The semiconductor industry was once the engine of hi-tech innovation and Silicon Valley buzzed with start-ups and new ideas. But that era may have ended. Silicon Valley is less and less about silicon; the costs of developing new chips are huge and scare off most venture capitalists; the established companies are in a cycle of consolidation; even Moore’s Law may be close to running its course. Yet in the mobile world, the need for new ideas and fierce competition is as urgent as ever – just as the networks have to evolve to support a wide range of behaviors from hyperscale cloud services to the IoT, so the underlying processors, modems and sensors need to do so too.
ARM’s CTO, Mike Muller, threw cold water on the fears that innovation is running out in the chip space. In the run-up to the processor IP firm’s annual conference, TechCon, he said: “I still think there’s an incredible amount of innovation ahead of us no matter what people say about flattening of Moore’s Law.” Certainly ARM is not content to sit in its comfort zone, and is pushing the boundaries of its Cortex architectures into every areas from cloud servers and Cloud-RAN, to consumer equipment, to IoT devices. In that endeavour, it is head-to-head with Intel, and the two companies are constantly upstaging one another. In the week of ARM’s TechCon, for instance, Intel drove home the reality of its rising presence in the mobile industry by announcing new alliances with Verizon and America Movil – the kind of operators which would more commonly work with ARM licensees such as Qualcomm.
But while this ARM/Intel battle royal is certainly driving both these very different companies to enhance their R&D efforts, is that really enough to guarantee the levels of innovation which are needed in the new wireless world? Both firms are making similar assumptions about how the platforms need to develop, even if their technical and business model approaches contrast. Both are pursuing common goals and where one leads, the other follows – Intel into low power SoCs, the stronghold of ARM’s customers; ARM into the server markets where Intel rules. Clever technologies are the result, and improved performance in well established applications – but the near-duopoly in processor cores, up and down the layers, leaves little air for firms with completely different takes on the challenges ahead. Radical new ideas will struggle to get any funding or to gain the ear of any powerful interests.
For now, ARM and Intel are setting the agenda, competing furiously but also squeezing out alternatives such as PowerPC and MIPS, as well as raising the barriers to any new entrants in their processor markets. The past week has seen a host of developments on both sides, though the TechCon audience was disappointed not to hear more details of ARM’s next generation processor, codenamed Artemis.
The centrepiece of the event was less exciting than that, but did take the Cortex-A microprocessor architecture further into the high performance, entry level smartphone – the only segment of the market where growth is still healthy.
The Cortex-A35 brings 64-bit capabilities to entry level devices and is the new low end member of the 64-bit range. ARM believes the affordable smartphone category will grow at 8% a year between now and 2020, and that its cores will be in one billion devices in this segment by then, out of a total of 2.8bn.
However, with smartphone growth slowing – with the effects felt in the quarterly results of many players from Qualcomm to Samsung – ARM is diversifying like mad, upwards into cloud and broadband infrastructure, outwards into the IoT and non-mobile devices. So it is pushing the power efficiency of the A35 as a good solution for other embedded products such as set-top boxes. The smallest configuration would consume less than 90mW of power when running at 1 GHz, and just 6mW at 100 MHz, said ARM. Up to four cores can be clustered together.
In a 28nm process an A35 should be around 25% higher in performance, 32% lower power and 25% more efficient than the midrange 64-bit core, the Cortex A53, the company claimed. Compared to the 32-bit A7, the A35 is rated to deliver 16% higher performance with 10% less power consumption.
The A35 has already been licensed by “several” unnamed vendors and should appear in commercial system-level products in about a year’s time.
The A35 will be strategically important in the Intel battle too. The smartphone SoC is, of course, the area where Intel has the least chance of ever catching up with ARM and its licensees. Last year, ARM won 85% of the sockets for mobile processors and Muller said it would have its 64-bit architecture in more than 50% of them next year. But the figure used to be over 90%, and Intel has made some bold moves to win at least a respectable slice of the next generation of mobile SoCs, particularly in the entry level space. It has particularly targeted China, investing in Rockchip and Spreadtrum alliances, which will help bring some scale to its first integrated processor/modem, SoFIA. The A35 is targeted very clearly at licensees which will need to up the ante with their own ARM-based SoCs, in order to limit the impact of Intel’s new friendships.
In servers of course, the positions are reversed and Intel’s dominance looks hard to crack. The ARM ecosystem is, in general, not going directly after the conventional server and other high performance infrastructure enabled by Intel Xeon. Instead, the vendors are trying to dance around that fortress, aiming to address emerging infrastructure categories with greater agility – and, crucially, better price and power consumption figures – than the incumbent. Coprocessors and accelerators to surround the servers and offload particularly demanding tasks; the blades for hyperscale cloud platforms; Cloud-RAN hardware; new base stations and network processors for the high capacity, agile, resource efficient, IP network.
Here, the two giants really will go head-to-head in chasing the new opportunities. Muller said ARM-based servers and infrastructure-oriented SoCs were now shipping from a range of companies, emphasizing how its licensing model enables its architecture to diversify rapidly and attack new markets nimbly on the back of a large number of chip suppliers with different skills and segment expertise. ARM now runs its own web sites on Hewlett-Packard servers running on its own processor cores. The firm said earlier this year that it believes it can secure 25% of the server processor market and 45% of the embedded processors used in communications gear by 2020.
But Intel is working hard to prevent this outcome, not just building new walls around its own kingdom, but expanding into the new surrounding segments too, with a particular enthusiasm for Cloud-RAN – its biggest ever opportunity to penetrate the closed world of the mobile access network with a controlling platform, rather than chipping at the edges with WiFi.
To rain on ARM’s parade, Intel announced a partnership with Latin American regional giant America Movil, helping it to transform and “cloudify” its network. The company joins the list of major operators, also including China Mobile and Verizon, with which Intel has loudly proclaimed the kind of strategic alliance it would not previously have had the assets to secure.
The world’s largest chipmaker also unveiled the Xeon Processor D-1500 product family and new Ethernet controllers as the latest in its growing portfolio of products which target, among other things, C-RAN and communications network systems. It said the D-1500 is focused on network and storage optimization and is particularly suited to helping operators extend intelligence from the core to the edge to improve performance, latency levels and resource efficiency.
It joins a Xeon D, launched earlier this year, which was optimized for microservers, the initial segment of the server market targeted by ARM-based processors, and the most vulnerable to their attack because of the requirement for very low power and cost, as firms look to tie together large numbers of boxes to scale up their cloud activities.
The new member of the Xeon D range is Intel’s third generation 64-bit SoC and the first based on Xeon processor technology. The family consists of eight CPUs with between four and eight cores. Lynn Comp, director of market development for Intel’s Network Platforms Group, said: “The goal is the to bring the capabilities of virtualization and high availability … further out into the edge of the network, so base stations and RANs.” She added that this was about “not just the control plane but starting to really extend into the data plane and towards the network edge to be able to support these new capabilities.”
Penetrating every link in the carrier network chain is also about having a full portfolio of surrounding products, since it will be many years before all the high performance functions of the RAN can realistically run on x86 servers. So Intel has been adding to its ranges of coprocessors, accelerators and controllers over recent years, and this week added a multihost and a single-host Ethernet controller to complement its processors. The Ethernet Multihost Controller FM10000 and Ethernet Controller X550 are designed to handle the dense east/west traffic between virtual machines, which will increase as clouds become more dense and as more and more network functions are run in software.
“Networks are facing extraordinary demands as more devices become connected and new digital services are offered,” said Sandra Rivera, general manager of the Network Platforms Group. “Building intelligence throughout the communications infrastructure and using a standards-based approach offers service providers the foundation to build agile, cloud-ready networks that allow them to expand their services capabilities.”
These announcements came a week after Intel significantly upgraded its IoT platform with a more flexible approach (see Wireless Watch November 5 2015), and the audience at ARM TechCon will have been very aware of that as Muller led the line-up of executives discussing the UK company’s push into this market. Its strength in the embedded microcontroller and real time processor markets may give ARM the entrenched advantage, but it knows Intel has some crown jewels of its own, notably the Wind River real time operating system and virtualization platform, and the security technology acquired with McAfee. And there is perhaps more opportunity here for other companies to spoil ARM’s party – Imagination’s MIPS is focusing heavily on markets such as wearables while some of the microcontroller experts are also turning to non-ARM cores.
The company’s announcements in this area were mainly about its software and services platform for IoT than actual cores. Muller announced a service that allows smartphones to authenticate and provision a consumer connected device securely. The aim is to “make IoT look like the mobile web not embedded control” to bring all kinds of app developers into the ecosystem.
ARM’s key asset in the IoT, like Intel’s, revolves around security, and the extension of its TrustZone architecture, initially conceived for the handset. This is coming into its own in the world of virtualized, multi-role devices and of mobile commerce, because it isolates an area of the chip where only trusted code can run, supporting sensitive tasks. The TrustZone area runs a very slim OS which provides little mileage for hackers and can be vetted constantly. It has been adopted by companies like Samsung (for Knox) and Netflix, and is included in all the Cortex-A processor cores.
Now ARM has expanded it into the IoT, and it will now come with Cortex-M microcontrollers along with the new ARMv8-M architecture. This brings hardware-based security to every part of the core, but with only slight impact on real time operations, said the firm, and as such, it extends the highly strategic TrustZone initiative into the darkest corners of the IoT.
TrustZone for ARMv8-M includes new instructions plus techniques that shift functions which would run in software on Cortex-A microprocessor cores into hardware to reduce software overhead where real time response is required. In 64-bit systems, a hypervisor handles the switching between secure processes in the TrustZone, and insecure processes outside it. In the new 32-bit version, no hypervisor is needed and the mechanisms run in hardware. Secured and unsecured code can run in parallel.
The technique relies partly on the latest interconnect specification in ARM’s AMBA family, AMBA 5 AHB5, in turn based on AMBA 3 AHB-Lite with support for more types of memory and transactions. This allows the TrustZone to securely with the rest of the chip package in an embedded device; and a part to accelerate data encryption, called CryptoCell, the result of its acquisition of Sansa Security earlier this year.
ARM, like Intel, knows that processor cores do not win markets on their own any more, and is developing entire platforms around the internet of things (IoT) and security. It has been moving to increase the functionality of the Cortex-M microcontrollers to support a middle way between more basic MCUs and full microprocessors. This middle ground will be suitable to a large number of emerging IoT device chip categories, and is also the target area for the mBed OS.
At TechCon, ARM added to that ambitious IoT operating system effort by releasing a technology preview version of the upgraded mbed OS together with additional reference designs. It claims there are now 55 partner companies in the initiative, and 150,000 developers.
The stripped-down OS for ARM Cortex-M microcontrollers is designed to add to the capabilities of those chips, bringing them closer to microprocessors while keeping the software overhead and power consumption very low. New features include management functions and the first native OS support for Thread, the Google/Nest-driven personal area network protocol which is also heavily backed by ARM for smart home applications, in particular.
The new reference platforms are for wearables, based on a wrist-based connected device with eight weeks of battery life; and infrastructure monitoring, using 6LoWPAN and CoAP mesh network protocols to manage large numbers of devices in a smart city or smart factory context.
Related new products include mbed Device Connector, a free service for companies to connect and manage their IoT endpoints securely. Available at connector.mbed.com, mbed Device Connector is free for developers to use on up to 100 devices, handling up to 10,000 events per hour.
Krisztian Flautner, general manager of ARM’s IoT business, said the platform was designed to accelerate the creation of new hardware so developers could focus on differentiation, while reducing cost and time to market, and access the tools “to scale deployments from a single cloud-connected sensor to a host of more intricate devices that enterprise-grade IoT solutions demand”.
“ARM is enabling developers to prototype complex end point interactions while simultaneously creating allied web services in the cloud,” said Paul Teich, principal analyst at Tirias Research, in the ARM release. “mbed Device Connector can be used for free with up to 100 devices, handling up to 10,000 events per hour, and this will allow developers to take products to a fairly advanced stage before they need to decide on their next step. This is going to have an effect on the rate of innovation as it reduces the risk for everyone.”
There is significant work ahead if ARM is to stay ahead of Intel and also convince its licensees, and new potential customers in the IoT and servers, that it is offering them the peak of innovation. Muller insists the rapid changing requirements of the market will make that essential. One of his examples, during his opening keynote, was to show off the equivalent of a six-inch wafer on printed plastic film. This is “not very powerful” compared to silicon, he admitted, but “opens up a whole new world of computing. It will always be 20 years behind silicon, but it does open up new applications — it’s not just about driving down to the smallest geometries you can do a lot with just 100,000 transistors,” he added.