webinar interview<\/span><\/strong><\/a> with Alif Semiconductor\u2019s founder and president, Reza Kazerounian. Reza is a veteran semiconductor industry executive who has seen various markets for microcontrollers emerge and evolve in his career at Freescale, Atmel, STMicroelectronics and elsewhere. Generative AI is the latest use case for MCUs,and is perhaps set to be the fastest-growing.<\/p>\n\n\n\nBut Reza sees danger lurking in the way that some MCU manufacturers are integrating proprietary neural processing units (NPUs) for AI acceleration into their devices. He sees two main risks in the proprietary approach.<\/p>\n\n\n\n
First, he says, \u2018developers need a rich ecosystem of model frameworks, toolchains and data sets to be able implement generative AI on highly constrained embedded devices. No single MCU or semiconductor manufacturer is going to provide the whole of this ecosystem for their customers \u2013 the ecosystem draws in contributions from across the range of technology companies.\u2019 Over time, Reza says, we will see much more standardization of everything to do with generative AI, including this ecosystem of tools and frameworks.<\/p>\n\n\n\n
But a proprietary NPU is likely to support only those tools and frameworks which are provided directly by the MCU manufacturer itself, or by companies with which the MCU manufacturer is partnered. But these are not necessarily the products on which the rest of the embedded community settles as its preferred choice, leaving users of proprietary hardware stranded with a diminishing set of poorly supported tools and resources.<\/p>\n\n\n\n
The second main risk arises from the incredibly rapid development of AI technology. As new generative AI capabilities emerge, IoT and edge device manufacturers will want to migrate from one generation of MCU hardware to the next to benefit from improvements in latency and power consumption, and to handle bigger and better models. When they do so, they will want to port AI algorithms from legacy designs.<\/p>\n\n\n\n
But algorithms optimized for one proprietary hardware platform are unlikely to work on new, improved hardware. Again, the developer risks being stranded on a hardware platform which has been superseded by competitive offerings.<\/p>\n\n\n\n
The answer, says Reza, is to embrace standardization. \u2018With a non-proprietary NPU based on an open standard, developers can easily migrate their algorithm from one hardware platform to the next generation, and retain the value of their investment in existing AI algorithms.\u2019<\/p>\n\n\n\n
Alif has always supported standardization in its Ensemble family of MCUs and fusion processors: the first generation E1\/E3\/E5\/E7 devices featured the Arm\u00ae Ethos\u2122-U55 NPU core, and now the second generation, the E4, E6 and E8, have the Ethos-U85 NPU. This standard Arm architecture is employed across the MCU industry, so OEMs can develop AI applications based on the Ensemble family safe in the knowledge that their algorithms will work on any Ethos-based hardware.<\/p>\n\n\n\n
And because the Ethos NPU is the emerging standard for edge AI acceleration, it will attract the widest support from toolchain suppliers, model frameworks and other parts of the generative AI ecosystem.<\/p>\n\n\n\n
So this is why the big lesson from this new webinar is, avoid the proprietary hardware trap!<\/p>\n\n\n\n
<\/p>\n","protected":false},"excerpt":{"rendered":"
Getting stuck in a hardware trap is one of the biggest risks facing embedded system developers as they embark on their journey into generative AI on edge or IoT devices. This is the message from a recent webinar interview with Alif Semiconductor\u2019s founder and president, Reza Kazerounian. Reza is a veteran semiconductor industry executive who<\/p>\n","protected":false},"author":1,"featured_media":1000002635,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"content-type":"","inline_featured_image":false,"footnotes":""},"categories":[107],"tags":[],"class_list":["post-1000002633","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-generative-ai"],"acf":[],"_links":{"self":[{"href":"https:\/\/alifsemi.com\/wp-json\/wp\/v2\/posts\/1000002633","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/alifsemi.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/alifsemi.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/alifsemi.com\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/alifsemi.com\/wp-json\/wp\/v2\/comments?post=1000002633"}],"version-history":[{"count":2,"href":"https:\/\/alifsemi.com\/wp-json\/wp\/v2\/posts\/1000002633\/revisions"}],"predecessor-version":[{"id":1000002637,"href":"https:\/\/alifsemi.com\/wp-json\/wp\/v2\/posts\/1000002633\/revisions\/1000002637"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/alifsemi.com\/wp-json\/wp\/v2\/media\/1000002635"}],"wp:attachment":[{"href":"https:\/\/alifsemi.com\/wp-json\/wp\/v2\/media?parent=1000002633"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/alifsemi.com\/wp-json\/wp\/v2\/categories?post=1000002633"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/alifsemi.com\/wp-json\/wp\/v2\/tags?post=1000002633"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}