A lot of investment has gone into development of specialist silicon for small cells in recent years. In this article, we've taken a step back and reviewed the landscape, talked to many of the major players, and developed our view of what's going on. It might be controversial for some, but should help navigate through the plethora of press releases and announcements.
Before we start
It's no small task to design and develop a proven, efficient and cost effective silicon solution. We're talking tens of millions of dollars here, sometimes more like $100 millions over the lifetime of a product family. Teams of hundreds of designers can be involved across multiple aspects. Proof points publicly confirming this include VC funding in excess of $100million for Picochip and the $86 million that Broadcom paid for Percello. Revenues can also be substantial - TI disclosed annual embedded processing revenues of $1.9 billion in 2012, of which a significant contribution is believed to come from mobile basestation components.
A complete reference design has multiple components. The baseband processing chipset attracts the greatest attention, probably because it's also the most costly. So-called SoC's (Systems on a Chip) may not actually include all of the components required, with different designs requiring any or all of separate RF front end chipset, GPS, Ethernet, memory and timing/sync elements.
Software is also an important part of the solution. While many Layer 2/3 stacks are multi-vendor, the PHY (Physical Layer or Layer 1) needs to be tightly integrated into the specific baseband processing in order to take full advantage of any silicon "accelerator" hardware. Timing/sync software may be a third party solution such as that from Symmetricom or developed internally. Layer 2/3 stacks and application software are available from several mainstream vendors and a few smaller niche players, and can be more independent of the hardware.
The core capabilities of a small cell chipset are baseband processing and signalling, which require DSP and traditional scalar CPUs respectively. Reusing an existing commercial silicon design is quite feasible – for example the DSPs can be adapted from CEVA or similar, while the scalar processors might be sourced from ARM or MIPS. Specific hardware accelerator circuitry may be designed for specific, highly demanding tasks. High speed busses which enable efficient and effective throughput are essential to make the most of the huge processing capacity.
Millions of 3G WCMA residential Femtocells have shipped today, with AT&T the largest with some 1 million in use. The majority are still based on Picochip's PC2xx and PC3xx series chipsets. They dominated the early designs for Femtocells and continue to reap the benefits. Mindspeed, who acquired Picochip in 2012, have stated that they'll continue to manufacture and support this product family as long as there is demand for it. However, we have not seen any major new releases or features on that platform.
Instead, Mindspeed are investing in their Transcede product, and have migrated the Picochip 3G PHY across to that product family. The Transcede also supports LTE and has been demonstrated with multi-mode 3G/LTE operation. It seems likely that customers will be encouraged to migrate to that platform in the medium term, with the original Picochip withdrawn in the longer term.
3G small cells vendors using the Mindspeed design include ip.access, Ubiquisys and Alcatel-Lucent although the latter two also use/support Broadcom. Taiwanese ODMs such as Tecom are understood to use the Broadcom BCM61630 as part of a Ubiquisys reference design.
Broadcom have continued to invest in their Percello acquisition, evolving their 3G solution in two ways. Their latest chipset family now supports LTE and will soon integrate the RF front end, GPS receiver and Wi-Fi driver onboard into the same chip package. This will continue to drive down total femtocell costs because of reduced chip count.
Infonetics estimates today a residential 3G market share split of about 2/3rds Mindspeed and 1/3rd Broadcom, but I could see Broadcom's share growing across a number of categories. Further evidence for that includes Huawei's recently announced selection of Broadcom's BCM61670 chip for their own 3G small cell ePico products.
The only other major player promoting a residential femtocell 3G WCMDA chipset at this stage is Qualcomm with their FSM solution. Perhaps they brought their product to market a little too late for the early residential femtocell designs and to date have not announced any 3G WCMDA design wins. They acquired DesignArts in 2012 to contribute to their solution. I would not rule out their gaining a market share in the future.
Some of the larger 3G enterprise small cells were designed with more powerful chipsets from the outset. TI had several design wins, making it clear that they weren't aiming at the low end residential market and instead targeted the larger enterprise and Metrocell opportunities.
Despite there being a significant volume of 3G CDMA femtocells sold - Sprint alone have installed over 1 million - few chipset vendors have invested to design a dedicated CDMA small cell on a chip. The main vendors of 3G CDMA femtocells - Samsung, Airvana and Airwalk - have had to rely on more expensive FPGA designs which inherently make the unit cost higher than comparable 3G UMTS products. Qualcomm offer a version of their Femtocell Station Modem specifically for CDMA which has been incorporated into Ubee/Airwalk product, and this could quickly capture market share once the cost/performance is validated.
Picochip developed a reference design for this Chinese variant of 3G from an early stage, opening an R&D office in Beijing. Almost all 3G TD-SCDMA femtocells make use of this, and Picochip (now Mindspeed) appear to have little competition in this field.
(Click below to read about 4G/LTE Small Cell Chipset vendor landscape)