Dell’Oro Group estimates the RAN market is currently generating between $40 billion and $45 billion in annual revenues. The market research firm forecasts that Open RAN will account for 15% of sales in 2026. Research & Markets is more optimistic. They say the Open RAN Market will hit $32 billion in revenues by 2030 with a growth rate of 42% for the forecast period between 2022 and 2030.
As the undisputed leader of microprocessors for compute servers, it’s no surprise that most of the new Open RAN and virtual RAN (vRAN) deployments use Intel Xeon processors and FlexRAN™ software stack inside the baseband processing modules. FlexRAN™ is a vRAN reference architecture for virtualized cloud-enabled radio access networks.
The hardware for FlexRAN™ includes: Intel® Xeon® CPUs 3rd generation Intel® Xeon® Scalable processor (formerly code named Ice Lake scalable processor), Intel® Forward Error Correction Device (Intel® FEC Device), Mount Bryce (FEC accelerator), Network Interface Cards – Intel® Ethernet Controller E810 (code name Columbiaville). Intel says there are now over 100 FlexRAN™ licensees worldwide as per these charts:
A short video on the FlexRAN™ reference architecture is here.
FlexRAN™ got a big boost this week from AT&T. In a February 24, 2022 blog post titled “Cloudifying 5G with an Elastic RAN,” Gordon Mansfield, AT&T VP Mobility Access & Architecture said that “AT&T and Intel had co-developed an industry-leading advanced RAN pooling technology freeing 5G radios from the limitations of dedicated base stations, while enabling more efficient, resilient, and green 5G networks. DU-pooling will eventually be usable by the entire 5G operator community to drive the telecom industry’s goals of green and efficient wireless networks forward.”
DU pooling technology was made possible by combining AT&T’s deep knowledge of Open RAN technologies as one of the co-founders of the O-RAN Alliance with Intel’s expertise in general purpose processors and software-based RAN through its FlexRAN™ software stack running on Intel 3rd generation Intel® Xeon® Scalable processors. The open standards for communications between radios and DUs that were published by O-RAN enabled its development, and the result is a technology demonstrator implemented on FlexRAN™ software.
Intel is now facing new Open RAN competition from several semiconductor companies.
Marvell has just unveiled a new accelerator card that will slot into a Dell compute server (which uses x86 processors). Based on a system called “inline” acceleration, it is designed to do baseband PHY layer processing and do it more efficiently than x86 processors. A Marvell representative claims it will boost open RAN performance and support a move “away from Intel.” Heavy Reading’s Simon Stanley (see below) was impressed. “This is a significant investment by Dell in open RAN and vRAN and a great boost for Marvell and the inline approach,” he said.
Qualcomm, which licenses RISC processors designed by UK-based ARM, has teamed up with Hewlett Packard Enterprise (HPE) on the X100 5G RAN accelerator card. Like Marvel’s offering, it also uses inline acceleration and works – by “offloading server CPUs [central processing units] from compute-intensive 5G baseband processing.”
There is also EdgeQ which is sampling a “Base Station on a Chip” which is targeted at Open RAN and private 5G markets. Three years in the making, EdgeQ has been collaborating with market-leading wireless infrastructure customers to architect a highly optimized 5G baseband, networking, compute and AI inference system-on-a-chip. By coupling a highly integrated silicon with a production-ready 5G PHY software, EdgeQ uniquely enables a frictionless operating model where customers can deploy all key functionalities and critical algorithms of the radio access network such as beamforming, channel estimation, massive MIMO and interference cancellation out of the box.
For customers looking to engineer value-adds into their 5G RAN designs, the EdgeQ PHY layer is completely programmable and extensible. Customers can leverage an extendable nFAPI interface to add their custom extensions for 5G services to target the broad variety of 5G applications spanning Industry 4.0 to campus networks and fixed wireless to telco-grade macro cells. As an industry first, the EdgeQ 5G platform holistically addresses the pain point of deploying 5G PHY and MAC software layers, but with an open framework that enables a rich ecosystem of L2/L3 software partners.
The anticipated product launches will be welcomed by network operators backing Open RAN. Several of them have held off making investments in the technology, partly out of concern about energy efficiency and performance in busy urban areas. Scott Petty, Vodafone’s chief digital officer, has complained that Open RAN vendors will not look competitive equipped with only x86 processors. “Now they need to deliver, but it will require some dedicated silicon. It won’t be Intel chips,” he told Light Reading in late 2021.
Inline vs Lookaside Acceleration:
While Marvell and Qualcomm are promoting the “inline” acceleration concept, Intel is using an alternative form of acceleration called “lookaside,” which continues to rely heavily on the x86 processor, offloading some but not all PHY layer functions. This week, Intel announced its own product refresh based on Sapphire Rapids, the codename for its next-generation server processors.
Simon Stanley, an analyst at large for Heavy Reading (owned by Informa), said there are two key innovations. The first involves making signal-processing tweaks to the Sapphire Rapids core to speed up the performance of FlexRAN™, Intel’s baseband software stack. Speaking on a video call with reporters, Dan Rodriguez, the general manager of Intel’s network platforms group, claimed a two-fold capacity gain from the changes. “In the virtual RAN and open RAN world, the control, packet and signal processing are all done on Xeon and that is what FlexRAN enables,” he said.
The other innovation is the promise of integrated acceleration in future Sapphire Rapids processors. Sachin Katti, who works as chief technology officer for Intel’s network and edge group, said this would combine the benefits of inline acceleration with the flexibility of x86. That is preferable, he insisted, to any solution “that shoves an entire PHY layer into an inflexible hardware accelerator,” a clear knock at inline rivals such as Marvell and Qualcomm. Despite Katti’s reference to inline acceleration, Stanley does not think it is Intel’s focus. “None of this rules out an inline acceleration solution, but it does not seem to be part of the core approach,” he told Light Reading. “The key strategy is to add maximum value to Xeon Scalable processors and enable external acceleration where needed to achieve performance goals.”
Both inline and lookaside involve trade-offs. Inline’s backers have promised PHY layer software alternatives, but Intel has a major head start with FlexRAN™, which it began developing in 2010. That means lookaside may be a lot more straightforward. “The processor is in control of everything that goes on,” said Stanley during a previous conversation with Light Reading. “It is essentially the same software and makes life very easy.”
Larger network operators seemed more enthusiastic about inline during a Heavy Reading survey last year. By cutting out the processor, it would reduce latency, a measure of the delay that occurs when signals are sent over the network. That could also weaken Intel, reducing power needs and allowing companies to use less costly CPUs. “If you use inline, you probably need a less powerful processor and less expensive server platform, which is not necessarily something Intel wants to promote,” Stanley said last year.
System on a Chip (SoC) start-up EdgeQ (Cupertino, CA) announced its launch from stealth with $51 million in total funding, including $38.5 million in a Series A round. Backed by investors Threshold Ventures (formerly DFJ), Fusion Fund, Yahoo! co-founder Jerry Yang (AME Cloud Ventures) and an unannounced strategic customer, EdgeQ will address the 5G infrastructure market with products aimed at delivering 5G connectivity with AI computing.
The company counts experience in cellular modem development from Qualcomm, Intel and Broadcom on its team. It sees a limited number of players in the market, focused especially on smartphones, leaving room for new providers targeting edge devices and infrastructure.
EdgeQ said it will deliver a converged 5G and AI silicon platform that is open and software programmable for both devices and edge infrastructure. By introducing open programmability to the baseband, EdgeQ wants to provides a new software-driven development model for OEMs and operators, supporting existing cellular protocols such as 4G and 5G as well as the next generation of networks.
EdgeQ’s AI-5G SoC is aimed at emerging 5G private networks that are viewed as the backbone of industrial Internet of Things and other data-driven enterprise deployments. Along with manufacturing, the AI chip maker said Tuesday (Nov. 17) it is targeting the automotive, construction, energy and telecommunications sectors.
“We are rapidly evolving from a smartphone economy to a constellation of intelligent edge devices,” said Vinay Ravuri, CEO and founder of EdgeQ. “This will cause massive disruption to the current paradigm, where existing fixed-function approaches are inadequate to meet the scale, speed, and breadth of new end connections.”
“We provide an open platform converging 5G plus AI, which abstracts much of the complexities for our customers working on 5G deployment—from supporting multiple chipsets, different software stacks, board design, cost, power, and latencies in transferring data in between, not to mention, potential security hazards. Though we are not ready to disclose the hardware details, our 5G chip architecture uniquely lends itself to AI in a way without needing an extra AI accelerator hardware, saving both power and cost to the end customer.”
The combination of 5G connectivity, AI hardware and a “software-friendly” design is intended to enable an “open and programmable platform that is adaptable, configurable and economical for 5G-based applications,” added Ravuri, a former Qualcomm vice president for product management.
Ravuri said Qualcomm’s 5G SoC design (targeted at 5G endpoints) was closed while EdgeQ’s was open. “Their chip technology does not support 5G connectivity and AI computing, making it inadequate for enterprise-grade 5G infrastructure, which needs robust computing capabilities in addition to 5G,” he said. “We can bring the best of breed here—the cellular, but also offer to the market what they’re really looking for, which is an open ecosystem where they are able to innovate and add/develop features on this chipset that they can’t do otherwise. That is what we see as a big departure from the existing Qualcomm offerings.”
The software-defined SoC is aimed at replacing existing wireless and legacy networks with edge components that can be used to divide and partition 5G spectrum for emerging private wireless networks. The networking equivalent of private clouds, those high-bandwidth connections are being promoted as “industrial-strength” platforms that could be used to link sensors, massive amounts of raw data and AI-enabled manufacturing platforms in real time.
Yang and other early investors assert that EdgeQ’s programmable silicon moves beyond custom AI chip designs with limited use cases. “This technology will disrupt the market for silicon and democratize access to 5G for the first time,” said Yang.
Industry analysts note that AI and 5G technologies are advancing in tandem as new automation and edge use cases emerge. Among the operational efficiencies provided by AI-powered 5G networks is “predictive remediation,” in which potential outages can be identified before networks crash. “We are getting there with the help of AI,” said Will Townsend, an analyst with Moor Insights & Strategy.
Other analysts have predicted emerging AI systems on a chip. The adoption of 5G “may someday lead to convergence of the radio spectra for these disparate radio channels and convergence of network interfaces down to single chips that are agile at maintaining seamless connections across multiple radio access technologies,” James Kobielus, research director at Futurum Research, wrote last year.