[et_pb_section fb_built=”1″ _builder_version=”4.16″ custom_margin=”0px||||false|false” custom_padding=”0px||||false|false” locked=”off” global_colors_info=”{}”][et_pb_row _builder_version=”4.16″ background_size=”initial” background_position=”top_left” background_repeat=”repeat” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.16″ custom_padding=”|||” global_colors_info=”{}” custom_padding__hover=”|||”][et_pb_text _builder_version=”4.27.6″ background_size=”initial” background_position=”top_left” background_repeat=”repeat” hover_enabled=”0″ global_colors_info=”{}” sticky_enabled=”0″]<\/p>\n
AMD recently announced its acquisition of MEXT. This is very exciting for us at Phison as it validates our position on an evolving trend: the industry is redefining the role of NAND flash as an indispensable component of the memory hierarchy in the AI era.<\/p>\n
For over two decades, NAND flash has primarily been positioned as a storage medium. However, the emergence of generative AI has altered this industry landscape. As AI models and KV caches grow in size, DRAM and HBM have increasingly become the most expensive and scarce resources in AI systems. Consequently, finding ways to effectively offload more data to SSDs and flash has become a shared priority across the industry.<\/p>\n
The technology developed by MEXT is an AI-managed predictive swap mechanism that runs on Linux. It identifies cold data and moves it to SSDs, leaving more room for hot data in DRAM. The real differentiation comes from accurately predicting what needs to return to memory before it\u2019s needed, helping ensure that applications can access relevant data from DRAM. However, enabling flash to fully augment DRAM cannot be achieved through software alone; it requires deep collaboration with SSD providers.<\/p>\n
Both MEXT and Pascari aiDAPTIV\u2122 are complementary and can easily collaborate. The MEXT technology focuses on predictive memory tiering for broad compute workloads, while aiDAPTIV works with inference and training runtimes using AI-workload-aware memory management. Another opportunity for collaboration comes from Phison\u2019s work to optimize SSD behavior for AI memory and cache workloads and the integration of AI accelerators directly into the SSD.<\/p>\n
This continuous expansion of NAND capabilities also closely mirrors the concept behind NVIDIA\u2019s CUDA libraries. From graphics rendering, and High-Performance Computing (HPC) analysis, to machine learning and generative AI, the introduction of each new CUDA library creates a new application scenario for GPUs, further driving up their value and demand.<\/p>\n
Similarly, Phison\u2019s aiDAPTIV technology is designed to continuously increase utilization and expand the application scenarios for flash within systems. If the objective of CUDA libraries is to broaden the use cases for GPUs, then aiDAPTIV helps broaden the use cases for flash as an additional memory tier.<\/p>\n
Phison has developed aiDAPTIV capabilities that help data traditionally requiring DRAM or HBM to be effectively managed using flash:<\/p>\n
Through these capabilities, flash transcends its traditional role as mere data storage. It can now directly participate in diverse workloads such as AI model training, inference, KV cache management, and generative AI applications on Linux and Windows. aiDAPTIV also supports a wide range of processing elements (CPUs, GPUs, XPUs), allowing flash capabilities to be integrated into a wide array of computing architectures.<\/p>\n
This is precisely where Phison holds a distinct advantage.<\/p>\n
Phison has been deeply rooted in the NAND flash industry for over two decades, accumulating profound technical expertise in controllers, firmware, reliability, and performance. Simultaneously, through the development of aiDAPTIV, we have gained extensive practical experience across multiple scenarios, including AI training, AI inference, KV cache reuse, generative AI, and memory extension.<\/p>\n
We anticipate that an increasing number of innovative software companies \u2013 like MEXT \u2013 will venture into NAND flash application innovation. Phison remains committed to close collaboration with these solution providers who are expanding the use-cases for flash. By combining our partners’ innovative capabilities in software and system architecture with Phison\u2019s deep expertise in flash technology, we aim to jointly create highly competitive, integrated hardware-software solutions.<\/p>\n
Through this collaborative model, Phison not only delivers industry-leading NAND flash reliability and performance but also helps partners maximize the value of their software technologies. Ultimately, this enables us to provide customers with complete products that offer the best balance of performance, reliability, and Total Cost of Ownership (TCO).<\/p>\n
Just as CUDA has continuously expanded the use cases for GPUs, Phison will continue to broaden the application scope of flash through the expansion of aiDAPTIV capabilities.<\/p>\n
We believe that the greatest future opportunity for NAND flash lies not merely in storing more data, but enabling more capable and memory-intensive workloads. This represents the next major growth trajectory for the flash industry.<\/p>\n
[\/et_pb_text][\/et_pb_column][\/et_pb_row][\/et_pb_section]<\/p>\n","protected":false},"excerpt":{"rendered":"
AMD recently announced its acquisition of MEXT. This is very exciting for us at Phison as it validates our position on an evolving trend: the industry is redefining the role of NAND flash as an indispensable component of the memory hierarchy in the AI era. For over two decades, NAND flash has primarily been positioned […]<\/p>\n","protected":false},"author":81,"featured_media":89708,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_et_pb_use_builder":"on","_et_pb_old_content":"","_et_gb_content_width":"","inline_featured_image":false,"footnotes":""},"categories":[23,116,118],"tags":[22],"class_list":["post-89705","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-all-posts","category-featured","category-thought-leadership","tag-long-content"],"acf":[],"_links":{"self":[{"href":"https:\/\/phisonblog.com\/zh\/wp-json\/wp\/v2\/posts\/89705","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/phisonblog.com\/zh\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/phisonblog.com\/zh\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/phisonblog.com\/zh\/wp-json\/wp\/v2\/users\/81"}],"replies":[{"embeddable":true,"href":"https:\/\/phisonblog.com\/zh\/wp-json\/wp\/v2\/comments?post=89705"}],"version-history":[{"count":5,"href":"https:\/\/phisonblog.com\/zh\/wp-json\/wp\/v2\/posts\/89705\/revisions"}],"predecessor-version":[{"id":89720,"href":"https:\/\/phisonblog.com\/zh\/wp-json\/wp\/v2\/posts\/89705\/revisions\/89720"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/phisonblog.com\/zh\/wp-json\/wp\/v2\/media\/89708"}],"wp:attachment":[{"href":"https:\/\/phisonblog.com\/zh\/wp-json\/wp\/v2\/media?parent=89705"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/phisonblog.com\/zh\/wp-json\/wp\/v2\/categories?post=89705"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/phisonblog.com\/zh\/wp-json\/wp\/v2\/tags?post=89705"}],"curies":[{"name":"\u53ef\u6e7f\u6027\u7c89\u5242","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}