with efficient, low-power and seamless AI integration solutions to break the limitations set by traditional technologies."