Much of the analysis of the mobile chip industry in the past year has centered on devices, particularly whether Intel could make Atom a major force at last, or if anyone could unseat Qualcomm. However, equally important changes have been happening on the wireless infrastructure front, where Intel and Qualcomm are both eager newcomers, and Texas Instruments and Freescale are the incumbents in the base station. The final weeks of the year are seeing M&A activity in this space, with Intel buying the small cell assets of Mindspeed (formerly Picochip) and Avago snapping up LSI for its storage and wireless infrastructure offerings. These deals highlight important trends for 2014, particularly the expansion of generic silicon platforms into many previously separate engines of the IP data boom, from servers to base stations. And that super-trend will certainly drive further acquisitions next year ‘ some are even whispering of an Intel bid for Broadcom.
For now, the emergence of new architectures is opening doors to new players. Small cells allow platforms originally developed for home equipment and devices, with their associated economics and ecosystem, to move into the carrier network, while Cloud-RAN and SDN (software defined network) provide opportunities for server silicon vendors to leap the divide from data center to network.
Situations as fluid as this encourage M&A by defending incumbents and new entrants. Most of the original small cell pioneers, who created platforms for low cost, low power femtocells ‘ architectures now being adapted for public access metrocells running on SoC (system-on-chip) designs ‘ have now been snapped up. Broadcom bought Percello, Qualcomm acquired fellow Israeli firm DesignArt, and Mindspeed took over Picochip. The last of these has now been sold again, this time to Intel, where it will play a far more strategic role in the giant’s ambitious bid to push x86 into almost every link in the network, from C-RAN servers to network processors to base stations (the rest of Mindspeed has been acquired by MACom for $272m, probably signalling the final end of a once-interesting development in small LTE base stations).
An interesting speculation is how far Qualcomm will reach into the infrastructure sector it quit in 1999 (apart from some CDMA products) when it settled its patents suit with Ericsson and sold its network chip business to the Swedish giant. It certainly won’t be selling whole base stations again, as it did in the 1990s, but it may have to remember some of the lessons of that episode. Qualcomm was a failure in CDMA networks because it had base station platforms but nothing for the switches, the area that made Ericsson and others into powerhouses in the early digital days. Now it is entering infrastructure again right at the edge, with chips and SoCs for small cells, Wi-Fi equipment and home gateways (courtesy of Atheros). However, while there will be a premium in carrier small cells for a couple of years, the economics of the metrocell demand that it evolves quickly towards consumer-class costs and margins, in order to drive the huge quantities that will create the true HetNet. Qualcomm understands consumer product margins and may choose to stay in familiar territory, adapting its device platforms for small-sized infrastructure and harnessing the advantages of coming from that angle (low power, highly integrated multimode and so on).
However, this will set it apart from its main rivals and it will have to compete against companies with a top-to-bottom story for the HetNet ‘ chips, tools and code which span base stations of all sizes; network processors; and increasingly, the hardware to support the virtualized network. This is the problem the femtocell specialists would have had if they’d remained independent ‘ their architectures are reaching up to more complex, carrier-grade equipment, bringing their disruptive economics with them, but TI and Freescale could always produce their ‘one platform for the whole HetNet’ argument to keep OEMs close. A truly open ecosystem for small cells could still change that story, enabling carriers to mix and match metrocells from different vendors, and not necessarily from the same vendors supplying other equipment. If that materializes, as many operators hope, it will favor new, consumer-class suppliers such as ODMs, and they will not have the same pressure to adopt an established wireless architecture. But as with anything related to openness in wireless, there is a big gap between dreams of a seamless multivendor network, and carrier confidence actually to deploy one.
While Qualcomm is, for now at least, staying at the edge, Intel is unlikely to use its new Picochip jewel to design a small cell SoC itself, or at least, not in isolation. The expertise and PHY layer engineering it will acquire could feed into several aspects of its newly unveiled mobile infrastructure strategy, which sees the giant aiming to push x86 everywhere for the HetNet and, in particular, the software defined network. Here, Intel can build on its existing strengths ‘ dominance of the server market; broad SDN roadmap including a switch platform; healthy offerings in some types of network processor; key acquisitions such as Fulcrum in switch-chips and now Picochip.
Of course, the firm has faltered many times before, buying and then discarding some useful assets (it sold its Dialogic DSP unit to Eicon in 2006, its ATCA board business to Radisys in 2007 and, of course, its ARM processor division, XScale, to Marvell a year before that). It often sells a unit and then starts reinventing the wheel again, and it battles with its lack of expertise or track record in actual radio technology, despite device-side efforts such as Centrino and the Infineon Wireless acquisition. However, the recent unveiling of Highland Forest by the head of Intel’s infrastructure division, Rose Schooler, spoke of a new determination and confidence not to waste the first real upper hand Intel has ever had in the mobile world ‘ the rise of virtualized networks and Cloud-RAN, where x86 has the edge.
‘This isn’t a strategy that started when SDN started to get attention,’ Schooler said on the conference call to launch Highland Forest. ‘We’ve been on this journey for a decade. We’ve been looking at the evolution for the Intel architecture franchise within communications and redefined networking to boil it down into all of the infrastructure and workloads that support a network function.’
This ambition is currently underpinned by various platforms which pair the Xeon processor with a widening variety of companion chipsets – Crystal Forest for the data plane, Jasper Forest for the edge router, and so on. Highland Forest is Intel’s third generation communications platform and the most aggressive in harnessing Intel’s power in enterprise data centers to shift the economics of networking. It is based around the Xeon E5-2600 v2 processor with the new Coleto Creek chipset, which is optimized for high end telco infrastructure, boasting the ability to process up to 255m packets per second.
This is part of a broader strategy to expand the platform, targeting four key areas of network infrastructure workload ‘ application processing, control plane processing, data plane packet transport, and signal processing. Schooler says communications infrastructure is a $16bn business, when telcos, enterprises and cloud servers are considered, and Intel has only 5% of that. It aims to increase the percentage dramatically with a processing and chipset platform to take networking equipment beyond ‘fixed function devices with customized silicon, towards more of a general purpose approach’. That pitches Intel not just against Broadcom but the habit of major vendors like Cisco to use their own customized chips.
The ex-Picochip assets could feed into various aspects of this plan, notably the next promised step, to produce the long awaited base station platform, built around a Xeon accelerator optimized for signal processing. Prototypes based on FPGA chips are being used in China Mobile’s huge Cloud-RAN market trial which, as Price put it, aims to ‘create synergy from big cells serving thousands of users to small cells in metro areas of public networks’. Intel has talked about producing a full base station offering, though to date it has mainly worked with partners to get x86 processors into cell site equipment, as seen in its alliances with NSN over the RACS/Liquid Apps offering, and Cisco/Ubiquisys for the Smart Cell.
Now it has certainly bought itself significant new expertise in this area. “The team and technology Intel is acquiring will make important contributions to how Intel Architecture (IA)-based solutions are transforming wireless access within mobile network infrastructure,” wrote Schooler in a blog post. “Our goal has always been to consolidate all four workloads to run on IA and we have already made significant steps towards enabling the last one – signal processing on Intel-based servers – through collaboration with China Mobile and SKT on designing Cloud RAN technologies.’
Schooler also said, in an interview with EETimes, that the deal would help Intel support baseband processing, including turbo N Code and Z Code functions. “It has been a decade-long journey [to support all comms workloads] and this acquisition is one of the final building blocks,’ she said. ‘One of the key gaps we wanted to close was in the area of wireless access, and that’s what this acquisition allows us to do.’
Another significant aspect of the deal is that Mindspeed/Picochip was the only firm to produce small cell chips supporting China’s 3G standard, TD-SCDMA, which gives Intel another route into the world’s highest growth wireless market, where it is already forging close ties via its C-RAN trials.
Intel was an early investor in Picochip via its venture capital arm and was linked with a possible acquisition bid two years ago, before the UK firm was bought by Mindspeed for $51.8m in 2012. Its second sale, for an undisclosed sum, should complete in February.
In pushing a general purpose platform for all kinds of network functions, Intel will be taking on specialized network processors, whether from vendors like Broadcom, or custom, but also the entire ARM ecosystem. It will have to battle ARM on every front as the processor IP firm, with its powerful customers, brings its low power advantage into base stations, switches and cloud servers and seeks to make similar ‘one platform suits all’ arguments to those of Intel. Increasingly, indeed, it will work with the same providers of the specialized functions to surround the processor (Intel and ARM work with common partners in many parts of this market, from CEVA for DSP cores to Asocs for C-RAN, and many others). However, ARM is more embedded in the base station/signal processing field thanks to long-standing alliances with Texas Instruments and the femtocell makers, while Intel has the incumbent position in servers (though, of course, ARM-based platforms are starting to penetrate, especially for cloud servers).
The arrival of 64-bit ARM platforms and significant interconnect technology highlighted ARM’s determination ‘ backed by key partners like LSI, itself acquired this week, by Avago ‘ to expand its offering to be truly infrastructure-class. In that battle, it is likely that PowerPC, the incumbent platform in this area, will be squeezed out over time as the two more generic offerings become increasingly capable, although at the high end, the tougher battle for x86 and ARM will be against dedicated custom architectures, like those used by Cisco and others in heavy duty switches.
LSI has been a very significant ARM partner in the networks world, being codeveloper and lead customer for the CoreLink technology and in the Linaro initiative. It will be interesting to see whether its new parent, Avago of Singapore, has equal enthusiasm for this sector. The $6.6bn takeover is being portrayed as being mainly focused on storage chips, but it highlights how the boundaries are blurring between many infrastructure silicon segments ‘ storage, servers, networks and so on ‘ and suppliers are consolidating, partly for scale and partly to be able to push common architectures across as many sectors as possible. So the acquisition not only creates economies of scale amidst slowing semiconductor growth, but also takes Avago into new areas. The two firms have little product overlap – LSI has a significant business in storage, as well as networking processors, while Avago does about half its sales in handset and base station RF chips (particularly power amplifiers and resonators/duplexers) and most of the rest in a mixture of optical components. The end result will be a company with “greater scale for investment and a broader product portfolio in a larger, more diversified organization,” said LSI’s CEO Abhi Talwalkar, on the analyst call.
Avago will reduce its exposure to the volatile wireless segments, which will become 25%, rather than 50%, of its sales, as well as becoming an overnight leader in enterprise storage as well as wireless infrastructure, positioning itself for the growing opportunities from the IP data boom. That boom needs to be targeted by a platform capable of serving diverse data handling products and even before the proposed merger, LSI had been working on an architecture which could be applied to nearly any element of the network.
Its Axxia 5500 family, introduced early this year, sports 16 ARM Cortex A15 cores, integrated with LSI’s networking accelerators and ARM’s low latency CoreLink interconnect technology, on a single 28n chip. The resulting communication processors aim to accelerate performance and reduce power consumption throughout an LTE or multiradio networks, including base stations, cell site routers, gateways and backhaul kit.
Networking acceleration functions support packet processing up to 50Gbps, 20Gbps security processing and 160Gbps of Ethernet switching via 16 10Gb Ethernet interfaces. Such capabilities are increasingly needed at the cell site as well as the back end in data-heavy networks, and LSI is paying more attention to bringing its capabilities to the base station, and supporting cells from small to large. LSI’s new chip offers the same key advantage claimed by TI and Freescale for their platforms ‘ a modular but common architecture spanning macrocell to picocell.
‘Mobile broadband growth is outpacing infrastructure capability, and service providers need solutions that will meet traffic performance needs while also minimizing cost and power consumption,’ said Jim Anderson, general manager of LSI’s networking solutions group, at the launch. The firm added: ‘The combination of LSI’s networking expertise, specialized acceleration engines and Virtual Pipeline technology with ARM’s power efficient processors and interconnect IP delivers communication processors that are uniquely suited for building intelligent, heterogeneous networks.’
LSI has worked closely with ARM over the past year and is a key partner in the UK firm’s bid to conquer the networking high end. This relies on a combination of new developments, all outlined last fall ‘ the move to 64-bit processors with ARMv8, which will appear in commercial chips this year; the big.Little architecture, which balances performance and energy efficiency using flexible combinations of high end and low power processors; and the CoreLink IP fabric. Together, these make a robust story for the high end space, and one in which LSI has played a critical role courtesy of the CoreLink development.
This technology is important because it allows many combinations of cores and other system elements to be integrated in a scalable way, up to massively parallel arrays. For instance, ARM recently showed off CoreLink with a memory controller targeted at networking and storage equipment. Last October it unveiled two intellectual property blocks, theCoreLink CCN-504 interconnect (in effect an on-die network) and the CoreLink DMC-520 dynamic memory controller. They work together, with the former providing the ‘glue’ to connect cores and cache components, within a chip and also to other system elements. The CCN-504 can scale up from single-core CPUs to four clusters of four cores each, meaning the common architecture can power infrastructure from small cells to macrocells to communications processors.
This is the kind of challenge which Intel needs to meet quickly and effectively, since so many of the broad goals are the same. Intel and the key ARM backers are not pushing a different philosophy in this market, just a different architecture. Both want to broaden their platform into as many products as possible which support the explosion of data traffic, as the industry moves towards more generic hardware to carry out all kinds of processing functions. That will inevitably lead to further M&A, since the winners will need skills, expertise and dedicated surrounding chips in a widening variety of function areas.
Good news for innovative start-ups, but some are whispering that Intel needs another really big acquisition to ensure its credibility in the networking area. One bold move would be a bid for Broadcom itself, which is powerful in several of the areas where Intel most wants to grow (the digital home as well as the network). That could be the superdeal of 2014 in semiconductors, and would certainly change the lines of competition significantly.
The LSI/Avago deal:
Avago will pay $1bn in cash and use a $4.6bn bank loan to fund the deal, while private equity firm Silver Lake Partners, which helped acquire Avago before its IPO in 2009, will provide a $1bn investment. The deal creates a business with about $5bn a year in combined revenue and $200m a year in potential cost savings. Avago was originally a division of Hewlett-Packard, and was part of the Agilent company which HP spun off in 2000. A group of private equity companies acquired it for $2.66bn in 2005 and it went public in 2009.