华为悬赏单项最高 100 万元攻克存储技术难题，第六届奥林帕斯奖启动全球征集

Group 1 - The OlympusMons Awards, established by Huawei in 2019, aim to encourage global researchers in the field of data storage to advance fundamental theoretical research and overcome key technical challenges, facilitating the industrialization of research outcomes and promoting collaboration between academia and industry [3][44] - Since its inception, the awards have attracted over 320 scholars from 12 countries, resulting in the awarding of 6 OlympusMons Awards and 18 Pioneer Awards [3][44] - In 2025, the program will offer 2 OlympusMons Awards, each with a prize of 1 million yuan, and 5 Pioneer Awards, each with a prize of 200,000 yuan, providing winners with opportunities for technical exchange and research support from Huawei [3][44] Group 2 - The 2025 OlympusMons challenges focus on innovative medium technologies for the AI era, addressing the increasing costs of processing cold, warm, and hot data, and promoting high-performance, large-capacity, and cost-effective storage systems [5][46] - The first challenge involves SSD-based computing and efficient indexing technology, emphasizing the need for high-density information recording and hierarchical large memory innovations [5][46] Group 3 - The bandwidth specifications for various storage technologies are as follows: HBM (1-4 TB/s), DRAM (100 GB/s), large capacity (7-32 GB/s), and SSD (costing $0.06 to $15 per GB) [6][47] - The challenges include near-storage computing power limitations due to high arithmetic intensity of operators, which creates performance bottlenecks in storage-computing integration architectures [8][49] - The expansion of SSD capacity leads to increased memory demands for FTL mapping tables, raising power consumption, costs, and reliability issues, which are core technical bottlenecks for SSD scalability [8][49] Group 4 - The second challenge focuses on storage channel modulation coding technology for ultra-high recording density, addressing the need to enhance storage channel capacity density as data scales reach the YB level [11][52] - The goal is to achieve a recording density gain of G≥3 and develop high-reliability detection and decoding technologies with an error rate deterioration of less than 10% [18][59] Group 5 - The third challenge targets hierarchical large memory network protocols and IO path optimization technologies, essential for meeting the demands of Agentic AI, which requires high bandwidth, low latency, and high IOPS [20][60] - The development of a hierarchical large memory system is proposed to overcome limitations and provide a robust storage foundation for Agentic AI applications [20][62] Group 6 - The fourth challenge involves knowledge extraction, multi-modal data representation, and knowledge retrieval technologies, addressing issues such as semantic loss during data transformation and the complexity of multi-modal knowledge alignment [28][72] - The aim is to enhance knowledge extraction completeness and improve semantic understanding consistency, achieving high accuracy in knowledge representation and retrieval [33][73] Group 7 - The fifth challenge focuses on semantic information condensation technology for efficient inference in large models, aiming for an end-to-end storage-computing collaborative information condensation system [34][78] - The objective is to achieve a compression ratio of ≥20 times while maintaining inference accuracy loss of less than 1% and improving throughput by over 5 times [38][79]