Intel exerts welcome discipline over its IoT strategy, sacrificing Quark

Intel may be having a late flowering in high end mobile device modems, but this does not mean it is anywhere close to threatening Qualcomm yet. Having finally admitted defeat in the mobile processor/SoC space, it urgently needs to steal a march on its rival in the emerging IoT market, or at least some of its key categories. Until now, Intel has been showing indecision and adopting a scattergun approach to the IoT, with mixed results, but now it looks set to become more focused on its key strengths in higher end devices and those for the industrial IoT.

That focus is welcome. Intel has made a string of investments and acquisitions in wearables, but that market is not living up to its much-hyped promise. The firm recently axed its wearables division, which was built around its Basis acquisition of 2014. Intel had touted the wearables market as a key area in which it could expand its business, but this became far more difficult once it backed away from mobile processors, which could have been adapted for smartwatches, glasses and so on. Intel recalled all of the Basis Peak watches in 2015, after some customers complained about the wearables overheating and causing blisters. Intel then stopped supporting the watch.

And in recent months, the company has axed several other IoT-focused products which had been launched with great fanfare, mainly offerings based on its Quark core, an x86 design specifically targeted at IoT maker boards and ultra-low power end nodes, including the microcontroller space.

Earlier this year, it killed off several Quark products for the IoT maker market – Edison, Galileo and Joule – and those were followed this month by the Arduino maker board, a would-be rival to Raspberry Pi, and its associated Curie module. The maker and hobbyist spaces are important to innovation in the fragmented IoT landscape, but they are very alien environments for Intel. It now seems to be focusing on higher value links in the IoT chain, including gateways, edge computing and automotive systems (in the latter, it will come up against Qualcomm if that firm completes its acquisition of auto chip market leader NXP).

It is not just the maker market which was a bridge too far for Intel. It has ended the wider attempt to drive the x86 architecture into the microcontroller space – very simple, very low power chips which will run many IoT end nodes and small devices. Companies like NXP and STMicro are strong in microcontrollers, while the margins, market structure and mechanics are very far from Intel’s norms. It was always going to be at a disadvantage to ARM, which can span full processors and microcontrollers with the different flavors of its Cortex platform (Cortex-A and Cortex-M, plus the Cortex-R real time processor design).

By contrast, Intel was trying to squeeze an architecture designed for fully fledged processors into unsuitable terrain. The Quark range, based on a stripped-down implementation of x86, was supposed to help it target the crossover territory between microprocessors and microcontrollers, where large chip volumes are expected to materialize.

“Intel tried to take the x86 everywhere but the MCU business is not amenable to Intel’s business model of high margins and volumes and they had no second sources for the modules,” Kevin Krewell, senior analyst at Tirias Research, told EETimes.

A year ago, Intel was describing work on the D2000, a 32-bit Quark processor which consumed only 35 milliwatts in active mode. The firm said it planned to scale Quark from microcontrollers “to right below the Atom X1000 for Linux with lots of implementation options in cores and SoCs”. There were two other variations (the SE and D1000) on the original Quark, a 32nm design unveiled in 2013, promising one-fifth the size and one-tenth the power consumption of Intel’s previous lowest-power core, Atom.

Now, it seems that Atom will once again be the low end of the range, despite its failure to capture the smartphone market. Atom-based gateways will be central to the strategy, which revolves, according to an Intel statement, around “IoT market segments that access, analyze and share data. These include retail, industrial, automotive and video, which will drive billions of connected devices.”

Certainly these sectors were stars in the Intel’s recent second quarter results announcement. Strong growth was seen in its IoT segment, despite recent setbacks. Sales in the industrial, video and automotive sub-segments drove 26% year-on-year growth to deliver $720m in IoT revenue – in comparison to just 2% growth in the same quarter last year.

In the video segment, Intel has been notching up early wins recently with its Movidius machine vision processor. Intel paid an undisclosed fee for Movidius in 2016 and has since been supporting the development of its low power vision processor. The product is already being integrated in the drone market with DJI, but it is also targeting small devices such as phones, VR headsets and cameras – any IoT application that needs eyes for the physical world.

Much of Intel’s IoT business revenue looks to be coming from the industrial sector, in which it builds custom solutions to monitor the data from automated industrial processes. The IoT industrial solutions business is advertised on Intel’s website as using a complete range of the company’s processors, from the Quark, Atom, Intel Core and Xeon, as the solutions involve analyzing data across a whole production line.

In terms of its automotive business, Intel had been building in-vehicle-infotainment (IVI) systems for automakers – but the platform looked to be declining in popularity as more automaker move over to BlackBerry’s QNX, Android Auto, Apple Car Play and AGL. Intel has also been working on autonomous vehicles and has acquired auto machine vision specialist Mobileye for a reported $15.3bn. Intel decided to include the revenue generated by Mobileye for the quarter under IoT in its revenue figures, even though it is yet to complete the acquisition – which has helped boost the figures.

Mobileye is currently the number one supplier of chips and software that power video-based image recognition in cars – a huge market for growth, and one where Intel has paid a huge premium to entre relatively securely. The IP portfolio could bring in patent license fees from autonomous driving for decades to come.

Intel made a point of highlighting to investors, in its earnings presentation, that its non-PC business – including data center, memory and IoT – now accounts for over 40% of revenue, raising hopes that the worst of the transition away from the PC is over. However, there have been casualties along the way.

Intel has pursued a strategy of dipping its finger in as many IoT-related pies as possible, which inevitably leads to some failures. Since the maker products were part of its New Technology Group, rather than its IoT Group, the fate of the former now looks uncertain.

MediaTek focuses on IoT rather than high end smartphones to restart growth:

MediaTek may be Qualcomm’s biggest challenger in the smartphone SoC race, but it has seen its market share slipping all year, and is now more focused on non-handset products as it tries to restart growth.

The company said that its IoT products are seeing higher growth than its smartphone chips. MediaTek now gets more than half of its revenues from outside the smartphone space, especially in IoT, set-top box and automotive chips, although the flagship Helio mobile SoC still accounts for between 10% and 15% of sales.

The Taiwanese company repeated an earlier statement that its pace of product upgrades has slowed, leading to falling market share, which SVP David Ku, announcing second quarter results, said would not be reversed until the fourth quarter of this year.

MediaTek reported Q2 sales of NT$58.1bn ($1.92bn), down almost 20% year-on-year, while net income was the worst for more than five years. According to analysts at Credit Suisse, the company has lost share to Qualcomm and to Chinese firm Spreadtrum, especially to their dual-core offerings.

Ku said that R&D spending and headcount would increase only slightly this year to help control expenses, and most of the investment would be for 16nm and 12nm products. It also expects to have some 7nm offerings next year.

“Given the limited resources we have right now, we will probably double down on entry level 4G smartphones,” said Ku. “If 4G featurephones become a trend, we can always jump back in later on.” The central offering for the handset market in the second half of 2017 will be the Helio P, which targets superslim, midrange smartphones, but has modest specs compared to Qualcomm’s designs for this space – for instance, only a Category 6 LTE modem (due to be upgraded to Cat 7 later this year).

MediaTek has a high end design, Helio X, but on the modem side, the gap with Qualcomm is gaping. While the US firm launched the Snapdragon 835, with the world’s first commercial 10nm processor and a Cat 16 modem, earlier this year, MediaTek will not introduce Cat 10 or Cat 12 until next year.