This week I’m attending the 15th annual Arm TechCon event in San Jose, CA. TechCon is Arm’s premier industry event, where developers, analysts, and media gather from all around to hear the latest from the semiconductor giant. Over the last several years, Arm has been working to stake out its claim in the so-called “5th wave of computing”—the place where AI, IoT, and 5G converge, and the new data consumption models that are arising as a result (see my coverage of last year’s conference here if interested). This story continued at Arm TechCon 2019. Let’s take a look at what all was announced and some of my hot takes.
Introducing the AVCC
One of the big announcements from Day 1 of the event was the launch of the Autonomous Vehicle Computing Consortium, or AVCC, a group united by the goal of creating a common computing platform for autonomous vehicles. In addition to Arm, founding members include the likes of Bosch, Continental, DENSO, General Motors, NVIDIA, NXP Semiconductors, and Toyota. Through collaboration, the consortium seeks to speed the arrival of safe and affordable autonomous vehicles at scale. According to the group, the first leg of the journey is the development of a set of recommendations for system architecture and computing platform that takes into account both vehicle-specific requirements and the performance requirements of autonomous systems.
I personally don’t usually get too excited about consortiums since they typically talk a lot and do very little. That said, the AVCC looks interesting—where companies have to work together to succeed usually determines success. It would be nice to see other industry leaders involved though, such as Qualcomm, Intel’s MobileEye and ON Semiconductor. If not the AVCC could look more like a chaebol of like-minded suppliers than an industry group. Software vendors will be important as well. I look forward to more details as they come out and expect to see the first output end next year to impact 2025 model designs.
One of the big picture items from the conference was the unveiling of Arm’s new Total Compute approach to IP design, which the company says will put a deeper focus on performance, security and developer access in the interest of driving more immersive digital experiences in fields like AI, XR, and IoT. Arm says this represents a shift towards a more use-case and experience-driven portfolio, through optimization across the company’s IP, software and tools. In order to accomplish this, the company says it will rely on software and tools such as Arm NN and the Arm Compute Library, as well as open standards and the open-source community.
As part of this new Total Compute approach, Arm shared details on the work it is doing to increase its CPU coverage for machine learning. It’s been building on these capabilities since the launch of the Cortex-A73. Arm says that its next-generation Cortex CPU, Matterhorn, will boast double the machine learning performance of previous generations, thanks to its new Matrix Multiply feature.
Arm also shared its vision of Total Compute security, which consists of three different layers: application security, which protects apps and data, in-process security, which detects and provides defense against vulnerabilities, and lastly platform security, which Arm says defragments via “transparent and accessible security standards.”
Additionally, Arm announced a broadening of its partnership with Unity Technologies, a video game software development company currently responsible for roughly 70% of VR content available today. This partnership was conceived to further the company’s goal of making Total Compute the developer “platform of choice” for creating immersive content. Arm says that the two companies together will work to optimize performance on Arm-based silicon—SoCs, CPUs, and GPUs—for Unity Technologies’ real-time 3D development platform.
All in all, Arm’s Total Compute strategy reminds me of heterogeneous computing on steroids, plus security—more of a platform approach than anything else. This is Arm raising the conversation to the platform level, putting it more in competition with platform IP companies like Qualcomm.
On Day 2 of the event, Arm announced Project Cassini, its new reference design for edge compute including the recommended hardware, software, and security. I believe Arm “gets” the evolution of the network edge as it transforms from a dumb control layer to a smart AI edge. Arm says Project Cassini will help ensure a secure, cloud-native experience across the diverse edge ecosystem, through the establishment of platform standards and reference systems. As part of this, Arm announced it is expanding its Platform Security Architecture from constrained devices to IoT devices and the infrastructure edge.
Last year, Arm’s Drew Henry told the audience to “watch the space” for cloud announcements with Arm infrastructure and like magic, Amazon Web Services (AWS) A1 instances magically arrived at re: Invent. This year, Henry advised the audience to “watch this space” for ecosystem announcements, and I surely will.
Arm getting easier to do business with
In the grand scheme, I believe Arm doesn’t make an amount of money from its IP licenses commensurate with its importance and investment. But there has been some industry chatter about the desire for Arm to be more flexible in its licensing and a customer’s custom instructions, particularly in lower-level controller and IoT applications.
One of the ways Arm is making itself easier to do business with is by allowing developers to try its technology before making a full purchase. Consider this “try before you buy”. According to Arm, businesses will only be charged for IP used at tape-out—which means no license fee if projects are paused, changed, or stopped. This was announced a few months back and applies to IoT IP, not smartphone, PC, automotive or infrastructure IP, but does constitute 75% of its licenses.
Arm also announced new, free custom instructions feature for the Cortex-M33 CPU (and all M series moving forward). This feature will allow SoC designers to use their own instructions for embedded and IoT apps while avoiding the risk of software fragmentation. Consider this “BYOI” or “bring your own instructions”. I was a bit disappointed Arm didn’t provide or even hint to an example of a custom instruction, but I know the company doesn’t want to hint, as it could potentially “out” a customer.
Consequently, all of this blunts some of the RISC-V value proposition and raises the stakes for the entire industry.
All in all, there was plenty of new information at Arm TechCon 2019 to chew on. I wrote last year that Arm TechCon was the place to be for anyone involved in IoT in any way, and that continues to be true. I look forward to seeing how Arm’s Total Compute initiative evolves, and what comes of the AVCC consortium. Generally speaking, I was glad to see the emphasis Arm put on partnerships and building out its ecosystem this year in software. Arm is slowly but surely transforming into a platform IP company. Nice work, Arm—I’ll see you at TechCon 2020.
Note: Moor Insights & Strategy writers and editors may have contributed to this article.