Which could be unlucky for some…

If history is anything to go by, it’s going to be unlucky for some, but it’s an indication of the momentum which is growing around cellular IoT that so many chip companies have jumped on the bandwagon. 

It’s not cheap to develop a cellular chip, even one that is moderately simple, such as is the case with the NB-IoT standard. Back in 2012, when I wrote about the cost of developing wireless standards, I put it at around $6 million for each chip and protocol stack. That was looking at Bluetooth and Wi-Fi. With the additional complexity of any cellular standard, along with network interoperability testing, it becomes far more expensive, as you need to test with as many operators as possible. Hence the development cost to get a chip and stack to market is probably at least $15 million. 

With thirteen different companies bringing chips to market, that’s an investment of around $200 million. Some of these have tried to cut their development time by acquiring start-ups which were already some way down the route. Sony purchased Altair, Huawei bought Neul, ARM bought NextG-Com and Mistbase, Goodix acquired CommSolid, while Nordic Semiconductor picked up around 60 engineers in Finland’s Oulu. We don’t know how much they spent on these acquisitions, but it’s probably well over $200 million. Add to that the costs of the standardisation process, infrastructure development and initial market trials and it’s clear that somewhere between $500 million and $1 billion has already been spent on getting NB-IoT to the point where it is today. That’s a level of investment that should be worrying competing standards like LoRa and Sigfox, as the NB-IoT companies will do all they can to recoup their investment.

Having said which, if you count up real NB-IoT deployments, it’s still early days. There are probably fewer than 10 million chips deployed. That’s the figure from Huawei, who is certainly leading the field. How many of those are actually connected and sending data back is questionable – the last year has largely been an exercise in getting things to work and spinning the PR. Nevertheless, Huawei is predicting that by the end of 2018 the number of chip shipments will reach 150 million, which, given the focus on NB-IoT within China, may well happen. The big question is what will happen in the rest of the world. To understand that, it’s interesting to look at the different companies which will be producing silicon.

The thirteen companies I’m aware of (please let me know if you know of any others) are HiSilicon (part of Huawei), Sanechips (a division of ZTE), RDA, Mediatek, Altair (owned by Sony), Sequans, Nordic Semiconductor, Goodix, Riot Micro, Qualcomm and Nesslab, along with ARM and ASTRI/CEVA. ARM and the ASTRI / CEVA partnership are IP vendors, but appear to be at a state where they are already behind some of the offerings, so are worth including, as if anyone plans to ship in volume, they’re an obvious destination. ARM is further differentiating itself by offering a wider-ranging IoT service including device management and aspects of provisioning. I need to apologise for missing GCT, which brings it up to fourteen. And since writing this I’ve been made aware of a further three - Pinecone Electronics (who have Xiaomi as an investor and appear to be building on ASTRI’s IP), Extra Dimensions Technology – a Beijing startup and Eigencomm – a Shanghai startup. That further highlights the China centric concentration and reflects the amount of Government support being put in to make China the leader in IoT. So we have a sweet seventeen, with probably more to come.

It’s slightly reminiscent of the early days of Bluetooth development, which saw a similar number of companies vying to become the leading silicon supplier. That included established chip vendors TI, Conexant, Broadcom, Motorola, Agere and Nat Semi along with a tranche of startups – Silicon Wave, Cambridge Silicon Radio, Brightcom, Zeevo, Transilica, Philsar, Zucotto, Parthus and Mobilian. Few are remembered, as most of the startups were acquired or went bust and very few of the majors went on to gain a significant market share. By the time the market had shaken out and volumes had grown to a billion chips a year, it was dominated by Cambridge Silicon Radio and Broadcom. I expect to see the same evolution within NB-IoT.

With both Bluetooth and NB-IoT, Intel were a prominent player at the beginning, but then appeared to give up before they shipped anything. That’s been a recurrent story with their wireless developments from HomeRF onwards. If anyone’s seen a real Intel XMM 7115 chip or any successor silicon, let me know. Otherwise I’ll continue to put Intel in the number 18 (dead parrot) position.

There are some significant differences between that early stage of Bluetooth excitement and NB-IoT. At the start of the Bluetooth journey, few believed that 2.4GHz chips could work using CMOS. Most of the major companies and startups were working on more exotic processes, such as SOI and GaAs. CSR managed to make CMOS work and prove them wrong. They also enraged the industry by predicting a price point of $5, when everyone thought that impossible, which makes the $5 PR pronouncements of the NB-IoT vendors sound a little déjà vu. Everyone followed CSR’s innovation and as a result a Bluetooth chip is now around $1 in volume. It means there’s some interesting legacy running through the HiSilicon chipset, as the Neul team which Huawei acquired had the same technical founders that started CSR – they know how to do low cost silicon. 

It’s not an exact comparison. Bluetooth requires interoperability between devices made by different companies, hence a phone manufacturer wants their phone to work with any headset or carkit, regardless of who makes it or what chip is within it. That requires extensive testing to ensure that every chip and host protocol stack works with every other one. With NB-IoT, chips only need to talk to base-stations, which are made by a more limited number of companies, many of whom also make the NB-IoT chips. However, different network operators may choose to use different frequency bands and enable different features. Moreover, the premise of much of the IoT is that devices will continue to work for ten years or more, so there are additional forward and backward compatibility testing issues. Whereas a consumer might replace a phone if it stops working, it’s another issue to reinstall millions of sensors within a smart city. They will be expected to work unattended for ten years or more. That needs a much higher level of confidence in the connection.

This brings us to the question of what will drive the IoT. Despite Ericsson’s 2010 prediction of over 50 billion connected devices by 2020, we’re probably not much above 100 million IoT devices. Whilst everyone believes that cheaper hardware and data rates will lead to a rise in deployments, we’ve not seen that so far. Whether that has been due to GPRS being turned off, or having a limited future, uncertainty about when we’ll have a global IoT network or just a lack of applications is open to debate. The reality is that for most of the IoT we’re seeing the same organic growth that characterised M2M, not a bright new world of connecting trillions of everything.

That uncertainly about what the killer application might be is evident in the different approaches to NB-IoT chips from the seventeen manufacturers.

Leading the charge is Huawei, via its HiSilicon chip division. They were a major contributor to the NB-IoT standard after their acquisition of Neul and brought the first chips to market. They’re also leading base-station deployments, promising full national coverage of China by the end of 2018. Huawei’s view of the market seems to be a policy-driven one of putting NB-IoT chips into as many things as possible, connecting them to the cloud to collect data and then thinking about what to do with the data at some point in the future.

There’s nothing wrong with that – it’s an approach that has worked very well for companies like John Deere, who put sensors on their farm machinery fifteen years ago, collected the data and are now a leader in smart agriculture applications. But the high cost of their tractors meant they could absorb that cost. Where you’re dealing with low cost sensors or meters, that’s more challenging. 

I illustrated the problem in an article on LoRa, Sigfox and LTE-M back in 2015. Today’s M2M applications are generally proprietary and expensive to deploy, hence there are few of them around and not much data generated. To get lots of data, which is what the IoT needs, you need to reduce the cost of sensors so that they’re deployed in volume. However, until they’ve been out and operating for a length of time, which may be a year or more, you don’t accumulate enough data to start extracting insight and generating value.

M2M grows organically. It needs a disruptive decrease in cost of deployment to move to the IoT model. However, bringing chip costs and data costs down by themselves isn’t enough, otherwise we’d have seen LoRa and Sigfox deploy far more devices than they have. For companies to bother with implementing IoT connections they generally need external forces to persuade them to add connectivity and do something with the data. To move from tens of thousands to tens or hundreds of millions of deployments normally requires Government mandates or regulatory requirements. 

Going back to our seventeen pioneers, Sanechips (the chip arm of ZTE) appears to be following the same approach as HiSilicon of “pile it high and sell it cheap”, as does Mediatek. Nesslab and Goodix are relative newcomers and are still some way away from shipping. It’s not clear exactly what they’re going to do to differentiate themselves. If you move to RDA you see a slightly different attitude. RDA has been successful with low cost GPRS chips and seems to be drawing on that M2M experience to add some extra functionality into their RDA8909 chip. As well as NB-IOT, they’ve added an 802.11 receiver for indoor positioning along with Bluetooth 4.2, both presumably to help with commissioning.  GCT’s GDM7243i is another interesting chip along the same lines, as it combines NB-IoT with Sigfox and Bluetooth 4.2, claiming concurrent NB-IoT and Bluetooth operation. The Sigfox / NB-IoT option gives manufacturers the flexibility to support different networks, although if the NB-IoT bandwagon takes off (and the presence of seventeen chip vendors suggests it will), that might not be quite the differentiator that GCT had hoped. Having said which, it supports both eSIM and soft SIM, which implies that GCT have thought about the provisioning issues.  I have no details on Pinecone Electronics, Extra Dimensions Technology or Eigencomm.  They’re ones to watch out for. If they’re bright, they’ll be looking at the current silicon offerings and seeking out niches, as it will be tough. The figures I’m hearing from Chinese module vendors is that they’re expecting an NB-IoT chip to be at $3 by the end of this year, leading to $5 modules, and silicon falling to $2 by the end of 2019.

Riot Micro – a Canadian startup, came out of stealth mode when they announced their RM1000 chip at the end of last year. They recently provided some further information, saying that they are working with PoLTE Corporation to integrate LTE positioning. That’s one of the advanced features coming into NB-IoT, which aims to use the cellular network for positioning, removing the need for GPS. It suggests that Riot Micro is positioning itself in the asset tracking sector. Incidentally I do like the contrast in company names of Sanechips and Riot. It makes you wonder whether the latter should rename themselves Insanechips?

Altair was one of the first companies to embrace NB-IoT, pairing it, as most do nowadays with LTE-M. Since their acquisition by Sony, they’ve incorporated low power GPS into their most recent ALT1250. It’s a nice package which looks attractive for automotive and asset tracking.

The other early starter is Sequans. Their Monarch series of chips is already incorporated within a number of modules. At MWC they announced an NB-IoT only chip, pushing the price point down and presumably betting on NB-IoT becoming the volume opportunity.

Then we have Qualcomm – the gorilla in today’s baseband market. They’re working with module partners to make it easy for developers to connect and gain access to web services via their Gizwits IoT platform. That’s an interesting acknowledgement of the fact that the NB-IoT chip is only a comparatively small part of the overall IoT value chain. It’s the same strategy that ARM is pushing.

The reality of cellular is that few companies will put chips directly into their products. That only becomes cost effective if you’re shipping in tens of millions. Instead, they will opt for pre-certified modules. All the chip vendors mentioned above are relying on module partners to get them to market. In most cases they’re also leaving it up the module vendors to provide differentiation. The chip vendors are keen on the mantra that they’re designing chips which will enable $5 modules, although module vendors invariably counter that for the foreseeable future the chip will be at least $5 and the module double that.

This makes Nordic Semiconductor appear as the odd one out. They’ve made their name with Bluetooth low energy chips, where they’ve generated a loyal and supportive developer community, which has helped them become a leading supplier of BLE chips. Their nRF91 module is their first foray into cellular. And it’s a module. Although they’ve designed their own chip, they’ve made the decision not to sell the chip, but to supply pre-certified modules, which also include GPS. They believe that their developer community is a key differentiator and that it will enable faster time to market.

Nordic’s decision reflects the fact that there is a difference in purchasers of NB-IoT and traditional baseband chips. The latter market is dominated by Qualcomm, with Mediatek popular in the Far East. These have a business model of selling tens or hundreds of millions of chips to a few phone manufacturers, who they support to death. The IoT is a different business. Whilst there are very large markets, such as metering, agriculture and smart cities, these are verticals which have traditionally been served by hundreds of different companies, often operating at a national level. Many have limited comms or cloud knowledge and much longer market cycles. They typically use a conventional distribution model, where they get support from the distributor’s Field Application Engineers, as the overhead of supporting them directly in unsustainable. In general, these companies have not embraced the IoT, as large-scale data insight is not part of their current business model. That’s the quandary for everyone trying to kickstart the IoT market.

So which silicon companies will win in NB-IoT? Nordic have publicly stated that their module will be around $9 in high volume, so there is a decent expectation that silicon will be hitting $3 - $4 by 2020, with modules between $6 and $10. Those are decent price points, and competition from other LPWAN entrants has already driven data contracts down to a few dollars per year. Which means neither hardware or data costs should be a major barrier.

Analysts don’t help much. According to Markets and Markets, sales of NB-IoT chips will rise to $181 million by 2020, which if we plug those numbers in, suggests rather sluggish growth to fewer than 50 million chips – way behind Huawei’s short-term predictions. It’s probably considerably less, as the Markets and Markets brochure implies that their figure includes modules as well. IHS Markit are more aggressive, suggesting worldwide NB-IoT nodes could reach 450 million by 2021. As that’s cumulative, it could be fairly realistic, particularly given Huawei’s scaling out plans. 

I’m more interested in where these chips and modules are likely to go. As I mentioned above, breaking the hundred million and then the billion number probably needs Government mandates or policy decision. Markets like smart metering have been cited as vital to this, but Western Governments show little inclination to pick emerging technologies such as NB-IoT, rather than trying to breathe life into geriatric GPRS and proprietary solutions. What I’m seeing is Chinese chips going into China. The headline applications are the smart city ones of water meters, electricity meters and smart parking. Whether they will be used for anything more insightful than billing is unknown, but it’s a route to commoditisation, along with the development of better provisioning and cloud services. At which point we will start to see NB-IoT connectivity beginning to appear in all sorts of devices, as it becomes an effective way of registering and monitoring how a product is actually used.

That leaves the rest of the world with a question of how to address the IoT? I’ve no doubt that there will be lots of low volume, pseudo-IoT devices – the kind of connected product that crowdfunding projects are so fond of. But they’re unlikely to reach volume. Asset tracking may become significant, although the US / European indecision over whether to deploy NB-IoT or LTE-M will not help. What Western companies need to do is to concentrate on data insight.

The implication is that unless something goes wrong, the silicon market will gravitate towards Chinese chip companies, simply because they will be closest to their customer base. To compete, other chip and module suppliers will need to move up the value chain – something which I see Altair, ARM and Qualcomm doing. Seventeen contenders are more than enough to get us to commoditisation, and more than will survive in this market. Who wins the silicon race may not matter. What will matter is who owns the higher layer IoT value stack, as that is where the real revenue resides. If China leads in deployments and hence data capture, we may find that it ends up owning the IoT.

Read more on NB-IoT at www.nickhunn.com