ARM: Pain And Happiness, Accelerate Marginal Artificial Intelligence Innovation

The happiness isAI will rewrite everything to bring opportunities; the pain is that the evolution of AI is too fast, and the models and applications are endless.

existAt the ELEXCON2024 Shenzhen International Electronics Exhibition held on August 27, ARM VP, China GTM, LOT & AMP; Embedded Line of Business Chloe MA delivered a speech on "Accelerating the Edge Artificial Intelligence Innovation".

We captured the following key information:

The speed of AI landing on the edge is amazing

ARM empowering the three aspects of the innovation focus of the edge AI: hardware, software and ecology

Will not stop:ARM Flexible Access

along withThe rise of ChatGPT and the rapid development of large models and generation AI (Genai), large models help the automation transformation of thousands of industries and the improvement of new productivity.

AI is changing people's way of life and work at an unprecedented speed.From Netflix to TIKTOK (Douyin), and to ChatGPT, the speed of user growth continues to refresh the record.In particular, ChatGPT reached one million users in five days, and easily exceeded the 100 million levels in two months. This feat fully demonstrated the huge potential and market acceptance of AI technology.With the continuous breakthrough of AI models and computing chip technology, the AI computing power is rapidly infiltrating the edge devices, bringing revolutionary changes to new computing fields such as smartphones, PCs, and autonomous driving, robots.

Chloe MA said that behind this rapid development is AI's strong technical support.The breakthroughs of AI models and computing chip technology, and the demands of real -time response costs, user experience, and security privacy all made the AI computing power quickly rush to the edge equipment.

MarginThe development of AI, the current sensor mainly undertakes information collection tasks, and the edge network is not supported by AI models, but this situation is rapidly changing.With the continuous improvement of the computing power of the AI PC and IoT gateway, the edge AI has shown huge potential.In the future, sensors will support more modes and have simple intelligent data analysis capabilities, and complex data analysis and decision -making planning will achieve collaborative scheduling cloud, edge, and end, and the overall AI computing power level will be significantly improved.

As the world's leading computing platform provider, ARM is actively empowering the innovation of marginal AI.ARM continues to make efforts in three aspects: hardware, software and ecology, and launched a number of products including Cortex-M embedded processor, Cortex-A application processor, and ETHOS edge AI accelerator to meet the AI calculations of different scenes.Demand.At the same time, ARM has also strengthened the CPU's vector or matrix processing capacity, and increases investment in software and tool chain to help developers easily release the potential of AI hardware.

 Hardware innovation: strengthen computing power, guarantee security

Chloe MA emphasizes the deep accumulation of ARM in the field of hardware.She pointed out that ARM's Cortex-M embedded processor and Cortex-A application processor provided strong computing power support for the edge AI.Facing the security challenges brought by the interconnection of all things, ARM launched security technologies such as Trustzone, providing solid protection barriers for sensitive data and operations.In addition, in response to the surge in the demand for AI reasoning, ARM not only enhanced the CPU's vector and matrix processing capacity, but also innovatively launched the ETHOS edge series AI accelerator to further meet high performance and complexity.Requirements for AI workload.

ARM's Ethos NPU is the world's first embedded AI accelerator.It has been adopted by leading manufacturers such as Yingfei, NXP, Olive, Qitian Optoelectronics, Alif Semiconductor, and Synaptics.From the first generation of ETHOS-U55 to the latest U85, it not only improved significantly in performance and energy efficiency, but also achieved native support for new operators and neural networks such as new operators and TRANSFORMER.ETHOS series NPU provides consistent tools, so it can bring seamless developer experience to the edge AI ecosystem, enabling partners to repeatedly use investment on AI AI.

 Software and toolchain: Empower developers, optimizeAI deployment

In terms of software and toolchain,Chloe Ma introduced the comprehensive support provided by ARM for developers.From ARM virtual hardware to IP Explorer, and then to ML Inflonce Adviser, ML ZOO model library and ML Evaluation Kit, ARM has built an efficient process from design to deployment for developers.

at the same time,ARM also supports a variety of mainstream ML development frameworks, including Pywatch, ONMX, TensorFlow, TensorFlow Lite, and Baidu flying paddle in China. It also provides optimized ML compilers and Runtime for different hardware targets to ensure that the ML model is on different hardware on different hardwareThe optimized operation.It is particularly worth mentioning that the open standards of TOSA (Tensor Operator Set Architecture) advocated by ARM are gradually simplifying the deployment process of training environments to different hardware platforms, paving the way for the popularization of AI technology.

In the near futureARM and Meta cooperate to introduce PyTorch into the edge -based motion and embedded platforms based on the ARM architecture through ExecuteTorch.The cooperation between ARM and META is to rely on ARM's huge investment in TOSA.

 Ecological construction: Accelerate chip development and promote market applications

In terms of building an open ecology,ARM is committed to the extensive processor IP and system IP to achieve pre -integration and verification, and provides partners in the form of a Core Stone reference design subsystem.These reference designs not only shorten the development cycle, reduce development costs, but also help partners to quickly launch marginal AI products that meet market demand.At the same time, ARM also cooperated with Nvidia to apply innovative technologies such as Etehos NPU to actual products, and further promoted the commercialization process of marginal AI technology.

Chloe MA said that only ARM can provide the characteristics and functions required from cloud to end, modern agile development deployment process.Let everyone transform AI on the consistent architecture of mass -based verification and a unified tool chain.

In the process of the rise of the big model in the past one or two years,ARM saw the rapid development of the entire industry.The industry is in a state of pain and happiness.The happiness is that AI will rewrite everything and bring opportunities.AI basically allows chip software in all fields to reorganize.It hurts that the evolution of AI is too fast, and the models and applications are endless.For chip design companies, how to choose tracks and specifications, accelerate the development and listing of products, and generate differentiation.In this AI gold rush tide, the ARM computing platform can provide a set of simple and universal universal weapons.Being able to make a good component from the cloud to the end, let everyone use it to use the hand -made AI gold mines.

In order to make the majority of developers access and use more easilyARM technology reduces the development threshold, ARM has continuously explored and innovated business models in the past few years.Chloe MA specifically introduced the ARM Flexible Access (AFA) program. This model has been widely praised overseas and officially implemented in the Chinese market.AFA is similar to Costco's membership services. Developers only need to pay extremely low annual fees, and they can be unlimited without restrictionaccessThe rich resources provided by ARM include Cortex-A, R, M series CPU processors, ETHOS-U NPU, ML GPU and ISP reference design, as well as simulation platforms and tool chains.Developers can independently choose the appropriate IP combination according to their own product needs to design and optimize, and settle the authorization and follow -up costs of the selected IP combination when they are mass -produced or stream.This flexible business model greatly reduced development costs, accelerated the process of product listing, and was warmly welcomed by many developers including Raspberry Pi.

at last,Chloe MA summarizes that the popularity of artificial intelligence and the rapid development of large models and generative AI have brought unprecedented opportunities to partners of the Internet of Things ecosystem.With its rich ecological resources, leading technical strength and extensive industry influence, ARM will fully support partners to seize the opportunities of marginal AI and promote the widespread application of intelligent technology.She firmly believes that only the close combination of cloud, edge, and end can build an omnipresent smart world.