Your browser is not supported. Please update it.

20 October 2020

Nokia spreads its net wider for FP4 and ReefShark 5G chips

Nokia has placed heavy strategic weight on its chip strategies and partnerships in the early years of the 5G era. It has had dramatic successes and failures in its development of its Reefshark silicon platform, which underpins a wide variety of infrastructure chips, and in its 5G choices.

On the failure side, a focus on FPGAs (field programmable gate arrays), in cooperation with Intel, for 5G base station chips was reportedly the reason for some early hiccups in 5G performance, and for product delays and high costs. Nokia pivoted towards a multivendor system-on-chip (SoC) strategy, working with Intel, Marvell and Broadcom, though not before the original problems had helped to force a change of CEO.

On the plus side, Nokia’s Reefshark chips have enabled it to achieve strong differentiation in high end routers, and it recently launched a chipset called Quillion, to target the same competitive edge in high speed fixed broadband networks.

In the router market, Nokia has built on the successful portfolio it acquired with Alcatel-Lucent, using its new silicon to move up the scale in terms of performance, claiming that its top end FP4 product was, three years ago, the world’s first ‘petabit router’. As such, it looks beyond Nokia’s traditional base of telcos and MNOs and also targets webscalers and cloud providers.

The Finnish company has just announced a contract to supply the FP4 routers to data center colocation and interconnection provider Equinix. The company will replace its legacy multivendor interconnection systems with a consolidated IP/MPLS network. It will use Nokia’s proprietary silicon in all its 210 colocation facilities worldwide, and Nokia is also supplying its Service Router Operating System, which will support Ethernet VPNs and segment routing.

These technologies are all part of Equinix’s move to support very low latency, high capacity 5G networks with a flexible, automated and software-defined architecture. And Nokia’s more traditional clients, MNOs, could leverage the shared Equinix infrastructure and backbone to reduce the cost of deploying their own cloud-based 5G networks.

“We see tremendous opportunity in providing our customers with 5G services, but this poses special demands for our network, from ultra-low latency to ultra-broadband performance, all with business- and mission-critical reliability,” said Muhammad Durrani, director of IP architecture at Equinix, in a statement. Nokia’s FP4 routers provide the “highly dynamic and programmable network fabric we need,” he added.

Nokia launched the FP4 routing platform in 2017, claiming the underlying processors were six times more powerful than any other network processors on the market at the time. In April 2019, it announced the UK’s BT as its first public telco customer for the fastest router based on FP4, the 7750 SR-14.

The FP4 processor is capable of 2.4Tbps of performance. It is a successor to the venerable FP3, which was itself a 400Gbps game-changer when it burst on the scene back in 2011. Several processors can be combined in one package to enable line cards capable of 12Tbps. These can be used in new routers – to achieve the sixfold traffic boost – or slotted into older ones of up to 10 years old. In that scenario, they could double the capacity of a 7750 SR-12 Service Router, or treble that of the newer 7750 SR-12e.

The new line-up is also designed to be more programmable than its predecessors. For instance, the FP4 can stream traffic statistics to an external analytics system, such as the Deepfield IP platform acquired by Nokia, in order to spot hacker attacks and other security issues, or gain better insights into traffic patterns overall. This is something the webscale players are good at, the vendor says, but service providers need to catch up, as their networks are “more opaque”.

Nokia designed its own memory and MAC address chips to drive performance still further, and uses a 2.5D packaging technology which was originally devised for the gaming industry. The decision to take ‘intelligent memory’ design inhouse was taken because very high throughput speeds are highly reliant on memory, and Nokia wanted to be able to control that key technology, and push it beyond what most merchant suppliers are doing.

Meanwhile, Nokia is still looking to enhance its new 5G SoC platform and make up for the missteps of late 2019, while continuing with its strategy of asserting control over crucial 5G components and attempting to differentiate performance from silicon upwards. It has announced a joint venture with Tampere University in Finland to develop inhouse 5G chips at a specific lab which will open next month. This will work on the “development of proprietary SoC chipsets, including their design and manufacture, to improve time to market and build a long term SoC development competence,” Nokia said.

The vendor will target annual chip releases from the new facility to address 5G use cases where a differentiated offering could drive significant new business, such as AI/ML, hardware security and open source SoCs.

Nokia has been working with researchers and engineers at the university for almost 20 years but is now turning the institution’s lab into a center of excellence to accelerate its own silicon efforts. The university is recruiting professors for the effort and according to Ari Kynaslahti, head of product management at Nokia, the collaboration “highlights our continued commitment to developing our  ReefShark chipset portfolio” – despite that commitment being questioned by some, during the fall-out from the FPGA issues.

It still has some catching-up to do on the 5G base station front, in terms of scale and credibility, though a string of recent 5G wins has helped to rebuild confidence. Performance issues with the FPGA-based solution it co-developed with Intel for its 5G baseband were blamed for teething problems with some customers, reportedly including Sprint, and for high costs, which ate into profits in late 2019. That proved to be a catalyst for a wider loss of market share and of investor confidence in Nokia at the turn of the year.

Late last year, it announced a U-turn in its chip strategy and said it would gradually migrate its 5G platforms to a new SoCdesign which would be far more cost-effective, power-efficient and reliable than the FPGA-based platform. That was chosen for its flexibility, since the programmability of FPGAs allows them to be adapted relatively easily when standards change, for instance – but they can be expensive and power-hungry. Nokia has only ever cited problems with an unnamed chip supplier for its performance and profitability issues with the 5G product, but that was assumed to be Intel.

Nokia is now working with multiple partners for various elements of its new-look ReefShark 5G baseband. Three ReefShark SoCs were developed, each from a different vendor and each focused on a different element of the vRAN architecture. It is working  with Broadcom on a custom ASIC-based solution, with Marvell for ARM-based multicore designs, and with Intel on some updated projects. It recently said that, like Ericsson, it is exploring graphical processing units (GPUs) as future accelerators for real time RAN functions.

However, the process of switching to the new, fully integrated SoC will be a multiyear task. In March, outgoing CEO Rajeev Suri said Nokia would provide quarterly updates on how many of its 5G shipments are powered by the new ReefShark SoC in order to show its progress in reducing 5G product costs.

“These new products made up approximately 10% of our 5G product shipments in the fourth quarter 2019, and we expect that percentage to increase progressively over the course of 2020, ending the year at more than 35%,” he said. “We expect to end 2021 at approximately 70%, and to essentially complete this transition in 2022.” In the first quarter, the ‘5G powered by ReefShark’ shipments had increased to 17%.

Despite the issues with the original chip, Intel reportedly still has a significant role. Nokia is using the Atom P5900, a 10-nanometer base station chip which Intel unveiled in February. The US giant also has a wide range of offerings which may play a part, from processors to FPGAs to custom ASICs (which are a bit more flexible than traditional single-purpose ASIC designs, though less so than FPGAs). Earlier this year, Intel announced a range of custom ASIC and SoC solutions that can support 5G NR basebands, and has said it is targeting 40% share of the market as it gets to scale.

But Marvell and now Broadcom are also in the mix. Broadcom will develop custom SoC processors for the RAN section of the ReefShark family. Like the Intel and Marvell SoCs, this will be developed using a mixture of Nokia and Broadcom IPR.

Nokia is using Marvell’s Octeon Fusion infrastructure processors, which add dedicated Layer 1 acceleration to support challenging real time and signal processing functions in the RAN, particularly in the distributed unit (DU) element of a disaggregated architecture. It is a companion for the Octeon TX2 chip, the latest  generation of the Octeon infrastructure processor range, announced in February. The TX2 claims to deliver 2.5 times better performance than the previous generation “to meet requirements for packet data processing, the incorporation of encryption algorithms for end-to-end security, and high end firewall crypto offoad,” as John Sakamoto, VP of the infrastructure processor business unit, put it.

The Octeon TX2 family combines up to 36 cores, based on the Arm v8-A architecture, with programmable hardware accelerator blocks, connected by Marvell’s coherent interconnect.

The Octeon Fusion wireless infrastructure processor can power baseband and radio units (RUs) supporting various functional splits between the CU, DU and RU in a disaggregated RAN. The product merges a Layer 1 processor and the Octeon TX2 and so is suited to Layer 1 processing in traditional integrated base stations, or the DU in a disaggregated 5G architecture.

Fusion adds programmable DSP cores and baseband accelerators to TX2. System designers can scale the number of cores to address different base station types, from the high end of small cells to large macrocells. The range includes processing solutions for smart radio heads, which require considerable compute power to support the complex beamforming associated with Massive MIMO.

Fusion is positioned as an alternative to vendors developing an ASIC inhouse, promising “the performance of an ASIC but still the flexibility of a processor”, as John Schimpf, senior director of product marketing, put it. He added that Octeon Fusion can support an upgrade from 4G to 5G in software. The two Octeon platforms sit alongside the high end ThunderX2 processor to form a triangle of chips to support RAN.

Both the TX2 and Fusion chips are available in custom varieties which allow equipment makers to integrate their own intellectual property (IP), as Nokia is doing.