2014全国社会治理创新案例颁奖典礼 暨经验交流现场会

This patent includes an all-new Adder Neural Network that goes beyond the constraints of existing AI computing frameworks to solve difficult mathematical bottlenecks in AI. It can reduce computing power consumption and circuit size by more than 70%, and support all-scenario AI application. This technology can also allow consumer electronics to save power when using AI functions.

This technology resolves some of the foremost challenges in autonomous driving, including positioning, perception, and decision-making challenges in complex urban road scenarios where people and vehicles are mixed. Our invention resolves 70%+ of challenges in these interactive scenarios and shortens vehicle pass-through time by 40%+. It also cuts computing power usage by 87.5%, reduces CPU usage by 85%, and slashes storage overhead by 200 times.

This solution helps carriers use public networks to meet the varied requirements of 5G networks across different industrial scenarios. It ensures that services are not interrupted when individual networks go down, and that local data is kept within campuses to meet security requirements. The highly reliable private networks enabled by this solution have been deployed in multiple mining sites, where it enables autonomous operations.

Optical distribution networks, or ODNs, serve as the underlying infrastructure for fiber broadband. The passive nature of optical fibers makes them difficult to manage and maintain. Achieving the digital management of optical fiber has been a decades-long struggle, ever since they were first invented by Charles Kao. Our Digital QuickODN solution enables just that by marking an "optical iris" on fibers, akin to a QR code for each connection. This innovation makes it possible to intelligently and digitally manage and maintain ODN networks. This means fiber broadband can be planned and deployed faster and more accurately and faults can be more easily located and rectified, making the O&M of ODNs 30%+ more efficient.

This technology enables a more flexible mix of computing precision and computing power. It can meet varying requirements for floating-point precision computing in scenarios from high-performance computing to AI training and inference that need large computing power. It has already been incorporated into our next-generation Ascend and Kunpeng chips and provides 2x computing power for AI training and inference compared with industry counterparts. The HPC + AI hybrid precision computing has become the latest direction of research within academia and industry, and the full-precision computing unit has also become critical to the success of the computing industry.

This compact, high-density, large-image AR head-up display (HUD) is built on innovative optics and algorithms. It solves issues with image distortion and vertigo commonly found in automotive head-up display systems due to limited size. The AR HUD turns the windshield into a screen that integrates all types of information while ensuring security. Its Optical Display Processor (ODP) defines the next generation of HUDs, offering a brand-new intelligent driving experience.

This innovation addresses major challenges in indoor and outdoor 5G deployment.

For outdoor scenarios, we have developed the See-through Antenna that can accommodate different antenna units that work on different frequency bands, and our BladeAAUs simplify 5G network deployment while ensuring optimal 2G, 3G, 4G, and 5G network performance. The solution has helped more than 10 carriers accelerate 5G network deployment.

For indoor scenarios, Distributed MIMO greatly improves 5G network capacity and user experience. This solution has been deployed by the National Centre for the Performing Arts of China, the Wuhan Metro, and the Chengdu East Railway Station, significantly improving 5G performance and experience.

As a 5G standard essential patent, the solution increases the proportion of inter-RAT spectrum resource sharing from 40% (during the 4G era) to 90% (for 5G). It is the industry's only millisecond-level dynamic spectrum sharing solution that's ready for commercial use. The solution is currently used commercially in 26 networks, with unmatched performance and deployment speeds.

This innovation enables deterministically low latency and low jitter for large-scale packet networks. Based on this, we were able to release the industry's first and only deterministic IP network solution. It has passed the world's first trial of deterministic wide area networks and wide-area cloud-based PLCs, enabling reliable, microsecond-level latency in remote industrial control scenarios. As the technological foundation, deterministic IP networks support emerging industry applications such as industrial control, connected vehicles, smart grids, telemedicine, Cloud VR, and holographic communications that require deterministically low latency and low jitter, enabling the next-generation industrial Internet.

This invention makes significant breakthroughs in storage speed, stability, and scalability, enabling a symmetric active-active high-end storage architecture with zero service interruptions. This architecture supports sub-millisecond latency and storage speeds of over 10 million input/output operations per second (IOPS). These features will allow it to play a crucial role in addressing key challenges that arise in various real-world scenarios that can lead to data bursts. Examples include the data bursts seen in the train booking systems during the Chinese Spring Festival, the storage of massive amounts of data in a short period of time on controlled explosions during oil exploration, and the surges in orders that regularly strain banking systems during the busiest online shopping days.

This patent portfolio reveals the innovative network and memory technologies widely used in Huawei device products, including the multi-network aggregation technology, LinkTurbo, for Wi-Fi and cellular networks (such as 4G and 5G mobile networks), and the dynamic memory-expansion technology HyperHold. LinkTurbo uses software only to stack and integrate the capabilities of multiple network modules. Using complementary multi-network capabilities, LinkTurbo improves concurrent download speeds by 83%, reduces gaming latency by 69%, and cuts video buffering time by 87% when there is limited network access or jitter, or for mission-critical services. HyperHold significantly expands available memory, reduces frame freezing caused by frequent memory swapping, improves basic read/write performance, and delivers an extraordinarily smooth user experience.