Two mobile network semiconductor announcements, made over the past couple of weeks, highlighted the contrast between the ‘new and old’ ways of approaching radio network design. Ericsson opened a design center in Austin, Texas, focused on ASICs (application specific integrated circuits) for 5G and representing a significant investment in very expensive chips. Meanwhile, Sigfox said it was working on an “ultra-cheap disposable radio” which could be priced around 20 cents.
Of course, these two radio efforts are targeting very different use cases and network types, but the Sigfox plan feeds into an increasingly lively debate. Are the high costs of optimized cellular radio networks really justified by their complexity and performance demands, or has the industry just been able to get away with charging so much because of the high barriers to entering the 3GPP ecosystem? And linked to that, is there the potential, in the cellular world, to achieve ultra-low cost, disposable systems that still deliver the required performance?
In the latter camp, there are various positive indicators. Many come from the WiFi, or other unlicensed radio, segments, where a more open ecosystem has always been in place. High levels of competition on a standards-based foundation have driven prices down – not resulting in an entirely commoditized platform, but a wide range of equipment for different purposes and business cases, from very cheap consumer gear to city or enterprise networks that are almost as expensive as cellular.
In the 3GPP world, there has never been this wide range of prices, but various trends are pointing towards a new expectation in terms of pricing. This has started in the devices, where the shift of handset growth towards lower-cost markets, and the emergence of IoT gadgets, have both made the chip prices of the past unsustainable except in premium models.
Now the same pressures are spreading to the network. MNOs will not be able to justify a 5G network built with the same cost levels as 4G. Virtualization and the software-driven network, implemented on commoditized white box servers and switches, are the big news. But what of the radios and antennas themselves? These remain physical items which rely on technology that has tended to be regarded as a dark art because of its complexity and general shortage of skills.
Sigfox says its ‘Admiral Ivory’ radio, for its low power WAN network, will be used in tracking applications such as monitoring packages. CMO Laetitia Jay said this would cost 20 cents, by contrast with the current Sigfox module at $3, and LTE-M alternatives currently around $7.50. Jay says unlicensed spectrum radios cost less than those for cellular network not because they are more complex or higher quality, but because there is a more competitive ecosystem – nine chipmakers now develop Sigfox-based modules or transceivers, including Texas Instruments and STMicro.
Another initiative that points to openness and commoditization even down to the radio level is that of Lime Microsystems, whose LimeSDR platform ticks many boxes for those hoping that the established supply chain of mobile networks will be broken apart in 5G – software-defined, open source, white box, even crowdfunded.
In July, UK-based RF transceiver maker Lime secured an alliance with Vodafone to develop solutions based on LimeSDR, which follows on the heels of a similar deal with the UK’s leading MNO, BT/EE.
Founded in 2005, Lime has toiled away in RF transceivers – hardly the rock stars of the mobile value chain – innovating in programmable devices which make it easier to support many frequencies. Then it burst into the limelight last year when it launched, and crowd-funded, LimeSDR. This has brought software-defined radio (SDR) to an open source base station platform, complete with app store courtesy of a collaboration with Linux major Canonical.
Lime has run two crowdfunding campaigns. The first supported the LimeSDR board itself and achieved almost double its $500,000 target in June 2016. The second, LimeNET, surpassed its $50,000 target in July but it still open. LimeNet combines the SDR with an Intel x86 processor to create a small cell base station, a gateway or an enterprise router, though the aim is to take it into macrocells too.
Lime claims it is the first company to take SDR into base stations, aiming to slash the cost of building a RAN and offering “the potential to completely transform the way telco networks run, shifting the emphasis and value away from proprietary hardware to open hardware with app stores on top”.
The Lime partnership is part of Vodafone’s Open RAN project to explore virtualized technology. The operator will use LimeNET to bring together a group of developers to work on innovations across many areas of the RAN, including LTE, NB-IoT, mobile enterprise and future networks.
Francisco Martin, head of radio product for Vodafone Group Technology, said: “Lime Micro is at the forefront of SDR wireless technology development, and the platform being app-enabled brings the concepts of agile and feature-rich systems together, unlocking new applications that leverage this radio flexibility and openness to build new services and a completely different radio.”
Lime CEO Ebrahim Bushehri said: “Radio access technology for wide area networks accounts for a significant portion of the overall deployment cost. We’re seeking to make the hardware an open source commodity sold for a fraction of current offerings, with the real value being in the software it runs. Doing it this way would effectively turn LTE, GSM or LoRa, or even 5G, into just an app.”
The first LimeNET product, the LimeNET Mini, is an app-enabled, small cell base station design targeted at localized IoT equipment, while the more-powerful LimeNET is for carrier-class wide area networks. EE is using it to support a ‘network in a box’ to bring cellular connectivity cost-effectively to remote areas or emergency situations. EE is partnering with several UK universities to develop these solutions.
Other applications developed by the LimeSDR community include IoT gateways, aviation transponders, smart meters, and systems for media streaming, radio astronomy, radar, drones and others.
At the heart of the platform is Lime’s LMS7002M RF transceiver, which allows the board to support a continuous frequency range from 100 Hz to 3.8 GHz, meaning it can be programmed to work with most mainstream wireless technologies including UMTS, GSM, LTE and WiFi, plus sub-1 GHz connections like Sigfox and LoRa, and personal area networks like Bluetooth, ZigBee and Z-Wave. The chip supports 2×2 MIMO. Much of the design of the board is open source, with the board schematics and layout available under a CC BY 3.0 licence, as well as the USB link and host software, and Altera’s Project Quartus software.
Under the Canonical partnership, developers are able to write and publish applications inside the Linux organization’s Snappy app store, where users can download them and then run them on their LimeSDR kits. Of course, developers are able to charge for these applications, with Canonical receiving a cut of that sale, but it seems likely that an ecosystem of free and open source apps will spring up around the LimeSDR. The LimeSDR is supported by Lime’s Myriad-RF community, which it set up as a family of open source hardware and software projects for wireless innovation.
Initiatives like Facebook TIP and OpenCellular, Vodafone Open RAN and Orange Telecom Track will send shivers down the spine of incumbent vendors, signalling MNOs’ determination to adopt a radically different cost structure for 5G, and to bring white boxes and open source into the heart of the network.
But the major mobile network vendors continue to invest in differentiating themselves through highly advanced radio technology, either directly or in partnership with chip suppliers. And some functions, they argue, will always require the optimized, single-purpose performance of an ASIC, despite the expense of these dedicated processors.
ASICs are designed specifically for mobile infrastructure’s demanding computation needs and are 100 times faster, more cost-efficient and less power-hungry than general processors such as an Intel CPU, according to Ericsson.
Ericsson’s new Texas design center will tap into the major ASIC and chip design community in Austin and, according to Sinisa Krajnovic, head of the networks development unit at Ericsson, it will “accelerate our development and our journey to 5G”. The company aims to have 30 designers and architects by the end of the year and 80 by mid-2018, augmenting its existing ASIC design centers in Sweden.
However, there are serious debates around the future of network ASICs. The more operators look to deploy most of their RAN network functions as VNFs on server hardware, the more there will be pressure to support the mobile network’s challenging requirements on more standardized processors. Running the entire RAN on commoditized CPUs and servers is, for most MNOs, a pipe dream (even excluding the obvious physical elements which remain at the cell sites, and will still need high performance semiconductors).
But Intel, Qualcomm, Cavium and other chip providers have recognized this, and are integrating FPGAs (field programmable gate arrays) and even GPUs (graphics processing units) with their general purpose server processors, so they can offload or accelerate the more challenging RAN functions. Though these additional chips can be programmed to address a specific task, they are less customized than an ASIC and can be reconfigured for new tasks as the network adapts.
“It is important to add custom-made silicon for different IT tasks,” said the VP of Intel’s data center unit, Shannon Poulin, in the wake of the company’s acquisition of FPGA market leader Altera. “We have to add programmable capabilities into our silicon for users to customize it to their workload needs and lead to software defined infrastructure.”
This shift has been seen in the transport network as well as the RAN – for instance, custom ASICs have declined for high performance switches, open doors for products like Broadcom’s Ethernet switch-chips. Such moves add a whole new dimension to standard platforms and have the potential to squeeze out custom manufacturers (except when it comes to the very strategic inhouse cloud developments of Google, Facebook and Microsoft).
It will be a slow death of the ASIC in the mobile network, and even if that day comes, it will be a mixed blessing for large OEMs. It will enable Ericsson and others to reduce the cost of semiconductor development considerably, but it will also rob them of one more way to differentiate themselves in their core network business, and to steal away the high margins of their traditional platforms, forcing them into a world of commoditization and disposability.