The three market sectors for small cell chipsets

Powerful chipsets developed for use in Small Cells have evolved dramatically over recent years, as have those for macrocells. Clear market segmentation is evolving associates specific silicon vendors to the different system architectures. We look at how these critical components are shaping up for the future.

The smartphone handset industry highlights the high stakes involved

Silicon vendors make multi-hundred million dollar bets on how the wireless industry will transform itself in a 3-5 year time window, and stand to win or lose huge returns. The smartphone industry saw Qualcomm capture a majority market share, with their RF baseband chipset embedded in the vast majority of new smartphones. Brutal price competition translated this to 66% revenue share in a market sized at $22.1 Billion during 2014 according to Strategy Analytics. LTE now accounts for more than half of total baseband revenue.

Factors affecting that sector include mainstream smartphone vendors developing their own chipsets (Apple, Samsung), high stake courtroom judgements on patent rights and the rapidly growing Chinese market. The evolving market share of upcoming smartphone vendors also plays a major role, as discussed by Technews.

A polarised Small Cell market

The scope of small cells has expanded from the original residential femtocell through to higher capacity urban outdoor metrocells. Some vendors even include up to 20W models in their “small cell” range. The Small Cell Forum uses classifications of residential, enterprise, urban and rural which help segment the market.

The small cell chipset industry has also seen consolidation through acquisition, with some half dozen companies left setting out their stall in the market. Broadcom acquired Percello, Intel bought Mindspeed (and thus Picochip), Qualcomm captured DesignArts – these and others continue to invest heavily as the market evolves.

As Radisys recently observed, some small cell vendors choose to select multiple SoCs from different vendors. This can be for dual sourcing to ensure continuity of supply, price point or radio technologies supported (eg 3G TD-SCDMA).

In-building Small Cells

In the residential sector, cost is the priority. With some 10 million units shipped to date, there is volume to offset early investments, but price pressure is intense. The entire BOM should be well under $100, leading to innovative cost cutting measures and high levels of integration.

Picochip dominated the early residential femtocell designs including AT&T’s popular 3G Microcell, but Percello (since acquired by Broadcom) quickly matched their capability and won over the major players. These residential designs evolved from the original four concurrent call capacity to handle 8 and then 16 or more, making them suitable for busier enterprise applications.

Broadcom then introduced LTE support, allowing the same chipset to be remotely configured for 3G or LTE. We also saw larger scale Enterprise RAN systems such as Spidercloud adopt their chipset, using a local controller to manage up to 100 radio nodes for the largest indoor buildings.

Picochip was acquired by Mindspeed in early 2012, and lost market share. In late 2013, Intel acquired Mindspeed’s wireless assets, investing and stabilising that part of the business to ensure continuity of supply and support. Contela selected it for their LTE product and the AT&T 3G Microcell is still using it.

Meanwhile, Qualcomm has been investing and developing in this market for some years, recently winning several key accounts. Well known brands including ip.access, Alcatel-Lucent and Airspan have all demonstrated or are shipping commercial products using their chipset. Unlike others, Qualcomm provide a tightly integrated solution with their own RF front end, GPS receiver and timing. They’ve made a major play of their Ultra-SON features that improve the performance of groups of closely located small cells.

I think it’s fair to say that this class of chipset is more focussed on indoor buildings than outdoor, although that's not exclusively so. These chipsets compete primarily with DRS (Distributed Radio Systems) and DAS (Distributed Antenna Systems) for some of the larger venues.

Macro equivalency and compatibility

TI and Freescale are both major players in the traditional macrocell business. TI invested heavily in their Keystone architecture which delivers macro-equivalency throughout a range that spans the higher end of Enterprise to Urban Public Access. Their approach is to allow OEMs to reuse the same software across their entire product portfolio from macrocell to microcell to picocell.

An example is NSN’s Flexizone, where their LTE picocell is understood to be based on the same TI chipset family as their macrocell products.

Freescale (formerly Motorola Semiconductor Division) have also made substantial investments and have been particularly successful in TD-LTE in China. They are to merge with NXP (formerly Philips Semiconductors) during 2015 to create a $40 Billion “Semiconductor Powerhouse” with over 34,000 employees.

ZTE’s public selection of Freescale’s BSC9132 chip is another example of reusing the same chipset across the whole product family.

The approach can reduce product development time, but involves more than simply porting software from one platform to another. Traffic profiles and signalling levels differ in dense urban environments.

Freescale have also won business from “greenfield” designs, such as Airvana’s OneCell and.

OEM vendors also tout the benefits of years of field experience, which ensures compatibility with the widest variety of deployed handsets. There is less emphasis on product cost per se, more on lifetime total cost of ownership. That involves ensuring adequate processor performance to handle future software upgrades, highly automated configuration and SON features which may not be present in macrocells.

Neither of these companies is targeting the low cost residential market, where consumer electronics price points and volumes apply. They promote the desire for macro equivalency, highlighting accelerators and optimisations designed from experience gained from their existing macrocell field experience.

Centralised approach using Cloud RAN

A third approach relates to Cloud RAN and Network Virtualisation, where the intensive RF baseband processing is shifted from the edge of the network to be co-located at central datacentres. This would typically need dark fibre from each antenna, shifting the economics from one of equipment cost to that of backhaul.  

Other benefits of this centralised approach are that multiple radio heads can be tightly synchronised, using some of the most advanced LTE-A features to squeeze the highest performance from available spectrum. This may require the backhaul not just to be high capacity dark fibre, but finely tuned for phase timing synchronisation.  In practical terms, this may involve physically extending or shortening the fibre cables by many metres to align the delay.

Co-located processing scales best when using the highest performance building blocks, and this is where Cavium and Intel would bring their expertise. While there’s nothing to stop other baseband vendors playing in this space, it’s these two that seem to be making the strongest marketing in this space.

Last year, Cavium announced a monster ThunderX_SC chip with up to 48 separate 64-bit ARM cores. They are present in many large scale datacentre and network processing systems and have been looking to get into the RF processing side of things. At Mobile World Congress 2015, they demonstrated several alternative RAN architecture configurations, trading off different amounts of baseband processing located centrally and at the edge.

Summary

Categorisations such as this are never hard and fast. None of the above precludes any individual vendor from playing in more than one segment - indeed, several already do. But it does help position the needs and focus for different parts of the market.

Which silicon vendors are most focussed on each of these three distinct market segments:

• In terms of volume, the in-building sector is most promising in the short term and probably where we’ll see the most upset. Broadcom and Qualcomm are both well positioned.
• Urban densification, especially for LTE, will be a natural market for the mainstream RAN vendors to evolve into. Protecting their investments and minimising risk naturally encourages them to stick with tried and tested platforms. TI and Freescale (soon NXP) are likely to continue to dominate.
• Cloud RAN is a new approach that may suit specific markets and applications where dark fibre is easily and cheaply available. Cavium is the one to watch here, with Intel a dark horse.

Some publicly announced chipset selections

 Further reading

Our more extensive and perhaps slightly outdated roundup of silicon vendors from 2013.