• NB-IoT modem landscape is crowded and will see inevitable consolidation
• On-device AI landscape is highly fragmented and the real AI chip story is yet to come
• Qualcomm makes the case for 5G but can it keep the lead?

Mobile World Congress 2018 is over now and it’s time to carefully assess what has happened in terms of mobile component announcements and what it means to the industry. Typically, IP companies and chip companies are well ahead in the value chain in terms of investments in next generation technologies and studying these companies provides a great insight into the timing of new technologies, new features and opportunities for the industry. In this blog post, we briefly look at some of the key mobile component industry themes we observed at the show and what it means in the near-to-medium term.

Crowded Cellular IoT Chipset Landscape: Over the last couple of years we have seen a key trend of bifurcation of LTE modem roadmaps from baseband chip companies. Cellular baseband companies have re-profiled their product roadmaps to address both high performance broadband devices including smartphones and low-power devices such as IoT. Leading modem vendors including Qualcomm, MediaTek, Huawei and Intel and smaller companies such as Sequans, Altair, Nordic Semiconductor and GCT Semiconductor all vying for a share of the pie in the cellular IoT chip market. By our count more than 17 cellular IoT modem chip companies and more than 3  cellular IoT modem IP companies are competing for share. Arm’s cordio NB-IoT IP will be available in commercial modules in 1H 2019 and this could potentially open flood gates in the NB-IoT market as Arm is targeting low-cost MCU (microcontroller unit) vendors with this. In 2017, the majority of cellular IoT chip volumes were LTE Cat.1 and EC-GPRS.

Based on our discussions with IoT chip vendors at MWC, we believe Cat.M1 will pick up momentum during 2018 and NB-IoT activity will start to gather pace by the end of this year. Both Cat.M1 and NB-IoT have different use cases as Cat.M1 can support VoLTE whereas NB-IoT cannot. NB-IoT is primarily targeted at fixed applications such as smart meters. We think consolidation will be inevitable in the NB-IoT modem market as entry barrier in this market appears to be low at this point. Our discussions suggest some network operators are also exploring NB-IoT chip design of their own! However, NB-IoT modem companies are differentiating with high integration (baseband, PMIC, RF, SRAM memory, GPU, MCU, security, GNSS/GPS, RF Front-end and passives), low power consumption and small area chip-scale package (CSP) or system-in-package (SiP) solutions that are as small as 8.8 mm x 10.8 mm. While cellular IoT is at odds with other LPWAN technologies such as Sigfox and LoRa, Korea-based GCT semiconductor announced a hybrid IoT modem with support for NB-IoT and Sigfox.  

Highly Fragmented On-device Artificial Intelligence (AI) and Machine Learning (ML) Solutions: Thanks to its huge scale, the smartphone market provides an ideal platform for on-device AI and ML. However, the MWC 2018 show was a bit underwhelming in terms of on-device AI and ML announcements in smartphone chipsets, in our view. While Apple, Intel, HiSilicon and Qualcomm have some head start, we believe there is no clear leader in the on-device AI and ML chip space currently. We at Strategy Analytics believe that the smartphone chip industry is still waiting for ‘proper’ on-device AI chips and meanwhile the industry is experimenting with multiple architectures. Some companies are repurposing existing GPU and DSP cores to address AI and ML use cases while others are experimenting with new architectures from 3^rd party IP companies. Arm, the smartphone processor IP market leader, has kept the industry waiting long for AI and ML-specific IPs and this has opened a window of opportunity for multiple architectures to address AI and ML in smartphones. However, Arm’s project Trillium and Object Detection processor IPs will be available by later this year in commercial chips. At the moment, the majority of ML on smartphone activity is centred around image processing, 3D face scanning and voice.

At the show, we have seen chip vendors partnering with specialist software companies such as SenseTime, Face++ and ArcSoft to enable AI and ML use cases on their chipsets.  At MWC, STMicro showed Alexa on MCU (near field) and also showcased its AI SoC (codenamed Orlando V1), manufactured in 28 nm FD SOI process technology, with deep learning capabilities. This chip is likely to be used in automotive applications. We feel that 2018 will be an experimental year for AI and ML chip architectures and 2019 is likely to see more powerful on-device AI and ML capabilities in smartphones. However, it remains to be seen which AI architecture will win higher share in smartphones as several companies with neural IPs competing in this space.

Multi-mode 5G NR Chip is Tangible but is there Market for it? 5G was everywhere at the show and it was hard to miss it with almost every company playing the 5G card. The cellular baseband market leader Qualcomm has been touting its progress in 5G NR basebands for a year or so and before coming to the show the company announced significant progress with OEMs and operators with its X50 multi-mode 5G NR chip. We believe this chip is likely to be manufactured in 10 nm process technology. The first phase of 5G devices in 2019 are likely to see dual basebands (4G SoC plus 5G NR slim modem) but we expect within 12-18 months fully-integrated 5G SoCs. Intel and HiSilicon also detailed their 5G chip progress at the show. Samsung LSI is also said to have Exynos 5G chip in development while Spreadtrum partnered with Intel for its 5G chip. MediaTek is working with infrastructure vendors and China operators to bring out its first 5G chip in 2020. At the moment, Qualcomm clearly has a lead in 5G modems, thanks to its work in gigabit LTE area and its early work with standard bodies, infrastructure vendors, network operators and OEMs. We note that Qualcomm had a significant lead in 4G basebands too but the competition caught up within 2-3 years and we expect a similar thing to happen in 5G. Typically, a network technology transition acts as a forcing function for industry consolidation and we expect 4G to 5G transition is likely to result in more consolidation. Despite the chip readiness, the market for 5G will have to evolve across the ecosystem as the increased mmWave complexity brought by 5G has to be tackled.

Broadcom-Qualcomm: Almost no meeting we participated in at MWC went without mentioning Broadcom’s attempt to acquire Qualcomm and its potential implications. While some competitors sense a rare opportunity, most people believed that the merger would be a bad one for ‘engineers’ as Broadcom is likely to downsize some of Qualcomm’s initiatives to accelerate margins. Broadcom is after Qualcomm’s cellular chip business and we expect the company to divest almost all other Qualcomm’s assets if the deal closes. We have analysed Broadcom-Qualcomm merger in a recent blog post.

This is just a high-level overview of MWC announcements from a processor perspective. Strategy Analytics will be doing a deep dive MWC 2018 component webinar on Thursday 15^th March 2018 at 4pm UK / 12 noon NY / 9am LA, covering AI, processors (basebands and apps processors), RF front-end, display, camera sensors, battery and memory. You can register for the webinar here.