UK processor and peripheral IP developer ARM is launching a series of virtual hardware tools intended to ease software development for the AI-enhanced Internet of Things.
A new technology can take five years to get to market in the IoT and then there is inefficient software development with a lack of scale, using software across multiple platforms. The ARM ‘Total Solutions for IoT’ combines the virtual hardware models that are used by chip designers with cloud-based tools, security and software abstraction technologies it calls Project Centauri.
“We will fundamentally transform how software is developed for the IoT,” said Mohamed Awad VP IoT and Embedded at ARM. “This is a direct result of the ecosystem coming to us to solve key problems.”
Awad points to 3m smartphone apps able to run on thousands of different smartphones, although this is down to just two operating systems, iOS and Linux-based Android that run on the ARM Cortex-A application processors. “We want bring the same economies of scale that apps bring to smartphones to the IoT,” he said. “Through a radical change in how systems are designed, Arm is uniquely positioned to fuel a new IoT economy that rivals the shape, speed and size of the smartphone industry’s app economy,”
The aim is to deliver the tools as apps running on Linux for mainstream software developers with hardware IP running in the cloud, software, machine learning (ML) models, advanced tools such as the new Virtual Hardware Targets, application specific reference code and support
“I’m not talking about the typical hardware players, I’m talking about IoT operators and millions of software developers who may not be embedded software developers,” said Awad. “We have been down this road before in mobile and infrastructure and this allows hardware and software to be developed in parallel with continuous integraiton and scalable cloud-based testing with well defined APIs and standards,” he said.
ARM will supply virtual hardware models for its Corstone sub-system that combines processor cores with machine learning accelerators and the AI frameworks that run on the models. The first platform is the Corstone-300 that combines an ARM Cortex M55 microcontroller core and Ethos U55 AI accelerator core, but the company plans to expand this to the M33 (in the Corstone-200) and A53 cores as well as new Cortex-M cores codenamed Olympus optimised for object recognition and other AI applications
“What we have done is created virtual drivers that can be customised by Perl scripts,” he said. “In the first instance there’s an Ethernet driver for connectivity and an audio driver to take a microphone input, and we expect partners to add peripherals,” said Awad. “Hardware Development Kits are about the low level driver development and there will continue to be a need for that.”
“This represents an entirely new way for software developers to innovate, entirely in the cloud,” said Awad. “This will collapse IoT development from five years to three. We are doubling down and increasing our investment in Corstone as foundational, everything we do will start with Corstone.”