In our special Mobile World Congress edition of Wireless Watch, we identified changes in the mobile network semiconductor landscape as one of the key trends of 2019. The opportunity for chipmakers is the gradual shift from vendor-specific developments to merchant chips, as suppliers seek the economies of the open market. But there are two challenges – ensuring that the commoditization of network hardware, driven by virtualization, does not mean a devaluing of the underlying silicon; and the growing need, for larger chip providers at least, to deliver an end-to-end platform rather than stay in one or two comfort zones.
To add a further dose of uncertainty to the market this year, there may be further consolidation – following a major wave in 2018 – and there will also be rising influence from open source initiatives, which are starting to drill down from the network into the chips themselves.
The most important aspect trend in the chip industry supplying mobile networks will be polarization. The complex infrastructure supporting 5G networks and associated virtualization and telco cloud technologies will require a wide variety of high performance semiconductors, often to enable specialized tasks. Increasingly, cloud economics mean these chips will come from merchant vendors rather than being developed inhouse by the OEMs.
That means the large chip providers (and their underlying platforms, like ARM’s) will need to offer full portfolios, directly or via partners, which address every point in the long chain of silicon, from base station and radio head to core and transport, and from central data centre to cell site to edge. This will provide opportunities at the other end of the supplier scale, for very specialized start-ups which can address niches that are not economic for the giants, directly or though alliances with those larger firms.
So, at MWC, we reflected that, despite its failure to acquire Qualcomm, Broadcom is still expanding into many areas of wireless infrastructure, while seeking to build on its dominance of the merchant switch-chip, exploiting the relative weakness of Intel here, and fending off open source while adapting to the white box revolution in key accounts like AT&T.
Marvell, courtesy of its Cavium purchase, now has a strong end-to-end offering from edge to cloud and small cell to macro site, with processors that are well optimized for demanding workloads like 5G virtualized RAN. ARM has been filling in gaps at the edge to offer its licensees options for every part of the telco cloud and network, as it looks to increase its infrastructure business. And Intel, which has been acquiring companies like Altera to surround its processors with specialist chips such as FPGAs, is also fleshing out its range, adding a 5G base station platform and no doubt casting around for its next acquisition.
In the short time since MWC finished, there have been further developments. On the hyperscale side, Nvidia acquired Ethernet interconnect firm Mellanox to boost its position in high performance computing (HPC). Two new open groups emerged, one hosted by the Linux Foundation (the CHIPS Alliance), the other led by Intel, to support its Compute Express Link (CXL) open interconnect technology. ARM extended the edge capabilities of its v8.1 architecture, while Samsung and Intel both added new base station chips.
And all the time, the semiconductor companies which profited from selling specialized products for proprietary base stations and appliances are now having to adapt to a world where a switch becomes a white box, a packet core runs in containers on COTS hardware, and even the RAN will evolve to be a set of virtual network functions on servers that may remain more rarefied than just Xeon boxes, but will increasingly become open, standardized platforms.