Why ARM wants to do more
However, Moore’s Law and the passage of time drives change as does corporate success, and for ARM the speed of layering up of technical activity appears to be accelerating. ARM is now a subsidiary of Japan’s Softbank Group and one of the reasons Softbank made its $32 billion acquisition of ARM in 2016 was because it liked the company’s active and comprehensive position on the Internet of Things (IoT). ARM is now a solutions company according to Ian Smythe, senior director of marketing programs at ARM; and that would also seem to reflect that things are changing.
Time was when ARM defined itself as a means by which the semiconductor industry could avoid repeatedly re-inventing the processor. Or looked at another way, ARM was akin to an outsourced processor R&D group for industry players. As a result, ARM designed processor cores and licensed them to multiple semiconductor companies – and that was it. The semiconductor companies saved money and ARM made a little.
To help it sign up semiconductor companies as licensees ARM pioneered a partnership model that extended up and down the value chain and Robin Saxby (then CEO and now Sir Robin Saxby) was notorious for cramming as many company logos as possible on to a slide, as he developed the now commonplace idea of “the ecosystem.” And ARM made it clear it would never compete with its semiconductor licensees.
But over time complexity has arrived not only in-core and on-chip in terms of what can be manufactured monolithically, but also in terms of the systems and applications these SoCs must serve.
A move into physical IP – with the acquisition of Artisan in 2004 – may have been seen by most as arcane and non-threatening; a good thing to help tune CPU cores around power, performance and area. But ARM has also moved into adjacent markets such as graphics processing units (GPUs) under the Mali brand and short-range radio via its Cordio IP family.
In Cordio it has blocks for the IEEE 802.15.4 standard, which is the basis for ZigBee, 6LowPan, Thread and others. It also has a Bluetooth 5 capable core and therefore almost by definition a core that can do both. This has accompanied dramatic success for the Cortex-M series of cores aimed at providing the intelligence for the world’s microcontrollers.
ARM drove the creation of Linaro a collaborative engineering organization, which helps develop open-source software for ARM cores and ARM-based chips and inhibit Linux fragmentation on ARM. And in a similar vein it should be noted that ARM is a founder member of home-oriented Thread Group.
And with particular regard to the Internet of Things (IoT) ARM is now highly active. There are such developments as driving the creation of the mbed standard for microcontroller boards and a community around them and the creation of the mbed OS.
Next: mbed at Embedded World
At the Embedded World exhibition in Nuremburg, Germany, ARM rolled out the latest release of the mbed OS – version 5.4 – with new features for applications such as connected buildings. With OS 5.4 developers have radio choices for cellular, Wi-Fi, 802.15.4, Bluetooth, Thread or sub-GHz 6LoWPAN. The ARM engineers behind mbed OS demonstrated NB-IoT connectivity at Mobile World Congress in February 2017.
And clearly for most of these choices ARM and semiconductor partners offer a mix of royalty-free and royalty-payable technology options.
In response to the suggestion that ARM was now doing a lot of forward thinking for the technology industry in general at chip-, software- and system-level, Ian Smythe, senior director of marketing programs at ARM said: “We are doing more of the firmware and software within ARM. But the reason is we need to understand the application profile on the way to creating the solution.”
Smythe added that the Internet of Things marked a watershed because it – and much of electronics equipment going forward – will be either connectable to the Internet, or permanently connected to it. He said that interconnectedness meant that such issues as security had to be engineered at the node and in the data center and from end-to-end. And that meant working on things like Thread, mbed OS and MCU cores together.
The Internet of Things may well be a game changer but does that make it a driver of cultural change within ARM? Smythe speaks of “solutions” rather than IP cores and solutions increasingly means software and core platforms which will be delivered by one semiconductor company or another.
Some might argue it is starting to matter less and less who the nominal chip supplier is. The chip and the firmware/software will have been engineered substantially by ARM and the chip made at one of a very few foundries.
It should also be noted that at Mobile World Congress in February 2017 ARM launched a machine learning software library that provides superior performance to OpenCV on Neon. The ARM Compute Library runs on any ARMv7, ARMv8 CPUs and BiFrost and Midgard GPUs and is free on liberal open-source license (ARM’s soft launch for machine learning library).
Next: Not your father’s ARM
Smythe said it is long time since ARM had been just a core licensing company. Smythe said ARM’s method of working now was to look ahead in application sectors and try and forecast what will be the next 100 million unit markets and then try with partners to engineer detailed solutions to enable those things to come to pass. That might be autonomous driving, augmented and virtual reality, drones and wireless sensor networks, smart homes, smart buildings, smart transport and so on. “Increasingly there has to be end-to-end thinking,” Smythe said.
Smythe denied that the result was ARM taking business from software-based ecosystem partners. “FreeRTOS, ThreadX and others are still out there.” Smythe also argued that because mbed OS is a free offering it is not the same as competing directly.
Smythe said it was really a case of ARM mapping out use cases to make sure that the end-to-end thinking was being done. That includes things like software updates over the air, wireless protocol stacks, shuffling between applications and radios, the implications for security so on. It also reflects different engineering competences within customers, Smythe said.
“For companies that know their markets and have strong engineering teams we may have minimal involvement.” But there are new challenges and new entrants and some customers value how much can be done for them and how much time saved. Why would they want to re-engineer it, argued Smythe.
However, there are certain red lines within the ARM universe. And its neutrality towards its semiconductor licensing partners and not competing with them is one of those, Smythe indicated.
Nonetheless, ARM continues to move forward and feed. Its $350 million acquisition of Apical Ltd. in May 2016 being one taking ARM into imaging and computer vision (see ARM buys Apical: all IP and running licensing deals for $350 million). The fact that ARM proved to be $30 billion meal for Softbank Group in 2016 may, if anything, increase the rate at which ARM seeks out and swallows up complementary companies and technologies (see ARM agrees to be bought by Japan’s Softbank).
Prior to its acquisition Apical’s IP blocks had been used in more than 1.5 billion smartphones through licensing deals with chip companies such as Samsung and Texas Instruments. ARM’s opportunity is to marry this IP with its ARM Mali graphics, display and video processor roadmap to create cores and firmware for next-generation vehicles, security systems, robotics, smart buildings and any industrial or retail application where intelligent image processing is needed.
Next: Machine learning
Which brings us round to machine learning. Surely the ARM Compute Library will be a temporary position while company works out one or more optimized or configurable machine learning cores. Adjustable balances of processor-memory resources created for different types of neural network and more general machine learning could be expected to provide greater energy efficiency than library routines running on a CPU or GPU. But the devil is in the detail and ARM’s business is to determine when it becomes worthwhile to turn such a technology into a licensible product.
Smythe acknowledges that in general such things are getting closer. He points to the scalable vector extension (SVE) capability provided within the ARMv8 architecture. “It fits for computer vision,” he asserts. This feels like a potentiality rather than the sort of hardware-firmware platform and end-to-end thinking Smythe has been espousing. “Can I say what our answer [for machine learning] will be in hardware. No!” said Smythe.
But that doesn’t mean that ARM does not have an answer and that it will not seek to make it as successful as possible when the timing is right.
It seems ARM always wants to do more. For ARM the more difficult task may be growing its engineering output while remaining everybody’s friend and avoiding the unpopular role that Intel fell into more than two decades ago; that of the technology supplier tail that wags the system company dogs. Avoiding that plight was easier when ARM was a just an IP core licensor and less successful than it is today.
Related links and articles: