Your browser is not supported. Please update it.

17 April 2020

ARM joins O-RAN, not letting Intel have open networks all its own way

ARM has joined the O-RAN Alliance, in an important, if belated, step forward for its own telecoms infrastructure ambitions, and for the goal of a truly multivendor mobile network platform from silicon to software.

The O-RAN Alliance – originally spearheaded by AT&T and now under the custodianship of the Linux Foundation – is the most influential cross-industry initiative focused on specifications for an open, disaggregated, multivendor RAN. Together with Telecom Infra Project (TIP), a close partner, the Alliance aims to define common interfaces and platforms that will allow equipment and software from any vendor to interoperate. That, in turn, could lower barriers for smaller suppliers to enter the closely guarded, proprietary fortress of the big five RAN vendors, creating a more open, WiFi-like ecosystem with high levels of price competition and innovation.

Many small suppliers of physical equipment and of virtual network functions (VNFs) have joined O-RAN, as have some of the big names. Nokia and Samsung are actively contributing, no doubt hoping to assert leadership of the new open architectures while the smaller challengers are unproven at scale; Ericsson is a less active member.

But while interoperable access points, multihaul products, antennas and VNFs are appearing to support an open network, nearly all the work at silicon level has been based on Intel processors. The company has invested heavily in supporting trials and its FlexRAN reference platform has played a significant role in development and tests of both open and traditional 5G systems. But this is ironic, if the quest for an open platform ends up with Intel in the same default position it has enjoyed in PCs and, until very recently, servers. Of course, Intel provides a common processor architecture which can be adopted by any equipment maker, unlike the proprietary chips that power current base stations – but x86 is still virtually a single-vendor platform.

Hence the importance of ARM’s entry into O-RAN. So far, projects to develop RAN designs on non-Intel platforms have been interesting but low-impact – for instance, the collaboration between Lime Microsystems and SiFive on a RISC-V base station.

ARM has far greater potential to influence the ecosystem in the near term. Its network infrastructure designs may be less prominent than its device and IoT processor cores, but they are well-established, with several generations of evolution behind them. According to Panch Chandrasekaran, marketing director of infrastructure at ARM, the Softbank-owned company’s processor designs account for about 28% of the overall infrastructure market today and have a growing presence in base stations.

This market share has been hard won. Early attempts to challenge Intel with ARM-based servers, mainly by start-ups by (now defunct) Smoothstone, had limited success, and even Qualcomm retreated from that effort. The base station sector remained largely closed (outside of small cells) to any merchant chips, Intel included.

But that is changing as operators move slowly towards virtualized RANs, in which some or all baseband processing is carried out by VNFs running on standard servers. These will not be commoditized, since they will have to support highly intensive and specialized processes, as the other articles in this Special Report describe. But they will run on general purpose processors (GPP), often coupled with specialist accelerators for demanding workloads.

ARM will be looking, like Intel, to use positions in more mainstream servers to push out into the peculiarly demanding RAN segment. Nokia’s recent problems with implementing an inhouse system-on-chip (SoC) for its 5G base stations highlighted the challenges of developing these new virtualized, disaggregated RAN architectures – it subsequently abandoned that approach after poor performance and has introduced a combination of Intel and Marvell/ARM designs to its AirScale virtualized base stations.

Marvell is the key standard bearer for and ARM-based 5G platform. ARM has gained a handful of sizeable server chip customers since those early, difficult days, but Marvell is by far the most successful vendor of ARM-based server processors and has now pushed into base stations, backhaul and other segments, to offer an end-to-end 5G solution.

Marvell uses ARM cores for many components of this platform, and is a member of the O-RAN Alliance, with commercial offerings supporting some O-RAN specs. Earlier this year, it announced its 5G base station offering, which consists of its Octeon TX2 infrastructure processor family, based on up to 36 ARM v8-A cores, coupled with the Octeon Fusion wireless infrastructure processor for base stations. That launch was greatly helped by Nokia’s announcement of its agreement with Marvell at the same time.

Such products will help to push ARM into wireless infrastructure and O-RAN, but it is important that the core IP provider itself both shows public support for the open initiative, and gets the inside track on contributing to specifications and ensuring its designs are optimized for them.

The company is already a member of Telecom Infra Project (TIP), which has a complementary role to that of O-RAN, focusing more on identifying service provider requirements and developing reference designs and practical deployment roadmaps. A year ago, for instance, ARM was one of 10 companies which set up a new TIP project group, called Open RAN 5G NR, led by Sprint and Vodafone (Intel and Qualcomm are also involved).

This group was focused initially on small cells, but ARM is keen to counter the perception that its designs’ famous power efficiency is their only advantage, and that they run out of steam in very high performance applications like macro base stations.

Here, ARM’s Neoverse architecture is central to its push into 5G networks. The company said: “As 5G evolves and expands, our infrastructure business is developing, and in many instances inventing, the technologies needed for a 5G world that includes a trillion edge devices. At the end of the day, we see 5G as a key market segment for ARM.”

Neoverse was unveiled in October 2018 as a brand for server processors, separate from the general Cortex brand for all CPU cores. While ARM’s initial server-oriented processor cores were fairly generic, those in the Neoverse family are far more specifically geared to cloud, edge and 5G workloads, and the architecture has found favor with Amazon AWS for its Graviton platform.

Neoverse was originally on the 16-nanometer Cosmos design, but now supports 7nm (codenamed Ares, with commercial branding of Neoverse N1 and E1) with plans for 7nm+ (Zeus), and then 5nm (Poseidon). Neoverse N1 is designed to provide a common platform from hyperscale to edge cloud, and points in between such as 5G base stations or Internet gateways. For a data center server SoC, ARM expects vendors to use 64 to 128 cores running beyond 150W. For storage, security, and large edge nodes, systems would typically run eight to 32 cores at 25W to 65W.

To meet the hefty demands of a cloud platform, Neoverse N1 has virtualization technologies, performance and management technologies, reliability and service support integrated.

Meanwhile, the Neoverse E1 is optimized for throughput performance – 2.7 times better than in its predecessors, according to ARM at launch, with 2.4 times better throughput efficiency and over twice the compute performance. This is aimed particularly at telco networks and at smoothing the transition from 4G to high throughput, fully virtualized 5G infrastructure. It will target equipment from 16-core SoCs running at 15W for gateways or 35W for 5G base stations; going up to 32-core versions which could run the data plane for routers with multiple 100Gbps Ethernet ports.

In a blog post to announce the O-RAN sign-up, Chandrasekaran wrote that Neoverse now delivers over 40% performance/power advantage over competing alternatives, which helps network operators cut operating expenses over the lifetime of their equipment.  But, he added, a broad ecosystem is also vital, and participation in industry initiatives.

He wrote: “It takes more than best-in-class processors to broadly deploy 5G. ARM works closely with a broad ecosystem of developers and makes significant contributions in more than 100 open-source initiatives including OS, CI/CD, virtualization and cloud-native technologies, end-to-end security and networking software stacks. ARM is already a key player in telco-centric organizations such as Telecom Infra Project, DPDK, fd.io, and Linux Foundation’s Akraino project.”